LBNL-6418E

ERNEST ORLANDO LAWRENCE

BERKELEY NATIONAL LABORATORY

Building Performance Simulation

Tianzhen Hong1, Hung-Wen Lin1, Wen-Kuei Chang1,

Liping Wang1, William Turner1

Da Yan2, Dandan Zhu2, Xin Zhou2, Chuang Wang2, Chen

Peng2, Xiaoxin Ren2

1
Environmental Energy Technologies Division, LBNL
2
Department of Building Science and Technology,
Tsinghua University, China

June 2013

This report is a summary of research work conducted by LBNL and Tsinghua

University, China in 2011 and 2012 under the 5-year research program of the
U.S.-China Clean Energy Research Center for Building Energy Efficiency.
The LBNL’s research was supported by the Assistant Secretary for Energy
Efficiency and Renewable Energy of the U.S. Department of Energy under
Contract No. DE-AC02-05CH11231. Tsinghua’s research was supported by the
Ministry of Housing and Urban-Rural Development and the Ministry of Science
and Technology of China, under Contract No. 2010DFA72740-02. Some joint
research work was also sponsored by Energy Foundation.
 Disclaimer

This document was prepared as an account of work sponsored by the United States
Government. While this document is believed to contain correct information, neither
the United States Government nor any agency thereof, nor The Regents of the
University of California, nor any of their employees, makes any warranty, express or
implied, or assumes any legal responsibility for the accuracy, completeness, or
usefulness of any information, apparatus, product, or process disclosed, or represents
that its use would not infringe privately owned rights. Reference herein to any specific
commercial product, process, or service by its trade name, trademark, manufacturer, or
otherwise, does not necessarily constitute or imply its endorsement, recommendation,
or favoring by the United States Government or any agency thereof, or The Regents of
the University of California. The views and opinions of authors expressed herein do
not necessarily state or reflect those of the United States Government or any agency
thereof, or The Regents of the University of California.

Ernest Orlando Lawrence Berkeley National Laboratory is an equal opportunity

employer.
Table of Contents 
Introduction ............................................................................................................................... 1 
The U.S.‐China Clean Energy Research Center ...................................................................... 1 
Research Background ............................................................................................................ 2 
Research Team and Collaboration ........................................................................................ 5 
Research Objectives and Technical Tasks .................................................................................. 6 
Research Objectives .............................................................................................................. 6 
Technical Tasks ...................................................................................................................... 6 
Task 1 – Key drivers of energy performance of buildings ................................................. 6 
Task 2 – Occupant behavior .............................................................................................. 6 
Task 3 – Comparison of building energy modeling programs ........................................... 7 
Research Findings ...................................................................................................................... 7 
Summary ................................................................................................................................ 7 
Building Operation ................................................................................................................. 8 
Building Maintenance .......................................................................................................... 11 
Weather Impact ................................................................................................................... 16 
Occupant Behavior .............................................................................................................. 17 
Description of occupant behavior ................................................................................... 17 
Integration of behavior model with energy modeling programs .................................... 18 
Statistical model of occupancy patterns ......................................................................... 19 
Markov‐chain model of occupant moving ...................................................................... 22 
Quantitative description and simulation of human behavior in residential buildings .... 23 
Comparison of EnergyPlus, DeST and DOE‐2.1E ................................................................. 23 
Introduction ..................................................................................................................... 23 
Building loads .................................................................................................................. 26 
HVAC systems .................................................................................................................. 27 
Future Research ....................................................................................................................... 32 
Acknowledgement ................................................................................................................... 33 
Appendix A – List of Publications ............................................................................................ 34 
Appendix B – List of Workshops and Major Activities ............................................................. 36 
Appendix C – Selected Publications ........................................................................................ 38 
 
Introduction 
U.S. and China are the world’s top two economics. Together they consumed one‐third of 
the  world’s  primary  energy.  It  is  an  unprecedented  opportunity  and  challenge  for 
governments,  researchers  and  industries  in  both  countries  to  join  together  to  address 
energy issues and global climate change. Such joint collaboration has huge potential in 
creating new jobs in energy technologies and services. 

The U.S.­China Clean Energy Research Center 
In  November  2009,  President  Barack  Obama  and  President  HU  Jintao  announced  the 
establishment  of  the  $150  million  U.S.‐China  Clean  Energy  Research  Center  (CERC, 
[Link] [Link] The  Protocol  formally 
establishing  the  Center  was  signed  at  ceremonies  in  Beijing  by  U.S.  Energy  Secretary 
Steven  Chu,  Chinese  Minister  of  Science  and  Technology  Wan  Gang,  and  Chinese 
National Energy Agency Administrator ZHANG Guobao. 

The  CERC  builds  upon  over  30  years  of  U.S.  and  China  science  and  technology 
collaboration.  Under  the  Science  and  Technology  Cooperation  Agreement  of  1979  and 
its  1991  amendment,  our  two  countries  have  cooperated  in  a  diverse  range  of  fields, 
including  basic  research  in  physics  and  chemistry,  earth  and  atmospheric  sciences,  a 
variety of energy‐related areas, environmental management and more. 

The  CERC  facilitates  joint  research  and  development  on  clean  energy  technology  by 
teams  of  scientists  and  engineers  from  the  United  States  and  China.  It  is  a  flagship 
initiative funded in equal parts by the United States and China, with broad participation 
from universities, research institutions and industry. U.S. funds will be used exclusively 
to support work conducted by U.S. institutions and individuals only, and Chinese funds 
will support work conducted by Chinese institutions and researchers. 

CERC  has  three  research  themes:  (1)  CERC  Building  Energy  Efficiency  (CERC‐BEE) 
focusing  on  research  and  development  of  building  technologies,  tools,  and  policy  to 
improve design and operation of buildings to reduce energy use in buildings, (2) CERC 
Clean  Vehicles  focusing  on  research  and  development  of  new  technologies  for  electric 
vehicles  and  alternative  fuels  to  reduce  air  pollution  and  carbon  emissions  from  the 
transportation  sector,  and  (3)  CERC  Advanced  Coal  Technology  focusing  on  research 
and development of technologies to improve efficiency and reduce air emissions of coal 
power plants and new technologies for carbon capture and storage.  

The CERC‐BEE Consortium conducts R&D on building energy efficiency technologies and 
practices in the United States and China. CERC‐BEE’s vision is to, “To build a foundation 
of knowledge, technologies, tools, human capabilities, and relationships that position the 
United States and China for a future with very low energy buildings resulting in very low 
CO2 emissions.” 

BEE  develops  innovative  technologies  and  strategies  for  use  in  new  and  existing 
buildings to improve efficiency, save energy, reduce greenhouse gas emissions, increase 
indoor  comfort,  and  reduce  stress  on  the  electric  grid.  As  new  construction  proceeds 
around the globe, collaborative BEE research efforts are helping to lock in tremendous 
potential  energy  savings  for  the  long  term  via  a  more  efficient  and  low  carbon  built 
1 
 
infrastructure. Figure 1 shows the five research areas and projects for the first two years 
under the CERC‐BEE.  

Figure 1 CERC‐BEE Research Areas and Projects 

Research Background 
Buildings in the US and China consumed about 40% and 25% of the primary energy in 
both  countries  in  2010  respectively.  Worldwide,  the  building  sector  is  the  largest 
contributor  to  the  greenhouse  gas  emission.  Better  understanding  and  improving  the 
energy performance of buildings is a critical step towards sustainable development and 
mitigation of global climate change.  

Buildings demonstrate very diverse performance based on measured energy use. Figure 
2 shows site energy use intensities (EUIs) of 100 LEED‐NC certified buildings from the 
2008  New  Building  Institute  Study,  Energy  Performance  of  LEED  for  New  Construction 
Buildings. At the same LEED certification levels, energy use of green buildings varies by a 
factor of up to 4 even excluding outliners. An ICF study shows similar divergence exists 
even for same type of buildings; Figure 3 shows the EUIs of big‐box retails in the US and 
Canada. Measurements done by Tsinghua University, by China (Figure 4) have disclosed 
large differences in energy use of campus buildings in similar climates between the US 
and  China,  even  though  the  buildings  in  the  US  were  designed  to  meet  more  stringent 
energy codes than those in China. 

 
Figure 2 – Measured Energy Use Intensities (kBtu/ft²) of LEED­NC Certified Buildings                                             
(courtesy New Building Institute) 
2 
 
  
Figure 3 – Measured Energy Use Intensities of Big‐Box Retails in US and Canada (courtesy ICF International) 

 
Figure 4 – Measured Electricity Use Intensities of Campus Buildings in the US and China                                                            
(courtesy Tsinghua University, China) 

As identified in the IEA ECBCS Annex 53: Total energy use in buildings – assessment and 
analysis methods, there are six driving factors that determine the energy performance of 
buildings  (Figure 5):  climate,  building  envelope,  building  equipment  (energy  and  water 
services  systems),  operation  and  maintenance,  occupant  behavior,  and  indoor 
environmental  conditions.  Understanding  how  these  factors  affect  the  energy 
performance  of  buildings  and  which  factors  play  more  significant  role  under  certain 
conditions  can  provide  insight  into  the  large  variations  of  building  energy  use.  This  is 
also a crucial step to improve the design and operation of buildings for lower energy use 
and lower carbon emissions.  

3 
 
  
Figure 5 – Six driving factors of energy performance of buildings (courtesy IEA ECBCS Annex 53) 

Measurement and simulation are the two approaches to obtaining the energy use data of 
buildings. While measurement can provide solid and more accurate data, it can be time 
consuming and costly. On the other hand, simulation is a quick and more cost effective 
way  to  get  more  detailed  energy  use  data,  but  the  simulated/predicted  energy  use 
usually  is  not  as  accurate  as  measured  data.  Both  approaches  are  needed  as  one 
supplements the other and usually both are used in a project.  

Measurements  of  real  buildings  tend  to  show  large  discrepancies  between  simulated 
and measured energy use of buildings. Figure 6 shows the measured and simulated site 
EUIs  of  LEED‐NC  certified  green  buildings.  Averaged  across  all  buildings,  simulated 
energy  use  is  within  a  reasonable  range  from  the  measured  data,  but  looking  at  the 
individual  building,  simulation  over‐predicted  energy  use  by  up  to  120%,  and  under‐
predicted by up to 65%. More astonishingly, simulation seemed to always under‐predict 
the  energy  consumption  of  low  energy  buildings,  with  site  EUI  of  40  or  lower 
representing  about  50%  better  energy  performance  than  the  2003  CBECS  average 
commercial  buildings  in  the  US  (EUI  of  90  kBtu/ft²).  This  can  be  observed  by  the  fact 
that no points fall into the shaded triangle in Figure 6. Besides how the six driving factors 
are captured in the simulations, the energy modeling tool used, the simulation user, and 
the simulation process can have strong impact on the predicted energy use of buildings.  

 
Figure 6 – Measured vs. Simulated Energy Use Intensities (kBtu/ft²) of LEED‐NC Certified Buildings                                 
(adopted, courtesy New Building Institute) 

4 
 
Computer‐based  building  performance  simulation  has  been  widely  and  successfully 
used to: (1) evaluate design alternatives during the design of new buildings and evaluate 
retrofit  measures  of  existing  buildings,  (2)  demonstrate  code  compliance,  and  (3) 
calculate  performance  ratings.  Although  less  common  in  the  past  and  gaining 
momentum  nowadays,  simulation  is  also  used  to  predict  energy  performance  of 
buildings. 

Among the six factors, the occupant behavior is the least studied and gets too simplified 
during  the  design  and  operation  of  buildings.  How  occupants  interact  with  building 
systems have direct and decisive impacts on energy performance of buildings. 

In this project, building performance simulation will be used to provide insight into the 
following important questions: 

1. Why buildings of same type in the US and China  have such large  differences in 

measured energy use? 
2. Why buildings of same type in similar climates between the US and China have 
such large differences in measured energy use? 
3. Why  the  measured  and  simulated  energy  use  of  buildings  have  such  large 
discrepancies?  
4. How to describe and model human behavior to better understand and quantify 
its impact on building energy performance? 
5. How building simulation can be improved to better capture the influences of the 
six factors in order to better guide the design and operation of buildings for high 
performance? 
6. With  various  building  energy  modeling  programs  to  choose,  which  is  the  right 
one to use?  
7. How  and  why  do  simulation  results  vary  so  much  from  using  different  energy 
modeling programs? 

Research Team and Collaboration 
The joint research team (Table 1) includes the LBNL team, the U.S. industry partners, the 
Tsinghua team and the China industry partners. Tianzhen Hong of LBNL led the U.S. side 
research and Da Yan of Tsinghua University led the China side research. Richard Karney 
and Yi Jiang served as the senior technical advisors for the project. 
 
The  research  team  had  bi‐weekly  conference  calls  to  discuss  project  progress  and 
resolve  issues.  The  team  organized  a  series  of  workshops  (Appendix  B)  to  exchange 
research  findings,  seek  inputs  and  comments  from  researchers,  practitioners,  industry 
partners, HVAC manufacturers, government agencies, and other stakeholders. The joint 
research  work  also  made  significant  contribution  to  the  IEA  Annex  53.  Exchanged 
students  from  Tsinghua  University  stayed  at  LBNL  for  a  few  months  to  work  on  joint 
technical tasks.  

 
5 
 
 Table 1 – The Joint Research Team 

  The U.S. Side  The China Side 

Research Team  The LBNL Team  The Tsinghua Team 

Principal Investigators  Tianzhen Hong, LBNL Da Yan, Tsinghua University 

Research Team Members  William  Turner,  Liping  Chuang  Wang,  Dandan  Zhu, 

Wang,  Hung‐Wen  Lin,    Xin  Zhou,  Xiaoxin  Ren, 
Wen‐Kuei Chang  Xiaohang  Feng,  Kaiyu  Sun, 
Chen Peng 
Industry Partners  Bentley Systems, C3 Energy,  The Xinao  Group,   
Energy Foundation  the CECEP Group 

Senior Technical Advisors  Richard Karney, USDOE  Yi Jiang, Tsinghua University 

Research Objectives and Technical Tasks 
Research Objectives 
This  project  (Project  A2  in  Figure 1)  aims  to  improve  and  expand  the  use  of  building 
simulation to support the design and operation of low energy buildings through better 
understanding  and  predicting  the  energy  performance  of  buildings.  The  research 
objectives are to: (1) develop methods and models to identify and evaluate the six key 
factors to improve understanding of the energy performance of buildings in China and 
US,  (2)  develop  methods  to  describe  and  model  occupant  behavior  in  buildings,  and 
integrate  the  behavior  models  with  the  energy  models  to  simulate  the  impact  of 
occupant behavior on building performance, and (3) evaluate and compare capabilities 
of  EnergyPlus,  DOE‐2.1E  and  DeST  to  better  understand  their  commons,  differences, 
strengths,  weakness  and  limitations,  and  to  guide  simulation  users  how  to  choose  an 
energy modeling tool for a specific application. 

Technical Tasks 
To  achieve  the  three  aforementioned  research  objectives,  three  research  tasks  were 
designed and completed. 

Task 1 – Key drivers of energy performance of buildings 
This  task  developed  simulation  methods  and  models,  and  performed  sensitivity  and 
scenario  analysis  to  quantify  the  impact  of  operation  practice,  maintenance  practice, 
occupant behavior, and weather on energy performance of typical office buildings in the 
U.S. Integrated consideration of these influencing  factors in  a  holistic system approach 
during  the  design  and  operation  of  buildings  is  the  most  critical  key  in  achieving  high 
performance buildings. 

Task 2 – Occupant behavior 
Occupant  energy‐related  behavior  in  buildings  has  two  aspects:  (1)  occupant  needs  of 
comfort:  thermal,  acoustical,  visual  and  indoor  air  quality  (IAQ),  and  (2)  occupant 
responses by interacting with building systems to restore comfort if their needs are not 
6 
 
met.  The  usual  occupant  interactions  include  opening/closing  windows,  operating 
blinds  (or  other  shading  devices),  switching  or  dimming  lights,  adjusting  thermostat, 
turning  HVAC  on/off,  turning  on/off  computers,  monitors  and  portable  devices,  and 
adjusting clothing. Occupant behavior includes occupant movement – how often and for 
how long an occupant moves around, and personal habit. ASHRAE Standard 55 defines 
comfort  range  for  typical  occupants  at  various  activities  and  environments.  The  latest 
additions to the standard allow the cooling setpoint of the indoor air temperature to be 
adjusted upper if indoor air velocity is elevated or based on the adaptive comfort model. 
ASHRAE  Standard  62.1  defines  minimum  outdoor  air  needed  for  healthy  ventilation. 
Differences in occupant behavior can lead to huge variations of energy use in buildings. 
On  the  other  hand,  changes  to  human  behavior  at  no‐cost  have  demonstrated  energy 
savings of 5 to 30% in office and residential buildings.  

Current  simulation  programs,  including  EnergyPlus,  DOE‐2.1E  and  DeST,  have  limited 
capability  to  model  human  behavior  considering  its  multidisciplinary  and  inherent 
complexity  and  uncertainty.  This  task  will:  (1)  identify  key  energy  related  behavior  of 
building  occupants  based  on  measured  and  survey  data,  and  literature  review,  (2) 
develop algorithms to model occupant behaviors, (3) integrate the behavior models with 
energy models: EnergyPlus and DeST, and 4) assess the impact of occupant behavior on 
energy use in selected buildings in the US and China.  

Task 3 – Comparison of building energy modeling programs 
EnergyPlus  is  a  computer  program  developed  by  USDOE  to  simulate  the  energy 
performance  of  buildings.  It  was  started  to  combine  the  best  features  of  two  former 
energy  modeling  programs:  DOE‐2  and  BLAST,  but  has  since  evolving  as  a  powerful 
simulation  engine  to  calculate  building  performance.  DeST  is  a  simulation  program 
developed by Tsinghua University since the early 1990s and has been getting more and 
more use in China. DOE‐2 was developed by LBNL in the early 1980s through mid 1990s 
to simulate energy performance of buildings. This task will compare the three programs, 
evaluate their capabilities, identify their commons and differences, strengths, weakness, 
and limitations in energy modeling. This provides guidance and support the simulation 
use in the building energy code development, code compliance, and performance rating 
and labeling. 

Research Findings 
This  section  summarizes  the  key  research  findings.  Detailed  description  of  research 
work,  technical  approaches,  and  results  were  published  in  11  journal  articles,  11 
conference papers, and two technical reports (Appendix A). Selected publications were 
included in Appendix C. 

Summary 
 Building  operation,  maintenance,  occupant  behavior,  and  weather  data  are  key 
drivers of  energy  performance of  buildings. Various  practices  of  building operation 
and  maintenance,  various  types  of  occupant  behavior,  and  long‐term  actual  yearly 
weather data should be used in building simulations when evaluating energy savings 
of  building  designs  and  technologies  in  order  to  understand  and  quantify  the 

7 
 
 sensitivity of such savings. Technologies sensitive to these factors may not work well 
when they are applied to real buildings. 
 Different  building  operation  practices  lead  to  different  building  performance,  and 
such  differences  can  be  greater  than  those  caused  by  different  design  efficiency 
levels  of  buildings.  Good  operation  is  crucial  to  achieving  low  energy  buildings, 
providing comfort for occupants, and extending equipment life.  
 Building  maintenance  has  strong  influence  on  energy  performance  of  buildings. 
Different  maintenance  practices  lead  to  large  differences  in  energy  use.  Good 
maintenance  is  another  key  to  achieving  low  energy  buildings,  providing  comfort 
and extending equipment life.  
 Yearly  variation  of  weather  is  significant  and  has  strong  impact  on  building 
performance.  Energy  savings  and  peak  electric  demand  reduction  of  energy 
technologies calculated using traditional weather data TMY can be significantly over 
or  under‐estimated  for  actual  weather.  When  evaluating  design  alternatives  and 
retrofit measures, long‐term actual yearly weather data should be used. This is now 
feasible and convenient considering the availability of historical AMY weather data 
and computing power of typical laptop and desktop computers nowadays. 
 Technologies alone may not lead to high performance buildings if occupant behavior 
is  ignored.  Occupant  behavior  is  complex,  stochastic,  and  multi‐disciplinary. 
Methods were developed to describe and model occupant behavior and to evaluate 
its impact on energy use in buildings. Various types of occupant behavior should be 
considered during the building design and operations. Changes to occupant behavior 
can be a no‐cost measure to save energy. 
 Markov‐chain  models  were  developed  to  describe  occupant  movement.  Statistical 
models were developed to describe occupancy patterns in single‐occupied offices. 
 Occupant behavior, related to opening/closing windows, turning on/off TVs and air –
conditioners,  and  turning  on/off  lights,  was  identified  and  models  were  developed 
for residential buildings by mining measured data. 
 Simulation  methods  were  developed  to  compare  the  energy  performance  of  four 
office  buildings  in  the  U.S.,  China,  Hong  Kong,  and  Taiwan  to  identify  and  analyze 
driving  factors  to  the  discrepancies  of  energy  use  of  these  buildings.  The  methods 
can be used for benchmarking analysis of building performance by simulations. 
 An in‐depth comparative study was conducted to analyze the three building energy 
modeling  programs:  EnergyPlus,  DOE‐2.1E  and  DeST.  Comparison  methods  and 
suites of test cases were developed to identify and quantify the discrepancies of the 
results from the three programs. In general three programs can provide consistent 
simulation results if inputs to them are the same or equivalent. Due to its limitations, 
DOE‐2.1E  can  have  large  errors  if  adjacent spaces  have very  different  load  profiles 
and  operating  schedules.  For  HVAC  simulations,  the  largest  discrepancies  again 
came  from  discrepancies  in  user  inputs,  followed  by  HVAC  equipment  control 
strategies. There is no simple rule of which program is the best; it all depends on the 
application. It is recommended to apply a program within its application scope and 
observe its limitations. 

Building Operation 
Energy  use  of  buildings  varies  in  a  wide  range.  Large  office  buildings  in  Chicago  and  built 
after 2000 were selected from the 2003 CBECS database and the high performance building 
database. These buildings’ measured total source energy use intensities (EUIs) were graphed 
at  the  top  portion  of  Figure  7.  The  size  of  the  bubble  represents  the  EUI.  These  buildings 
were  compared  to  the  DOE  commercial  reference  large  office  building  (baseline  building) 

8 
 
compliant  with  ASHRAE  90.1‐2004  in  Chicago,  and  their  relative  performance  in  EUI  were 
calculated as a percentage (compared to the simulated EUI of the baseline building which is 
1.39 MJ/m2) and shown in Figure 7.  

 
 
Figure 7 – Impact of building design and operation for a large office building in Chicago 

As Figure 7 shows, the source EUIs of the seven large office buildings from the CBECS vary by 
a  factor  of  four.  The  only  one  high  performance  large  office  building  found  from  the  HPB 
database consumed about half the total source energy compared to the baseline  building. 
Why  would  the energy  performance  of these buildings of  same  type (large  office)  in same 
climate  (Chicago)  built  in  same  era  (after  2000)  vary  by  a  factor  of  eight?  A  rational 
assumption  is  these  buildings  use  different  technologies  with  different  energy  efficiency 
levels (insulation or walls and roofs, windows, lighting power, COP of chiller etc.) and these 
buildings are operated very differently. To understand the impact of building operation and 
building  design  on  energy  use,  key  parameters  (Table  2)  were  selected  to  represent  the 
building design and operation.  

9 
 
 Table 2 – Design and operation parameters 

Design parameters  Operation parameters

WWR (Window‐wall‐ratio)  Daylighting control 
Window  Shading control 
Wall insulation  Infiltration rate 
Internal loads (lighting + plug‐load)  Infiltration schedule 
Economizer  VAV box damper setting 
Chiller efficiency  Heating thermostat 
Boiler efficiency  Cooling thermostat 
Heating setback during unoccupied hours 
Cooling setback during unoccupied hours 
Supply air temperature reset 
Lighting schedule 
Plug‐load schedule 
 

Typical values (Figure 8) of the design parameters are grouped into three categories: (1) high 
efficiency  design,  based  on  ASHRAE  Standard  90.1‐2010  requirements  and  best  design 
practice,  (2)  standard  design  compliant  with  ASHRAE  Standard  90.1‐2004,  and  (3)  low 
efficiency  design  based  on  characteristics  of  old  office  buildings.  Typical  values  of  the 
operation parameters are grouped into three categories too: (1) good operation practice, (2) 
standard  operation  practice,  and  (3)  poor  operation  practice.  These  various  design  and 
operation practices were applied to the baseline building and simulated with EnergyPlus.  

When the three levels of building design efficiency were combined with the three types of 
operation  practices,  their  compound  effects  on  the  building  total  source  energy  were 
simulated.  The  bottom  portion  of  Figure  9  shows  the  simulated  source  EUIs  of  the  nine 
combined cases (the middle bubble represents the baseline building). 

It can be seen that: 

 The design efficiency plays a significant role. Under standard operation practice, the 
low  efficiency  building  consumed  4.5  times  of  the  total  source  EUI  of  the  high 
efficiency building.  Under  the good and bad operation practices, the ratio  is  about 
3.7 and 4 respectively. The ratios are not very different. 
 The operation practice is as important. Under the standard design efficiency, the bad 
operation practice consumed 2.6 times of the total source EUI of the good operation 
practice.  Under  the  high  and  low  design  efficiency  levels,  the  ratios  are  3  and  3.2 
respectively.  
 Combined  together,  the  building  with  the  low  design  efficiency  and  the  bad 
operation practice consumes 12 times of the source energy of the building with the 
high  design  efficiency  and  the  good  operation  practice.  Although  these  ideal 
combinations may not exist in real word, it fully explains why the actual/measured 
energy performance of building varies by a factor of 8. 

10 
 
  To  achieve  high  performance  building,  using  latest  products  and  technologies  to 
improve  design  efficiency  is  necessary,  but  as  important  is  the  improving  of  good 
operation practice. 
 When evaluating energy savings of energy efficiency technologies, various operation 
practices should be considered, at least as sensitivity analysis to look the persistency 
and variation range of the calculated energy savings. 
 Similar trends were observed in other US climates. 

 
Figure 8 – Typical values of design and operation parameters and their impact on total source energy use 

Building Maintenance 
Different  practices  of  HVAC  system  maintenance  can  result  in  substantial  differences  in 
building energy use. If a piece of HVAC equipment is not well maintained, its performance 
will degrade. If sensors used for control purpose are not calibrated, not only building energy 
usage  could  be  dramatically  increased,  but  also  mechanical  systems  may  not  be  able  to 
satisfy indoor thermal comfort. Properly  maintained HVAC  systems can operate efficiently, 
improve occupant comfort, and prolong equipment service life.  

The maintenance practices are categorized into three levels depending on the maintenance 
effort  and  coverage:  1)  proactive,  performance‐monitored  maintenance;  2)  preventive, 
scheduled maintenance; and 3) reactive, unplanned or no maintenance. Table 3 summarizes 
the  three  practices  of  HVAC  maintenance  and  their  implications  on  equipment  operating 
efficiency and energy use, equipment life, short term maintenance cost, and life cycle cost 
11 
 
including  maintenance  cost,  energy  cost,  and  equipment  replacement  or  repair  cost.  The 
good practice will lead to lowest life cycle cost, while the bad practice seems to save short 
term maintenance cost, it will result in the highest life cycle cost. 
Table 3 ‐ Three types of HVAC maintenance practices 

 
A  list  of  maintenance  issues,  including  cooling  tower  fouling,  boiler/chiller  fouling, 
refrigerant  over  or  under  charge,  temperature  sensor  offset,  outdoor  air  damper  leakage, 
outdoor air screen blockage, outdoor air damper stuck at fully open position, and dirty filters 
are  investigated  in  this  study  using  field  survey  data  and  detailed  simulation  models.  The 
energy impacts of both individual  maintenance  issue and combined scenarios for an office 
building  with  central  VAV  systems  and  central  plant  were  evaluated  by  EnergyPlus 
simulations using three approaches:  

1) Direct modeling with EnergyPlus (Direct Modeling)  
Maintenance  issues  are  directly  modeled  using  existing  inputs  (either  design  input 
parameters  or  performance  curves)  in  the  current  version  of  EnergyPlus.  This  modeling 
approach  can  be  applied  to  such  maintenance  issues  as  supply  air  sensor  offset,  zone 
thermostat offset  and outdoor  air damper  leakage.  This  approach is also applied to  model 
simplified maintenance issues such as chiller or boiler fouling by introducing a degradation 
factor to the chiller or boiler efficiency inputs to the EnergyPlus models. The  advantage of 
this approach is easy implementation.  

2) Using the energy management system (EMS) in EnergyPlus 
EMS  is  an  advanced  feature  of  EnergyPlus  and  designed  for  users  to  develop  customized 
high‐level, supervisory control routines to override specified aspects of EnergyPlus modeling 
in  the  EMS  program.  The  EMS  feature  in  EnergyPlus  is  flexible  to  allow  users  to  simulate 
equipment operating with some maintenance issues by overwriting or adding algorithms in 
EnergyPlus within the specified aspects of current EMS capability. Use of EMS feature may 
require  advanced  knowledge  of  EnergyPlus  and  computer  programming.  EMS  is  used  to 
model  maintenance  issues  like  dirty  filters  which  increase  pressure  drop  across  the  filter 
with operating hours.  

3) Modifying EnergyPlus source code (Modified Code)  
Modifying the existing EnergyPlus source code, the third modeling approach, is used when 
both  direct  modeling  and  EMS  approaches  cannot  be  applied  to  simulate  any  particular 
equipment  or  system  deficiencies.  This  approach  requires  users  to  have  a  thorough 
understanding  of  the  existing  EnergyPlus  source  code  and  to  write  your  own  custom 
computer  program  based  on  existing  code.  Such  HVAC  maintenance  issues  as  cooling  coil 
fouling, outdoor air and return air temperature sensors offset adopt the third approach.    

12 
 
Table 4 shows a list of common HVAC maintenance issues with their potential impacts and 
modeling  approach  according  to  maintenance  types,  including  sensor  calibration,  filter 
replacement,  heat  exchanger  treatment,  mechanical  repair  and  refrigerant  charge,  are 
investigated  using  detailed  simulation  models.  A  description  of  the  implement  model  for 
selected maintenance issues is as follows. 

 Temperature sensor offset 

Control  sensors  such  as  supply  air  temperature  (SAT)  sensors,  zone  thermostats,  and 
outdoor air temperature (OAT) sensors may be out of calibration over a long term operation 
period. In this study, it is assumed that temperature sensors are offset by ±2˚C. For example, 
if a SAT sensor is offset by +2 ˚C and a designed supply air temperature to control is 13˚C, the 
actual supply air temperature due to sensor offset is 11 ˚C. 

 Dirty filter 

In  terms  of  filter  replacement  for  reactive  maintenance,  it  is  assumed  that  filters  in  air 
handler units have not been replaced over a year. Therefore, pressure drop for air handler 
units has been increased and the maximum additional pressure drop is 500 Pa. 

 Fouled cooling tower  

Cooling  towers  can  become  fouled  due  to  unfavorable  conditions.  The  study  assumes 
certain  fouling  condition  that  overall  heat  transfer  coefficient  is  reduced  to  85%  of  design 
value.  

 Fouled Chiller/Boiler/Coils 

Fouling on heat transfer surfaces of boiler and chiller increases the thermal resistance and 
leads to reduced heat transfer. For the scenario of chiller/boiler fouling, both chiller COP and 
boiler efficiency are assumed to be reduced by 10%. For fouled cooling/heating coils, overall 
heat transfer coefficients are assumed to be reduced to 50% of design UAs. 

 Outdoor air damper (OAD) leakage 

In the study, it is assumed that OAD leakage level is 30%. When the commanded outdoor air 
fraction  is  smaller  than  the  leakage  level,  leaky  damper  cannot  effectively  control  the  air 
intake.  

 Stuck outdoor air damper (OAD) 

Stuck  OAD  due  to  control  and  mechanical  failure  is  another  common  fault  in  field.  In  this 
study,  OAD  is  assumed  to  get  stuck  at  fully  open  position.  Cooling  and  heating  energy 
penalties are introduced when outdoor air is not favorable for free cooling.  

 Clogged OA screen 

Outdoor  air  intake  screens  may  get  clogged  due  to  unfavorable  locations  or  weather 
condition. The maximum percent of intake fresh air is assumed to reduce to 70%. 

 Stuck outdoor air damper (OAD) 

Stuck  OAD  due  to  control  and  mechanical  failure  is  another  common  fault  in  field.  In  this 
study,  OAD  is  assumed  to  get  stuck  at  fully  open  position.  Cooling  and  heating  energy 
penalties are introduced when outdoor air is not favorable for free cooling.  
13 
 
 Table 4 – A list of HVAC maintenance issues 

Maintenance  Maintenance  Simulated  Modeling 

Impacts 
Types  Issues  Scenarios  Approach 

Supply  air 
temperature  Direct  modeling, 
sensor  (SAT)  adjust SAT setpoint 
offset 
controls,  heating  temperature  Direct  modeling  , 
Sensor  Zone temperature  and  cooling  sensors  are  adjust  thermostat 
Calibration  sensor offset  energy  offset by ±2˚C  settings 
Modified  Code, 
Outdoor  air 
modify  the 
temperature 
economizer 
sensor offset 
controls 
pressure  drop,  additional  500Pa  EMS,  adjust  fan 
Filter 
Dirty filter  fan  energy,  of  air  pressure  power  for  VAV 
replacement 
airflow  drop  systems 
overall  heat 
transfer  Direct  modeling, 
Fouled  cooling 
efficiency  coefficient  is  adjust  cooling 
tower 
reduced  to  85%  tower UA 
of design UA
Direct  modeling, 
Chiller:  fouled  chiller  COP  is 
efficiency  adjust  chiller 
Heat  exchanger  tubes  reduced by 10% 
efficiency 
cleaning/treatme
boiler  efficiency  Direct  modeling, 
nt  Boiler: hard water 
efficiency  is  reduced  by  adjust  boiler 
scale 
10% efficiency 
overall  heat 
transfer 
Fouled  heating  efficiency,  Modified  Code, 
coefficient  is 
/cooling coil  comfort  adjust coils UA 
reduced  to  50% 
of design UAs
Direct  modeling, 
Outdoor  air  heating  and  30%  OAD 
adjust minimum OA 
damper leakage  cooling energy  leakage 
flow 
OAD  is  stuck  at 
Stuck  outdoor  air  heating  and  EMS,  set  constant 
fully  open 
damper (OAD)  cooling energy  OA flow 
position
Mechanical 
outdoor  air  flow 
repair 
is less than 100%  
maximum 
during  Direct  modeling, 
Clogged  OA  percent of intake 
economizer  set  maximum  OA 
screen  fresh  air  is 
mode  thus  flow 
reduced to 70% 
increasing 
cooling energy
Chiller:  over  or  Direct  modeling, 
Refrigerant  chiller  COP  is 
under  10%  efficiency  adjust  chiller 
charge  reduced by 10% 
refrigerant charge efficiency 
 

   

14 
 
 Clogged OA screen 

Outdoor  air  intake  screens  may  get  clogged  due  to  unfavorable  locations  or  weather 
condition. The maximum percent of intake fresh air is assumed to reduce to 70%. 

The energy penalty introduced by HVAC maintenance issues varies by a few factors including 
building  and  HVAC  systems  types,  vintage  (design  efficiencies),  and  climates.  In  the  study, 
the  USDOE  commercial  building  reference  model  for  a  large‐size  office  building  in 
compliance  with  ASHRAE  Standard  90.1‐2004  is  used  as  a  baseline  representing  good 
maintenance practice.  The  large‐size office  building  consists of  one basement level  and  12 
floors above ground served by 4 built‐up VAV systems with 2 water‐cooled chillers and one 
natural gas hot‐water boiler.  

The results, shown in Figure 9, demonstrated the energy penalty introduced by the reactive 
maintenance practice for the built‐up VAV system located in Chicago. The percentages are 
derived by comparing the total source/primary energy use of HVAC systems for the reactive 
maintenance  practice  to  those  of  the  good  practice  (baseline  reference  model).  The 
maintenance issues with significant energy impacts for Chicago are OA damper stuck at 100% 
position, blocked OA screen, supply air temperature offset, boiler/chiller fouling, and chiller 
refrigerant  under/overcharge.  Although  there  is  no  significant  energy  impact  due  to 
heating/cooling coil fouling, the numbers of unmet thermal comfort hours for both heating 
and cooling are significantly increased due to reduced system cooling and heating capacities. 
Two  combined  scenarios  (#1  and  #2)  with  different  temperature  sensor  offsets  were 
simulated in the study. The overall energy penalty by combining the sampled maintenance 
issues including sensor offset by +2 ˚C can reach 85% of overall HVAC energy consumption 
for Chicago climate.  

Mechanic
al Repair

Heat
exchanger
cleaning/tre
atment

Sensor
Calibrati
on

 
Figure 9 ‐ Impact of poor HVAC maintenance on HVAC source energy use of a large office building in Chicago 

Modeling  and  simulation  of  building  maintenance  can  help  practitioners  and  building 
operators  to  gain  the knowledge  of  maintaining HVAC  systems  in efficient operations,  and 
prioritize HVAC maintenance work plan. 

15 
 
Weather Impact 
Yearly  variations  of  weather  are  significant,  which  can  even  change  the  climate  zone 
classification  of  a  location.  Figure  10  shows  the  variations  of  weather  data  on  the 
ASHRAE  climate  zone  map  for  the  17  representative  cities  based  on  the  AMY  (Actual 
Meteorological  Year)  data  from  1980  to  2009.  It  can  be  seen  that  most  cities  do  not 
belong to only one climate zone. For the 30‐year period, the climates of some cities vary 
across two zones and some even across three or more zones.  

Yearly  variations  of  weather  have  strong  impact  on  energy  performance  of  buildings. 
The  simulated  energy  savings  and  peak  electrical  demand  reduction  from  the  use  of 
energy efficiency technologies are sensitive to the weather data used in the simulation, 
this is in contrary to the traditional thinking of weather data only impacts the absolute 
energy  use  but  not  the  relative  performance  between  two  different  designs.  Figure 11 
shows the variations of peak demand reduction (in %) and the HVAC and building total 
source  energy  savings  (in  %)  by  comparing  the  performance  of  the  prototype  office 
buildings  designed  to  meet  the  ASHRAE  standard  90.1‐2010  with  those  meeting  the 
90.1‐2004, using the TMY3 and the 30‐year AMY weather data across the 17 climates.  In 
Figure 11,  the  green  bars  represent  the  variation  in  the  demand  reduction  and  source 
energy  savings,  using  the  30‐year  AMY  weather  data.  The  red  marks  represent  the 
corresponding results using the TMY3 weather data.  

Generally  the  weather  impact  on  the  peak  demand  reduction  is  much  greater  than  on 
the  HVAC  source  energy  savings.  For  HVAC  source  energy  savings,  larger  weather 
impacts  occur  for  the  mixed  to  cold  climates,  from  San  Francisco  to  Fairbanks.  The 
savings based on TMY3 weather files are usually within the ranges of savings based on 
the AMY weather files, except for over‐estimates in San Francisco, Albuquerque, Boise, 
Vancouver, and Helena, where the red marks are usually at the very right end or outside 
of the green bars. The peak demand reduction can vary significantly year‐over‐year for 
most climates. The differences in demand reduction can be as high as 15% for Chicago 
and Fairbanks across the 30‐year period for the large office building. 

With the availability of long‐term AMY weather data and sufficient computational power 
of personal computers, it is feasible and necessary to run simulations with AMY weather 
data  covering  multiple  decades  to  fully  assess  the  impact  of  weather  on  the  long‐term 
performance  of  buildings,  and  to  evaluate  the  energy  savings  potential  of  energy 
conservation measures for new and existing buildings from a life cycle perspective. Main 
findings  are: (1)  annual  weather variation  has  a  greater impact  on the  peak  electricity 
demand than on the energy use in buildings; (2) simulated building energy use using the 
TMY3 weather data is not necessarily representative of the average energy use using the 
AMY  data,  across  the  30‐year  period.  The  TMY3  results  can  be  significantly  higher  or 
lower than those from the AMY data; (3) the weather impact is greater for buildings in 
cold  climates;  (4)  the  weather  has  the  greatest  impact  on  the  medium‐size  office 
building, followed by the large office and then the small office; and (5) simulated energy 
savings and peak demand reduction by energy conservation measures using the TMY3 
weather  data  can  be  significantly  lower  or  higher  when  compared  to  the  results  using 
the  AMY  data.  These  findings  can  support  energy  policy  making,  energy  code 
development, building technologies evaluation, and utility incentive programs planning. 
16 
 
  
Figure 10 ‐ Variations of yearly weather data on the ASHRAE climate zone map 

 
Figure 11 – Variations of percentage reduction of peak electricity demand, and percentage savings of HVAC 
source energy and total source energy of the ASHRAE 90.1‐2010 models over the ASHRAE 90.1‐2004 models.            
(a)‐(c) large office; (d)‐(f) medium office; (g)‐(i) small office. 

Occupant Behavior 
Description of occupant behavior 
Occupant behavior refers to an occupant’s movement and responses to discomfort when 
his/her  comfort  needs  are  not  met  by  interacting  with  building  systems,  including 
windows,  shades,  lights,  appliances,  thermostat,  and  HVAC  systems.  An  occupant’s 
comfort  needs  include  thermal,  acoustic,  visual,  and  indoor  air  quality.  Occupant’s 
behavior  varies  with  time,  space,  and  individual.  It  is  stochastic,  complex,  and 
multidisciplinary.  A  framework  is  developed  to  describe  occupant  energy‐related 

17 
 
behavior (Figure 12), which captures the three key elements: Drivers, Needs, and Actions. 
Drivers determine the needs which further determine the actions. 

 
Figure 12 – A framework to describe occupant energy‐related behavior 

The  framework  will  be  represented  in  an  XML  schema  to  provide  a  standard  definition  of 
occupant behavior. Typical occupant behavior, actions on building systems, personal habit, 
and  adjustment  of  personal  clothing,  moving,  etc,  will  be  represented  as  mathematical 
models suitable for simulation with energy modeling programs such as EnergyPlus and DeST. 

Integration of behavior model with energy modeling programs 
Typical  occupant  behavior  in  single  occupancy  private  offices  was  studied,  including 
how an occupant sets comfort criteria (cooling and heating thermostat), operates lights 
and office equipment, and turns on/off HVAC systems. The behavior is categorized into 
three workstyles (Table 5): (1) austerity – occupants are proactive in saving energy, (2) 
standard – average occupants, and (3) wasteful – occupants don’t care about energy use.  

The  three  types  of  occupant  behaviors  were  modeled  using  EnergyPlus  for  a  single‐
occupancy  office  room.  The  simulation  results  demonstrate  the  impact  of  occupant 
behavior on building energy use is significant, and even so at the energy end use levels 
such  as  lighting,  space  cooling  and  heating.  Compared  to  the  standard  or  average 
workstyle,  the  austerity  workstyle  reduces  energy  use  by  50%,  while  the  wasteful 
workstyle increases energy use by up to 90% (Figure 13).  

 
18 
 
 Table 5 ‐ Occupant behavior categorized into three workstyles 

Occupant behavior Austerity Standard Wasteful

workstyle workstyle workstyle
Cooling setpoint (°C) 26 24 22
Heating setpoint (°C) 18 21 23
Follow ASHRAE
Adaptive Comfort Yes None None
Model
If unoccupied, turn Leave everything
off lights and
Occupancy controls Scheduled on: lights, HVAC,
HVAC, turn down and plug-load
plug-load 30%
Daylighting Control 3 Steps Dimming None None
Turn on 1 hour late Scheduled on: Same as the whole
HVAC operation time and turn off 1 hour building schedule:
8am to 5pm
early: 9am to 4pm 6am to 10pm
Cooling turns on
Follow HVAC
when space air operation Follow HVAC
temperature operation schedule
Cooling startup schedule (8am
control reaches 28°C, then to 5pm) to (6am to 10pm) to
maintains at 24°C. maintain 24°C.
Cooling turns off maintain 24°C. Same as above.
Same as above.
when unoccupied.
All above All above All above
Combined behavior behavior behavior  

 
 
Figure 13 – Impact of occupant behavior for a private office 

Three methods were used to model occupant behavior depending upon the complexity: 
(1)  use  EnergyPlus  directly,  (2)  use  the  advanced  feature  of  EnergyPlus  ‐  Energy 
Management  System,  and  (3)  modify  source  code  of  EnergyPlus.  A  software  module  is 
being  developed  to  allow  its  co‐simulation  with  EnergyPlus  for  various  types  of 
occupant behaviors. 

Statistical model of occupancy patterns 
Two important aspects of the occupant behavior research are: (1) measure indoor and 
outdoor  environmental  parameters  and  occupant’s  interactions  with  building  systems 
and  personal  movement,  (2)  analyze  the  measured  data  to  identify  occupant  behavior 
19 
 
and  develop  mathematical  models  for  such  behavior.  Occupancy  profile  is  one  of  the 
driving  factors  behind  discrepancies  between  the  measured  and  simulated  energy 
consumption  of  buildings.  The  frequencies  of  occupants  leaving  their  offices  and  the 
corresponding  durations  of  absences  have  significant  impact  on  energy  use  and  the 
operational  controls  of  buildings.  Statistical  methods  were  used  to  analyze  the 
occupancy status, based on measured lighting‐switch data in five‐minute intervals, for a 
total of 200 open‐plan (cubicle) offices. Five typical occupancy patterns were identified 
based  on  the  average  daily  24‐hour  profiles  of  the  presence  of  occupants  in  their 
cubicles. These statistical patterns were represented by a one‐square curve, a one‐valley 
curve,  a  two‐valley  curve,  a  variable  curve,  and  a  flat  curve  (Figure  14).  The  key 
parameters that define the occupancy model are the average occupancy profile together 
with probability distributions of absence duration, and the number of times an occupant 
(Figure 15 and Figure 16) is absent from the cubicle. The statistical results also reveal that 
the  number  of  absence  occurrences  decreases  as  total  daily  presence  hours  decrease, 
and  the  duration  of  absence  from  the  cubicle  decreases  as  the  frequency  of  absence 
increases. The developed occupancy model captures the stochastic nature of occupants 
moving  in  and  out  of  cubicles,  and  can  be  used  to  generate  more  realistic  occupancy 
schedules  (Figure 17  and  Figure 18).  This  is  crucial  for  improving  the  evaluation  of  the 
energy  saving  potential  of  occupancy  based  technologies  and  controls  using  building 
simulations.  

 
 
Figure 14 ‐ The occupancy patterns: (a) single‐square curve, (b) one‐valley curve, (c) two‐valley curve, (d) 
variable curve,       (e) flat curve 

20 
 
  

 
Figure 15 ‐ The curves of occurrences, probability distribution function (PDF), and cumulative distribution 
function (CDF) of Pattern 1: (a) number of daily absences; (b) absence duration 

 
Figure 16 ‐ The curve of cumulative distribution function (CDF) of daily absence section for Pattern 2 

21 
 
  
Figure 17 ‐ Three generated weekday occupant schedules for Pattern 1 

 
Figure 18 ‐ Three generated weekday occupant schedules for Pattern 2 

Markov­chain model of occupant moving 
Building occupancy is an important basic factor in building energy simulation but it is
hard to represent due to its temporal and spatial stochastic nature. A novel approach
for building occupancy simulation based on the Markov chain was developed. In this
22 
 
study, occupancy is handled as the straightforward result of occupant movement
processes which occur among the spaces inside and outside a building. By using the
Markov chain method to simulate this stochastic movement process, the model can
generate the location for each occupant and the zone-level occupancy for the whole
building. There is no explicit or implicit constraint to the number of occupants and the
number of zones in the model while maintaining a simple and clear set of input
parameters. From the case study of an office building, it can be seen that the model
can produce realistic occupancy variations in the office building for a typical workday
with key statistical properties of occupancy such as the time of morning arrival and
night departure, lunch time, periods of intermediate walking-around, etc. Due to
simplicity, accuracy and unrestraint, this model is sufficient and practical to simulate
occupancy for building energy simulations and stochastic analysis of building HVAC
systems.

Quantitative  description  and  simulation  of  human  behavior  in  residential 

buildings 
This study gives a quantitative description of human behavior in residential buildings.
The quantitative description method can be used to forecast the impact of the human
behavior on the indoor built environment and energy use. Human behavior influences
the energy use directly and indirectly by changing window openings, air-conditioner
usage, lighting, etc. The quantitative description method describes these behavioral
effects. Behavior can be divided into several types according to the usage with time
related, environmentally related and random modes used to quantitatively describe the
behavior. The method is then applied to describe a household in Beijing with
comparison to on-site observations of the occupant behavior and measurements of
energy use to validate the method. The results show that the human behavior in the
real world can be quantified by the quantitative description method. These simulation
tools can greatly facilitate building energy conservation by describing the influence of
human behavior on building performance and energy use.

Comparison of EnergyPlus, DeST and DOE­2.1E 
Introduction 
Building Energy Modeling (BEM) programs play a significant role in the design of
energy efficient envelopes and HVAC systems for new buildings and retrofit of
existing buildings, the development and compliance of building energy codes, and
implementation of building energy rating/labeling programs. However, the issue that
large discrepancies exist in simulation results between different BEM programs, even
for the same building modeled by the same person, leads to many users and
stakeholders lack confidence in building simulation methods and the results from
simulations. This is a major barrier for the wider adoption and effective application of
building energy simulation, and represents a challenge to the industry. The large
discrepancies of simulation results from different BEM programs mainly come from
three factors (Figure  19): first is the simulation engine which is the unchangeable core;
second is the GUI to the simulation engine that usually simplifies, hides or hard-wires
23 
 
some inputs that can be important; third is the fact that users may model the building
or system inaccurately as they may not be familiar with the chosen BEM program, or
input poor data due to constraints of budget and resources. In order to address the
issue of large discrepancies between different BEM programs, the impact of the above
three factors must be identified and quantified.

Figure 19 ‐ Top three influencing factors of the discrepancies in simulation results

EnergyPlus, DOE-2.1E and DeST were selected for the in-depth comparative analysis
to qualitatively and quantitatively assess the main influencing factors driving the
discrepancies of simulation results. EnergyPlus was chosen because it is widely used
and continuously being developed and supported by USDOE. DOE-2.1E was chosen
as it is still widely used in the U.S. DeST was chosen due to its popular use in China
and a few Asian countries. Top-level key features of DOE-2.1E, DeST and
EnergyPlus are summarized in Table 6.

Due to the complexity of BEM programs and the intention of isolating influencing
factors, two separate comparisons were made: the building loads and the HVAC
systems.

24 
 
 Table 6 ‐ Comparison of top-level key features of DOE-2.1E, DeST and EnergyPlus

Feature DOE-2.1E DeST EnergyPlus

Tsinghua University,
Developer LBNL/DOE, US DOE/LBNL et al. US
China

Development
Early 1980s 1989 1996
Started

Development No more development or

On-going On-going
and Support support

Users Worldwide Mostly China Worldwide

Database, Microsoft
Inputs Text, BDL Text, IDF
Access

Extensive summary &

Summary & hourly Summary & hourly
Outputs detailed reports with user
reports reports
specified time steps

Simulation engine only; Simulation engine only;

GUI Coupled with AutoCAD
3rd party GUIs available 3rd party GUIs available

Surface heat transfer:

Zone heat balance: State Surface heat balance: CTF;
Algorithms CTF; Zone weighting
Space Method Zone heat balance
factors

Time Step 1 hour, fixed 1 hour, fixed 1 to 60 minutes

Weather Data Hourly Hourly Hourly or sub-hourly

A few pre-defined User configurable with

HVAC 28 pre-defined systems
systems some limitations

User Energy Management

User functions N/A
Customization Systems

Co-Simulation N/A N/A External Interface

Language Fortran 77 C++ Fortran 2003

Limited user
Lack zone air heat Potentially long run-time
Limitations customization, linear
balance, linear systems for large models
systems

Free download; Free download; Source Free download;

Licensing
Source code available code not open to public Open source

25 
 
Building loads 
Detailed comparison of the building thermal load modeling capabilities and
simulation results of the three BEM programs was made with the goal to identify and
quantify the influences of the simulation engines and input values or algorithms. Test
cases, with modifications to the ASHRAE Standard 140 tests, were designed to isolate
and evaluate the key influencing factors responsible for the discrepancies in results
between EnergyPlus and DeST. This included the loads algorithms and some of the
default input parameters. It was concluded that there is little difference between the
results from EnergyPlus and DeST (Figure  21) if the input values are the same or
equivalent despite there being many discrepancies between the heat balance
algorithms. DOE-2.1E can produce large errors for cases when adjacent zones have
very different conditions, or if a zone is conditioned part-time while adjacent zones
are unconditioned. This was due to the lack of a strict zonal heat balance routine in
DOE-2.1E, and the steady state handling of heat flow through interior walls and
partitions.

This comparison study did not produce another test suite, but rather a methodology to
design tests that can be used to identify and isolate key influencing factors that drive
the building thermal loads, and a process with which to carry them out. Figure  20
summarizes the method used to develop the tests and perform the comparisons. For all
test cases, EnergyPlus Version 7.0, DeST Version 2011-11-23 and DOE-2.1E114
were used. For all EnergyPlus tests, the CTF (Conduction Transfer Function) method
was used with a simulation time-step of 15 minutes.

Figure 20 ‐ Methodology to build the tests and perform the comparisons

26 
 
 Figure 21 ‐ Annual cooling and heating loads from the ASHRAE Standard 140-2007 tests

HVAC systems 
For  comparison  of  HVAC  systems  modeling,  a  comparative  test  method  is  used  mainly  for 
the inter‐program comparison. First, the HVAC system module calculations for each program 
are summarized, analyzed, and compared to identify differences in the solution algorithms 
and  main  assumptions  (Table  7).  The  component  models  are  important  parts  of  the  HVAC 
system  calculations  and  have  an  important  influence  on  the  HVAC  calculation  results.  The 
calculation  methods  and  the  main  assumptions  of  several  of  the  main  components  are 
discussed and compared in detail. As the HVAC control strategies of supply air temperature, 
supply  air volume,  and other parameters affect the operation of HVAC  systems, they have 
large  impacts  on  the  simulation  results.  The  basic  simulation  methods  for  the  control 
strategies  in  the  three  BEM  programs  (Table  8)  are  summarized  and  the  differences  are 
discussed. Secondly, the limitations of existing HVAC system calculation tests are discussed, 
and additional tests are designed to allow the HVAC systems to be compared in depth. CAV 
(constant  air  volume)  and  VAV  (variable  air  volume)  systems  are  tested  in  this  study  to 
analyze  the  HVAC  system  performance  and  control  strategies  under  different 
heating/cooling  load  ratios.  All  inputs  for  the  test  cases  are  kept  constant  where  possible. 
For the parameters that cannot have the same values, equivalent conversions are made, in 
the interest of consistency. The tests include the system‐side and plant‐side, so that the test 
process is similar to the real cases. Analytical tests are first conducted to make the load‐side 
calculations consistent, so the differences in the following calculation results can only come 
from  the  HVAC  system  module.  Through  the  CAV  tests  under  full  load  and  part  load 
conditions,  the  component  models  and  their  influence  on  the  calculation  results  are 
compared. Then, based on the VAV test cases, the control strategies used by the three BEM 
programs  are  analyzed  in  detail.  Finally,  a  case  study  based  on  a  real  office  building  is 
presented and differences in the simulation results between the three BEMPs are analyzed. 
Based  on  the  load  calculation  results,  drivers  of  the  differences  in  the  HVAC  side  are 
analyzed and discussed. Figure 22 shows the method used for the HVAC system simulation 
comparison. 

27 
 
  
Figure 22 Methodology of the HVAC calculation comparison 

Theoretical comparison 
The three BEMPs are reviewed in terms of HVAC simulation methods, and their advantages 
and  disadvantages  are  summarized.  Then,  focusing  on  the  main  HVAC  components  and 
HVAC  control  strategies,  the  differences  between  simulation  methods  (including  solution 
algorithms, modeling assumptions, and simplifications) are discussed and analyzed in detail.  
Integrated test cases 
Based on the review of existing HVAC system tests, an integrated test method is proposed 
and  used.  Due  to  the  similarity  of  EnergyPlus  and  DOE‐2  in  the  use  of  steady‐state  HVAC 
models, the test process is only applied to EnergyPlus and DeST. Two types of HVAC systems 
(CAV and VAV) are tested under different load conditions. Comparisons of each component 
model and control strategy are made and analyzed in detail. 
Case study with a real building 
Based  on  the  findings  of  the  previous  two  comparisons,  a  real  building  case  study  is 
conducted using the three simulation programs. The differences in the load‐side calculations 
are compared first. Then, on the foundation of the load‐side results, the errors in the HVAC 
system energy consumption results are compared and analyzed to ascertain the differences 
between measured and simulated results, for each program. 

28 
 
 Table 7 Summary of HVAC systems simulation 

Features  EnergyPlus  DeST  DOE‐2.1E 

Sub‐hourly,  auto‐adjusted  from  zone  time 

Time‐step  Fixed hourly  Fixed hourly 
step 

Advanced features  Energy Management Systems (EMS)  None  User functions 

DeST  separates  the  heating 

DOE‐2 is a program with sequential simulations. It has 
and  cooling  station  from  the 
Integrated  solution.  BUILDING,  SYSTEM,  one  subprogram  for  the  translation  of  users’  inputs 
supply side, dividing them into 
PLANT  are  integrated  and  controlled  by  the  (BDL  processor),  and  four  simulation  subprograms 
Simulation structure  two  modules:  equivalent  user 
integrated solution manager. Uses a predictor‐ (LOADS,  SYSTEMS,  PLANT  and  ECON).  LOADS, 
terminal  and  heating/cooling 
corrector approach.  SYSTEMS and PLANT are executed in sequence. Their 
station.  The  two  modules 
outputs are used as inputs to the ECON subprogram. 
iterate to calculate results. 

Algorithm  All the equipment models are forward, quasi‐steady models that use equipment/system performance curves 

For  new  types  of  HVAC  Cannot model new HVAC system types; 

For  complex  system  configurations, 
Limitations  systems,  the  equivalent  user  When  the  SYSTEM  subprogram  cannot  meet  the 
convergence sometimes is an issue 
terminal is limited  loads, space temperatures are estimated. 

Predefined  HVAC  templates: 

Flexible,  user  definable,  HVAC  templates 
System configurations  VAV,  CAV,  FCU,  VRF,  PTAC,  25 fixed system types with optional components 
covering most traditional designs 
PTHP 

29 
 
 Characteristics  of  equivalent 
Pressure calculation  Two types of pressure drop curves  User inputs 
user terminal 

Take terminals as different equipment models  Use  the  equivalent  user  Take terminals as different equipment models and the 

to get the information about flow rate, return  terminals  to  reflect  the  inputs  are  based  on  average  parameters,  then  the 
Terminal model 
water temperature and so on  average  of  terminal  flow  rate  overall features can be achieved 
and thermodynamic state 

successive  substitution  iteration  to  reconcile  The  zone  temperatures  from  the  previous  hour 
Interaction between the 
all  elements  using  the  Guass‐Seidell  calculation  are  used  to  approximate  the  heat  flow 
HVAC  equipment  and  equivalent user terminal curve 
philosophy  of  continuous  updating  with  across  internal  walls.  Temperature  histories  are  used 
the building loads 
predictive system energy balance method  in the calculation of equipment capacities. 

Handling  of  given 

According  to  the  control  strategy  defined  by  According  to  the  control  Model  them  as  if  all  the  sizes  operating  were 
types  of  equipment 
the users  strategy defined by the users  combined together into one large unit. 
with multiple sizes 
 

30 
 
 Table 8 Comparison of VAV system control strategies 

Program  Control  Description 

Supply  airflow  rate  stays  at  a  constant  minimum  value  for  normal  acting  dampers  and  modulates  higher  for 
EnergyPlus  Heating 
reserve acting dampers, supply air temperature may vary 
(normal acting damper) 
Cooling  Supply air temperature may vary, supply airflow rate varies 

An  optimization  technique  is  employed  to  search  for  the  best  supply  air  temperature;  when  the  range  of  best 
DeST  Heating & cooling  supply  air  temperature  (SAT)  exists,  determine  the  supply  air  volume  (SAV)  to  make  sure  the  air  flow  rate 
minimum; otherwise, determine the SAT to make deviation minimum 

The  actions  are  sequential:    1)  increase  supply  air  temperature,  2)  increase  the  baseboard  output  if  exist,3) 
Heating 
increase reheat coil output;3) increase supply air volume 
DOE‐2 
Cooling  The actions are sequential:1) reduce supply air temperature;2) increase the airflow rate  

31 
 
Based  on  the  comprehensive  test  cases  and  results  analysis,  the  main  findings  are  summarized  as 
follows: 
1. EnergyPlus,  DOE‐2.1E,  and  DeST  all  have  fundamental  capabilities  and  appropriate  modeling 
assumptions  for HVAC  system simulations. The results  from the comparative tests on component 
models show small differences, which are mainly due to the input settings and algorithms used in 
each  program.  Differences  between  the  total  energy  consumption  calculation  results  of  HVAC 
systems from DeST and EnergyPlus can be limited to 5%, if all component models are similar, and 
the same or equivalent inputs for the HVAC systems are used. It is found that the main influencing 
factors on HVAC discrepancies between DeST and EnergyPlus are the algorithms used for the HVAC 
component models and their control strategies. For the case that simulates the real building, using 
design  and  default  values  for  the  inputs  of  each  simulation  program,  the  errors  in  both  load 
calculations  and  HVAC  system  calculations  are  within  15%  of  the  measured  values.  This 
demonstrates good agreement between the simulation programs and the real building. 
2. EnergyPlus has more comprehensive component models than DOE‐2 and DeST. The three programs 
have consistent component models for pumps, fans, and boilers. The coil models in EnergyPlus and 
DeST  are based on engineering equations while the  coil  model  in DOE‐2  is based on  assumptions 
and  empirical  data.  The  influences  of  load  ratio,  condenser  inlet  water  temperature,  and 
evaporator outlet water temperature on the chiller efficiency are considered in all three programs. 
Three chiller performance curves with user‐specified coefficients are used in EnergyPlus and DOE‐2, 
while  one  hard‐wired  performance  curve  is  used  in  DeST.  In  EnergyPlus  and  DOE‐2.1E,  the  fan 
power of the cooling tower is related to the load ratio, so the fan can cycle during a particular hour 
if the load is small. In DeST, the fan power draw remains constant whenever the cooling tower has a 
load for any particular hour. 
3. To  complete  a  comprehensive  comparison  of  the  three  different  simulation  programs,  several 
requirements  are  needed:  1)  the  test  cases  should  be  broad  enough  to  cover  most  modeling 
features; 2) the test cases should be detailed enough to isolate influencing factors; 3) special cases 
should  be  designed  to  test  the  unique  limitations  of  the  programs.  Based  on  the  current 
development  of  HVAC  system  tests,  a  testing  concept  is  introduced  in  this  report  to  develop  a 
better method of comparison. As each component in a HVAC system is connected and influenced 
by one another, the whole HVAC system should be considered when the comparison is conducted. 
This  means  that  both  air‐side  and  plant‐side  components  should  be  tested  together.  Imposing 
steady‐state conditions makes it possible to compare each component model in detail and calculate 
the analytical results. Considering the whole system makes the test process more practical. 

Future Research 
The  research  findings  from  the  project  directly  feed  into  the  two  new  CERC‐BEE  Simulation 
Projects:  Integrated  Design  and  Operation  for  Very  Low  Energy  Buildings,  and  Human  Behavior. 
Both  projects  aim  to  further  understanding  driving  forces  of  building  performance,  integrated 
building design and operation, and occupant energy‐related behavior, and to create new scientific 
knowledge,  toolkit,  guideline,  and  case  studies  for  building  designers,  engineers,  researchers  and 
policy makers to improve design and operation of high performance buildings. 

32 
 
Acknowledgement 
This  work  was  sponsored  by  the  U.S.  Department  of  Energy  under  the  U.S.‐China  Clean  Energy 
Research Center for Building Energy Efficiency, and the China Ministry of Housing and Urban‐Rural 
Development and the Ministry of Science & Technology. The energy modeling program comparison 
work was sponsored by the China Sustainable Energy Program, Energy Foundation. Special thanks 
to the U.S. and China CERC‐BEE management, our industry partners, the Industry Advisory Board, 
and the Technical Advisory Committee for the strong support on the project.  .
   

33 
 
This work was supported by the Assistant Secretary for Energy Efficiency and Renewable
Energy, Building Technologies Program, of the U.S. Department of Energy under Contract No.
DE-AC02-05CH11231.
Appendix A – List of Publications 
Journal Articles 

1. T. Hong, W.K. Chang, and H.W. Lin. A Fresh Look at Weather Impact on Peak Electricity Demand 
and Energy Use of Buildings Using 30‐Year Actual Weather Data, Applied Energy, Vol.111, 2013.  
2. H.W. Lin, T. Hong. On Variations of Space‐heating Energy Use in Office Buildings, Applied Energy, 
Vol. 111, 2013. 
3. D.  Zhu,  T.  Hong,  D.  Yan,  C.  Wang.  A  Detailed  Comparison  of  Three  Building  Energy  Modeling 
Programs: EnergyPlus, DeST, and DOE‐2.1E. Journal of Building Simulation, 2013.  
4. W.K. Chang and T. Hong. Statistical Analysis and Modeling of Occupancy Patterns in Open‐Plan 
Offices using Measured Lighting‐Switch Data, Journal of Building Simulation, 2013 
5. X. Zhou, D. Yan, T. Hong, and D. Zhu. Building energy modeling programs comparison: Research 
on HVAC systems simulation, Journal of HV&AC, 2013. In Chinese.  
6. D.  Zhu,  D.  Yan,  C.  Wang,  T.  Hong.  Comparison  of  Building  Energy  Simulation  Programs:  DeST, 
EnergyPlus and DOE‐2. Building Science, 28(S2):213‐222. 2012. In Chinese. 
7. H.W. Lin, M.S. Zhen, T.  Hong. Analysis of  impact of occupant behavior on variations of  energy 
use in office buildings. Journal of Energy & HVAC Engineering, 76:31‐37. 2012. In Chinese. 
8. C.  Wang,  D.  Yan,  Y.  Jiang.  A  novel  approach  for  building  occupancy  simulation.  Journal  of 
Building Simulation, 2011. 
9. C.  Peng,  D.  Yan,  R.  Wu,  C.  Wang,  X.  Zhou,  Y.  Jiang.  Quantitative  description  and  simulation  of 
human behavior in residential buildings. Journal of Building Simulation, 2012. 
10. X.  Zhou,  D.  Yan,  X.  Ren,  T.  Hong.  Data  analysis  and  model  study  about  lighting  energy 
consumption  in  large  offices.  Submitted  to  the  CERC‐BEE  issue  of  Building  Research  and 
Information, 2012. 
11. X. Ren, S. Hu, D. Yan, C. Peng. Development of energy models for residential appliances based 
on measured data. Building Science, 28(S2), 2012. In Chinese. 
Conference Papers 

1. T. Hong, WK Chang. A Sensitivity Study of Building Performance Using 30‐Year Actual Weather 
Data. IBPSA Building Simulation, France, 2013. 
2. W.K.  Chang,  T.  Hong.  Statistical  Analysis  of  Occupancy  Behavior  in  Open  Office  Spaces  Using 
Measured Lighting Switch Data. First Asia Building Simulation Conference ASim, Shanghai, 2012. 

34 
 
 3. L.  Wang,  T.  Hong.  Modeling  and  Simulation  of  HVAC  Faulty  Operations  and  Performance 
Degradation  due  to  Maintenance  Issues.  First  Asia  Building  Simulation  Conference  ASim, 
Shanghai, 2012. 
4. H.W.  Lin,  T.  Hong.  Occupant  Behavior:  Impact  on  Energy  Use  of  Private  Offices.  First  Asia 
Building Simulation Conference ASim, Shanghai, 2012. 
5. X. Zhou, D. Yan, X. Ren, T. Hong. Data analysis and modeling of lighting energy use in large office 
buildings. First Asia Building Simulation Conference ASim, Shanghai, 2012. 
6. H.W. Lin, T. Hong. An In‐Depth Analysis of Space Heating Energy Use in Office Buildings. ACEEE 
Summer Study on Building Energy Efficiency, Asilomar, 2012. 
7. M. Levine, W. Feng, K. Jing, T. Hong, N. Zhou. A Retrofit Tool for Improving Energy Efficiency of 
Commercial Buildings. ACEEE Summer Study on Building Energy Efficiency, Asilomar, 2012. 
8. D.  Zhu,  C.  Wang,  D.  Yan,  T.  Hong.  A  Comparison  of  DeST  to  EnergyPlus.  Annual  China  HVAC 
Simulation Conference, Beijing, 2011. In Chinese. 
9. Q. Shen, J. Xia, Y. Jiang. A case study on energy performance and retrofitting technologies for a 
high  standard  resort  hotel.  the  7th  International  Symposium  of  Heating,  Ventilation  and  Air‐
conditioning, Shanghai, 2011. 
10. X. Ren, D. Yan, X. Zhou. A Study of Lighting Energy Consumption Model of Office Buildings. First 
Asia Building Simulation Conference ASim, Shanghai, 2012. 
11. G.  Deng,  D.  Yan,  Y.  Rui.  Commissioning  and  analysis  of  usage  patterns  of  water‐source  heat 
pump system based district heating systems. Annual China HVAC Simulation Conference, 2012. 
In Chinese. 
Technical Reports 

1. X.  Zhou,  T.  Hong,  D.  Yan.  Comparison  of  building  energy  modeling  programs:  HVAC  Systems. 
2013. LBNL Report, submitted, 2013. 
2. D. Zhu, T. Hong, D. Yan, C. Wang. Comparison of building energy modeling programs: Building 
Loads. LBNL Report 6034E, 2012. 
   

35 
 
Appendix B – List of Workshops and Major Activities 
 

Workshops/Activities  Date  and  Summary 

Location 

CERC Conference  1/19‐20/2011,  CERC Kickoff meeting 

DOE, USA 

CERC‐BEE  3/24‐25/2011,  Research  teams  from  both  countries  and  CERC 

Conference  management  attended  to  discuss  the  research 
Tsinghua, China  plan and scope of work etc. 

Project  Kickoff  7/21/2011,  About 24 people attended the meeting, including 

Meeting  LBNL, USA  the  LBNL  and  Tsinghua  research  teams,  industry 
partners,  DOE,  and  researchers  from  other 
institutes.  Research  objectives,  plan  and 
deliverables were presented and discussed. 

Industry  Advisor  8/16/2011,  The  LBNL  Team  presented  research  progress  and 
Board Meeting  addressed comments and questions from IAB. 
LBNL, USA 

CERC  Management  9/21/2011,  The  Tsinghua  Team  reported  research  progress 

visited Tsinghua  and  had  broader  discussion  with  US  CERC 
Tsinghua, China  directors, Robert Marley and Michaela Martin, to 
plan and improve the next step. 

Project  Progress  12/27/2011,  The LBNL and Tsinghua research teams had a one‐

Meeting  day  meeting  to  exchange  and  discuss  research 
Tsinghua, China  progress and address issues. 

China  CERC‐BEE  1/10/2012,  China  CERC‐BEE  Annual  Review.  The  Tsinghua 

Annual Review   ShenZhen,  Team presented the research work and addressed 
China  comments and questions from reviewers. 

US  CERC‐BEE  2/14/2012,  The LBNL Team presented joint research progress 

Technical Review  DOE, USA  and  addressed  comments  and  questions  from 
CERC reviewers and management. 

Industry  Advisor  3/20/2012,  The LBNL Team presented joint research progress 

Board Meeting  and addressed comments and questions from IAB. 
LBNL, USA 

36 
 
CERC  Simulation  7/16/2012,  The LBNL and Tsinghua research teams presented 
Workshop  the research findings and exchanged ideas with 20 
Hong  Kong,  participants  from  design  firms,  government 
China  agencies, and universities. 

Joint  CERC‐BEE  7/18‐20,  Sanya,  The  LBNL  and  Tsinghua  Teams  presented  the 
Annual Conference  China  research  results  and  addressed  comments  and 
questions from reviewers and CERC management. 

CERC  Simulation  11/26/2012,  During  the  first  Asia  IBPSA  Building  Simulation 
Workshop  Conference,  the  LBNL  and  Tsinghua  research 
Shanghai, China  teams  presented  the  research  findings  and 
exchanged ideas with 50 participants from design 
firms,  government  agencies,  universities,  and 
other organizations. 

Project  Progress  11/29/2012,  The LBNL and Tsinghua research teams had a one‐

Meeting  day  meeting  to  exchange  and  discuss  research 
Tsinghua, China  progress and address issues. 

CERC  Technical  12/3/2012,  The LBNL research presented the research results 

Review  and addressed comments and questions from the 
LBNL, USA  CERC TAC (Technical Advisor Committee). 

 
   

37 
 
Appendix C – Selected Publications 
 

38 
 
 Applied Energy 111 (2013) 333–350

Contents lists available at SciVerse ScienceDirect

Applied Energy
journal homepage: [Link]/locate/apenergy

A fresh look at weather impact on peak electricity demand and energy

use of buildings using 30-year actual weather data
Tianzhen Hong a,⇑, Wen-Kuei Chang b, Hung-Wen Lin b
a
Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA
b
Green Energy and Environment Research Laboratories, Industrial Technology Research Institute, Hsinchu, Taiwan, ROC

h i g h l i g h t s

 Weather has a significant impact on both the peak electricity demand and energy use.
 Weather impact varies with building type, building efficiency level, and location.
 Simulated results using TMY3 weather data can under or over estimate those of AMY.
 It is crucial to assess performance of buildings using long-term actual weather data.
 Findings enable building stakeholders to make better decisions on weather impact.

a r t i c l e i n f o a b s t r a c t

Article history: Buildings consume more than one third of the world’s total primary energy. Weather plays a unique and
Received 20 November 2012 significant role as it directly affects the thermal loads and thus energy performance of buildings. The tradi-
Received in revised form 4 May 2013 tional simulated energy performance using Typical Meteorological Year (TMY) weather data represents the
Accepted 7 May 2013
building performance for a typical year, but not necessarily the average or typical long-term performance as
buildings with different energy systems and designs respond differently to weather changes. Furthermore,
the single-year TMY simulations do not provide a range of results that capture yearly variations due to
Keywords:
changing weather, which is important for building energy management, and for performing risk assess-
Actual meteorological year
Building simulation
ments of energy efficiency investments. This paper employs large-scale building simulation (a total of
Energy use 3162 runs) to study the weather impact on peak electricity demand and energy use with the 30-year
Peak electricity demand (1980–2009) Actual Meteorological Year (AMY) weather data for three types of office buildings at two
Typical meteorological year design efficiency levels, across all 17 ASHRAE climate zones. The simulated results using the AMY data
Weather data are compared to those from the TMY3 data to determine and analyze the differences. Besides further dem-
onstration, as done by other studies, that actual weather has a significant impact on both the peak electricity
demand and energy use of buildings, the main findings from the current study include: (1) annual weather
variation has a greater impact on the peak electricity demand than it does on energy use in buildings; (2) the
simulated energy use using the TMY3 weather data is not necessarily representative of the average energy
use over a long period, and the TMY3 results can be significantly higher or lower than those from the AMY
data; (3) the weather impact is greater for buildings in colder climates than warmer climates; (4) the
weather impact on the medium-sized office building was the greatest, followed by the large office and then
the small office; and (5) simulated energy savings and peak demand reduction by energy conservation mea-
sures using the TMY3 weather data can be significantly underestimated or overestimated. It is crucial to run
multi-decade simulations with AMY weather data to fully assess the impact of weather on the long-term
performance of buildings, and to evaluate the energy savings potential of energy conservation measures
for new and existing buildings from a life cycle perspective.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction encing factors on building energy performance, including climate,

building envelope, building equipment, operation and mainte-
Buildings consume more than one third of the world’s total pri- nance, occupant behavior, and indoor environmental conditions.
mary energy. The IEA Annex 53 [1] identified and studied six influ- Among these influencing factors, climate plays a unique and signif-
icant role. Weather contributes directly and significantly to the
⇑ Corresponding author. Tel.: +1 510 4867082; fax: +1 510 4864089. variations of thermal loads and energy use of HVAC (heating, ven-
E-mail address: thong@[Link] (T. Hong). tilation, and air conditioning) systems, lighting (for buildings with

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
[Link]
334 T. Hong et al. / Applied Energy 111 (2013) 333–350

daylighting controls), and energy production from solar-based model, can be integrated with a building thermal model to obtain
renewable systems. In residential and commercial buildings in directly the probability distribution of building performance to
the US, heating and cooling accounts for more than 40% of end- investigate the uncertainty caused by the random meteorological
use energy demand. It is important to understand and estimate processes and internal heat gains [6].
the impact of weather on the long-term performance of buildings A new online weather data service with immediate access to
in order to support policy making, and to allow building operators precision, localized weather history, current conditions and fore-
and owners to respond better to climate changes in terms of build- casts are presented by Keller and Khuen [7]. Localized weather
ing energy supply and demand. Additionally, considering the im- data is created by integrating all available ground station observa-
pact of yearly variations in weather can improve the evaluation tions with high-resolution datasets from NOAA (National Oceanic
of investment risks of energy conservation measures (ECMs) for and Atmospheric Administration). Both historical and forecast time
new and existing buildings by taking into account their life-cycle series data are available for direct user access and application/sys-
energy and cost savings. tem access through Web Data Services and API interfaces.
The accuracy of building energy simulations and economic Selecting appropriate weather data to be used in building
assessments of renewable energy systems depend on the availabil- performance simulation is important. The use of inappropriate
ity of reliable weather data. There are two primary sources of weather data can result in large discrepancies between the pre-
weather data that are used to generate weather data files used in dicted and measured performance of buildings. In the late 1970s,
building simulation: measured weather data using physical sen- Freeman [8] evaluated how well TMY represents actual long-
sors and observations, and simulated data using mathematical term weather data based on simulations of an active residential
weather models. Various methods to generate annual hourly space solar heating and cooling system for six US climates, Albu-
weather data have been developed in the past. Such weather data querque, Fort Worth, Madison, Miami, New York, and Washing-
include the Typical Meteorological Year (TMY), the test reference ton DC. High variability of the weather and solar heating
year (TRY), the weather year for energy calculation (WYEC), the de- system performance year to year was noted. Crawley [9] com-
sign reference year (DRY), as well as the synthetically modeled pared the influence of the various weather data sets on simu-
meteorological year (SMY). However, the lack of long-term weath- lated annual energy use and cost. Using different weather data
er records usually limits the generation of typical annual weather sets can cause significant variations in annual energy consump-
data files in any format [2]. tion and cost from simulation results. The results show that
A TMY weather file contains hourly values of solar radiation the TMY and the WYEC data sets represent the closest typical
and meteorological elements for a 1-year period. The 12 typical weather patterns. Simulated results using the TMY weather data
meteorological months (TMMs) are selected from various calen- provides the average/typical energy use for buildings, but the
dar months in a multi-year weather database. The criteria for peak electricity demand predictions and uncertainty analyses
TMM selection is based on the statistical analysis and evaluation based on TMY are often not reliable because a single year cannot
of four weather parameters: the ambient dry-bulb temperature, capture the full variability of the long-term climate change [10].
the dew-point temperature, the wind speed and the global solar In view of the long-term climate change, the time period as-
radiation. Algorithms are used to smooth discontinuities from signed for TMY selection should include the most recent meteo-
the data to avoid drastic changes between two adjacent months rological data and should be reasonably long to reflect well the
selected from different years. The first generation of TMY weath- weather variations [11]. Most of the available TMY weather data
er data for the US is derived from the 1952–1975 SOLMET/ER- are from weather stations located at airports. It is possible to
SATZ database, while the second generation of data (TMY2) is create a new TMY file localized to a building location by inte-
derived from the 1961–1990 National Solar Radiation Database grating the weather station observations with gridded reanalysis
(NSRDB) covering 239 US locations. The latest, third generation data. However, there are limited complete weather data col-
data (TMY3) is derived from the 1976–1990 and 1991–2005 Na- lected by weather stations over 15–30 years, so TMY data is only
tional Solar Radiation Data Base (NSRDB). TMY3 covers 1020 US available for only 1020 locations. Furthermore, some of the TMY
locations. TMY, TMY2 and TMY3 data sets cannot be used inter- weather data files were created up to 20 years ago. They are less
changeably because of differences in the data structure such as representative of the typical present day climate and do not de-
time (solar vs. local), formats, elements, and units. The intended scribe extreme weather conditions. Compared with the TMY
use of TMY weather data is for computer-based building perfor- weather data, the AMY is created from actual hourly data for a
mance simulations of solar energy conversion systems and build- particular calendar year. AMY weather data is particularly useful
ing systems to facilitate performance comparisons of different for modeling years with extremes in weather and verifying the
system types, configurations, and locations in the US and its ter- energy performance of buildings. However, as with the TMY
ritories. Because they represent typical rather than extreme con- weather data, the AMY weather data needs to be chosen as close
ditions, they are not suited for designing systems to meet the to the building location as possible.
worst-case conditions occurring at a location [3]. For the calcula- The potential impacts of various types of weather forecast mod-
tions of peak cooling and heating loads of buildings, and sizing els, weather data, and building prototypes have been studied from
HVAC equipment, design day weather data are used. Design- a number of perspectives. A prototypical small office building was
day weather data tend to represent more extreme weather con- modeled operating at three energy efficiency levels, using typical
ditions in order to guarantee that HVAC systems can meet peak and extreme meteorological weather data for 25 locations, to study
loads for most of the time during their life cycle. Various meth- various predicted climate change and heat island scenarios [12].
ods are used to create design-day weather data [4]. The largest change to the annual energy use due to climate change
As TMY data may not be available for some cities or sites, SMY was seen in the temperate, mid-latitude climates, where there was
weather data provide a practical and useful alternative. SMY a swapping of energy use from heating to cooling. The heating en-
weather data can be generated from monthly average or total val- ergy was reduced by more than 25% and cooling energy was in-
ues of weather parameters using stochastic models and auto- creased by up to 15%. The TMY weather data provides more
regressive moving average processes to represent the seasonal localized and comprehensive climate indicators to further support
and daily weather variations [5]. Such stochastic weather models the HVAC system design in buildings [3,13]. The space cooling
can be used to generate AMY weather data for use in deterministic plays a major role in determining the magnitude and timing of
building simulations, or together with a stochastic internal loads peak electricity demand. The archived General Circulation Model
 T. Hong et al. / Applied Energy 111 (2013) 333–350 335

(GCM) projections were statistically downscaled to the site scale, (1) How significant is the weather impact on both the peak elec-
which were then used for input to building cooling and heating tricity demand and building energy use?
simulations to study the California specific impact of global warm- (2) Does the simulated building energy use using the TMY3
ing on building energy consumption [14]. The IPCC’s different car- weather data represent the average or typical energy use
bon emission scenarios predict that climate change will lead to a over a 30-year period?
25–50% increase in space cooling electricity use over the next (3) Building simulation results from which climates are greater
100 years. Under the worst case carbon emission scenario the total affected by using different weather data sets?
energy consumption will increase between 8% and 20%. The energy (4) What types of office buildings are subject to the greatest
performance of an office building in Hong Kong, using multi-year impact of weather?
weather data sets was simulated to investigate the diversity in (5) What are the risks from using the TMY3 weather data in build-
simulation predictions [15]. The results concluded that the choice ing simulations to evaluate the energy savings and electricity
of weather data sets was not crucial for the comparative energy demand reduction of energy efficiency technologies?
studies during the initial design stage. However, it becomes impor-
tant to select a particular standard weather year data set when Through better understanding of which building technologies
absolute energy consumption data are required. Similar studies and system designs are more sensitive to yearly weather variation,
on office buildings were conducted in five major climate zones in building designers, owners, operators, and policy makers can make
China by using multi-year weather databases as well as TMY data more informed decisions on energy efficiency implementations to
[16–18]. The results showed a decreasing trend for heating loads reduce peak electricity demand and building energy use.
and an increasing trend for cooling loads due to predicted climate
change. The monthly loads and energy use profiles calculated using
the TMY and long-term means profiles fell well within the maxi- 2. Methodology
mum and minimum ranges of the 30-year individual predictions.
It was concluded that building performance predictions using 2.1. Overview
TMY weather data can be used in comparative energy efficiency
studies. To study the impact of weather on building performance, the
In recent years, various types of weather data have been used in most typical commercial buildings located in typical climate zones
building simulation to evaluate energy performance and demand are the natural starting point. The US 2003 Commercial Building
response. Accurate estimation of building performance relies on Energy Consumption Survey (CBECS) [20] indicates that office
the appropriate selection of accurate weather data. The quality of buildings are the most common building type, comprising the larg-
weather data and their impact on building cooling and heating est floor area, and consuming the most energy in the commercial
loads and energy consumption were studied by comparing three building sector. Therefore, the prototypical office buildings with
weather datasets for a specific location for the calendar year three different sizes at two design efficiency levels for 17 climates
2010 [19]. The three sources of data included site measured data are chosen from the PNNL’s prototype buildings. Three building
and AMY weather data provided by two vendors. Key weather vari- sizes represent large, medium, and small office buildings based
ables from the three datasets were compared statistically, and on the statistics of the 2003 CBECS. The 17 climates represent all
building loads and energy use were simulated using EnergyPlus of the ASHRAE climate zones. The two design efficiency levels cor-
version 6.0. The study concluded that the maximum difference in respond to the ASHRAE Standard 90.1-2004 and 2010. ASHRAE
individual hourly weather variables can be as high as 90%, annual standard 90.1 is an energy standard providing prescriptive and
building energy consumption can vary by ±7%, while monthly mandatory requirements for energy efficiency levels of major
building loads can vary by ±40% when using different weather building systems including building envelopes (opaque construc-
datasets. tion and fenestration), lighting systems, service water heating,
Using TMY weather data to calculate the energy use in buildings and HVAC systems. The 90.1-2004 standard was published in
aims to represent the average or typical values. However, different 2004 and represented the minimum performance of recently built
types of buildings with different energy service systems and oper- new constructions that comply with the standard. While 90.1-
ation strategies have different responses to weather. Furthermore, 2010 [21] represents more efficient designs, with about 30% energy
a single set of energy use results from TMY simulations does not savings over 90.1-2004 [22].
provide the range of variations due to the change of weather from The TMY3 weather data and 30 years of AMY weather data
year to year. The typical life of a building is more than 50 years; (1980–2009) are used in the building performance simulations.
therefore the assessment of long-term building performance be- The simulations were run using EnergyPlus 7.1. There was a total
comes very important. TMYs are often recommended to be used of 3162 simulation runs: 3 office building types, 2 design efficiency
in building simulations to evaluate and compare performance of levels, 17 climates, and 31 weather files. The HVAC equipment is
design alternatives under the assumption that energy savings from autosized by EnergyPlus to meet peak cooling and heating loads
a design alternative would not vary noticeably with yearly weather based on the 2009 ASHRAE design day weather data. The structure
variations. This assumption is not necessarily true. Although previ- of the simulation runs is illustrated in Fig. 1. Performance metrics,
ous studies have demonstrated actual weather has a significant im- including building total source energy (including all end uses), HVAC
pact on peak electric demand and energy use in buildings, there are source energy (including end uses of cooling, heating, and ventila-
limited studies that focus on investigating the sensitivity of energy tion), and peak electricity demand, of each simulation run were ex-
savings and peak demand reduction of energy conservation mea- tracted from the EnergyPlus output reports. The performance
sures to the yearly variation of weather, using multi-decade AMY metrics of each AMY run were then compared with those of the cor-
weather data across a complete coverage of climate zones for typ- responding TMY3 run to calculate the percentage changes, equal to
ical commercial buildings. This study aims to address that gap in 100  (AMY_Results–TMY3_Results)/TMY3_Results, as indicators
the literature. of deviations from the TMY3 results. The ranges of these percentage
This study does not touch the topics of previous studies on im- changes are graphed as key results for analysis and discussions. To
pacts of long-term climate change or local heat island effects on filter out the extreme weather years, the variation ranges excluding
building performance; instead it focuses on providing insights to those of the top three and the bottom three weather years were
the following important questions: overlapped on the same graphs. The variation ranges of the percent-
336 T. Hong et al. / Applied Energy 111 (2013) 333–350

Fig. 1. The structure of simulation runs.

Table 1
Climate zone classification based on ASHRAE Standard 90.1-2010.

Climate zone Climate type Criteria Representative city

1A Very Hot–Humid 5000 < CDD10 °C Miami, USA
1B Very Hot–Dry 5000 < CDD10 °C Riyadh, Saudi Arabia
2A Hot–Humid 3500 < CDD10 °C 6 5000 Houston, USA
2B Hot–Dry 3500 < CDD10 °C 6 5000 Phoenix, USA
3A Warm–Humid 2500 < CDD10 °C 6 3500 Memphis, USA
3B Warm–Dry 2500 < CDD10 °C 6 3500 EI Paso, USA
3C Warm–Marine CDD10 °C 6 2500 and HDD18 °C 6 2000 San Francisco, USA
4A Mixed–Humid CDD10 °C 6 2500 and 2000 < HDD18 °C 6 3000 Baltimore, USA
4B Mixed–Dry CDD10 °C 6 2500 and 2000 < HDD18 °C 6 3000 Albuquerque, USA
4C Mixed–Marine 2000 < HDD18 °C 6 3000 Salem, USA
5A Cool–Humid 3000 < HDD18 °C 6 4000 Chicago, USA
5B Cool–Dry 3000 < HDD18 °C 6 4000 Boise, USA
5C Cool–Marine 3000 < HDD18 °C 6 4000 Vancouver, Canada
6A Cold–Humid 4000 < HDD18 °C 6 5000 Burlington, USA
6B Cold–Dry 4000 < HDD18 °C 6 5000 Helena, USA
7 Very Cold 5000 < HDD18 °C 6 7000 Duluth, USA
8 Subarctic 7000 < HDD18 °C Fairbanks, USA

age changes of building total source energy, HVAC source energy and weather data and the other is the AMY weather data. The TMY
peak electricity demand give a clear picture on how the AMY results weather data is usually used for annual energy simulations during
differ from the TMY3 results. The smaller the range of difference, the the building design process, either to evaluate the energy and cost
closer of TMY3 results to AMY results. effectiveness of design alternatives, to demonstrate code compli-
To investigate the weather impact on energy savings and de- ance, or to calculate credit points towards building rating systems
mand reduction of building technologies, two office models under or utility incentive programs. The AMY weather data, containing
two design efficiency levels (ASHRAE standard 90.1-2004 and 90.1- measured data for a particular year, is usually used in simulations
2010) were simulated using the TMY3 and 30-year AMY weather post occupancy to verify and diagnose the actual building energy
files. The energy savings and demand reductions of the 90.1- performance. The AMY weather data can be obtained from several
2010 models over the corresponding 90.1-2004 models were sources, including Weather Bank, National Climatic Data Center
determined using the same TMY3 or AMY weather files. (NCDC), Weather Source, Weather Analytics, and Meteonorm.
Furthermore, values of key weather parameters, such as annual Weather Bank maintains hourly and daily historical data records
average ambient air temperature, global horizontal solar radiation, from every National Weather Service reporting station in the Uni-
and heating and cooling degree days, were extracted from the ted States, as well as other locations around the world. The weath-
EnergyPlus weather statistics (stat) files and used to identify po- er data are archived on a real-time basis and updates are made
tential variation patterns and trends. hourly. NCDC is the world’s largest active archive of weather data.
In this study, source energy (also referred to as primary energy) The Integrated Surface Database (ISD) consists of global hourly and
is used because it considers the energy loss during energy genera- synoptic observations compiled from numerous sources. Currently
tion, transmission, and distribution. EnergyPlus calculates the there are over 11,000 stations active and updated daily in the data-
source energy by multiplying the calculated site energy with corre- base [23]. Weather Source provides historical and real-time digital
sponding source factors, which depend on types of energy sources weather information for tens of thousands of locations across the
and building location. US and around the world. Weather Analytics [7] provides site-spe-
cific TMY and AMY weather files based on the last 30 years of
2.2. Weather data hourly data. The files combine hourly weather station observations
and the new NOAA reanalysis data sets. Meteonorm is a weather
In general, two kinds of weather data packaged in weather files data generation tool. It integrates a climate database, a spatial
are used in building performance simulation. One is the TMY interpolation tool and a stochastic weather generator. The typical
 T. Hong et al. / Applied Energy 111 (2013) 333–350 337

Table 2 prototype buildings [25] covering 80% of the commercial building

Commercial reference buildings. floor area in the United States for new construction. These proto-
Building type Subtype type buildings were derived from the US Department of Energy
Offices Small office; medium office; large office (USDOE) [26] but with substantial modifications based on exten-
Retails Stand-alone retail; strip mall sive inputs from ASHRAE 90.1 Standing Standards Project Commit-
Schools Primary school; secondary school tee members and other building industry experts. The prototype
Hospitals Outpatient healthcare; hospital models include the 16 building types in 17 climate locations for
Hotels Small hotel; large hotel
Restaurants Quick service restaurant; full service restaurant
three editions of ASHRAE Standard 90.1 (90.1-2004, 90.1-2007
Apartments Mid-rise apartment; high-rise apartment and 90.1-2010). Table 2 summarizes the building types. The Ener-
Others Warehouse (non-refrigerated) gyPlus models of these buildings are available; including Energy-
Plus model input files (.idf) and output files (.html). The
description of the building, HVAC systems, internal loads, operat-
years with hourly or 1-min time resolution can be calculated for ing schedules, and other key model inputs are summarized in
any site [24]. scorecard spreadsheet files that are also available from the web
In this study, the weather data for 17 climate zones, including site. The detailed methodology and modeling strategy used to de-
the 30-year AMY weather files covering 1980–2009 from Weather velop these prototype models as well as the energy and cost saving
Analytics and the TMY3 weather data, were used in the simula- analysis is presented in [22].
tions to investigate the weather impact on building performance. From these prototype buildings, the three types of office build-
Table 1 lists the climate type, criteria, and representative cities ings with different sizes, small, medium and large, were chosen for
for the 17 climates – major US cities except Riyadh in Saudi Arabia this study. Office buildings represent the most typical commercial
and Vancouver in Canada. buildings in the United States in terms of buildings numbers and
total floor area [20]. The large-size office building is served by a
2.3. Prototype buildings central built-up variable air volume (VAV) system with a central
plant. The medium office has packaged VAV systems, and the small
To calculate the impact of ASHRAE Standard 90.1, researchers at office has packaged single zone systems. The key features of these
Pacific Northwest National Laboratory (PNNL) created a suite of 16 office buildings are summarized in Table 3. The EnergyPlus models

Table 3
Summary of key features of the three types of office buildings.

Large-size office Medium-size office Small-size office

Total floor area (m2) 46,320 4980 510
Number of stories 12 3 1
% Perimeter zone area 30% 40% 70%
Envelope
Window-wall-ratio (WWR) 40% 33% 24.4% for South and 19.8% for the
other three orientations
Walls, roofs, floors: U-factor ASHRAE 90.1 Requirements, nonresidential; walls, above-grade, steel-framed; roofs, insulation entirely above deck
Windows: U-factor and SHGC ASHRAE 90.1 requirements ASHRAE 90.1 requirements ASHRAE 90.1 requirements
nonresidential nonresidential; vertical glazing, 31.1– nonresidential; vertical glazing, 20–
40%, U fixed 30%, U fixed
HVAC systems
System type Central built-up VAV systems Packaged VAV systems Packaged single zone systems
Heating source Gas boiler Gas furnace Air-source heat pump with gas
furnace as back up
Cooling source Water-cooled centrifugal chillers Air-cooled direct expansion Air-source heat pump
Air distribution and terminal units VAV terminal box with hot-water VAV terminal box with hot-water No terminal unit
reheat coil, minimum damper position reheat coil, minimum damper position
set at 30% set at 30%
Thermostat setpoint 24 °C Cooling/21 °C Heating 24 °C Cooling/21 °C Heating 24 °C Cooling/21 °C Heating
Air-side economizer Applicable based on 90.1 Applicable based on 90.1 None
Internal loads
Average lighting power density (W/m2) 90.1-2004: 10.76 90.1-2004: 10.76 90.1-2004: 10.76
90.1-2010: 8.99 90.1-2010: 8.87 90.1-2010: 9.15
Average plug-load power density (W/m2) 7.8 8.07 6.78
Average occupant density (m2/person) 18.6 18.6 16.6
Operating schedules
Lighting, plug-loads, and occupants

Misc.
Exterior lighting peak power (W) 90.1-2004: 62787 90.1-2004: 14385 90.1-2004: 1634
90.1-2010: 43305 90.1-2010: 7476 90.1-2010: 896
338 T. Hong et al. / Applied Energy 111 (2013) 333–350

for the three office buildings in 17 climates based on ASHRAE Stan- 3. Results and discussion
dard 90.1-2004 and 90.1-2010 were downloaded and converted for
use with EnergyPlus version 7.1. The 90.1-2010 models represent 3.1. Variations of weather data
high energy-efficiency designs, with better insulation and win-
dows, more efficient lighting and HVAC systems, exceeding the Variations of weather data and climate zone classification for
performance of the 90.1-2004 models by approximately a 30% each of the 17 cities based on the annual HDD18 (Heating Degree
reduction in site energy use. Days with base temperature of 18 °C) and CDD10 (Cooling Degree
Days with base temperature of 10 °C) of the AMY data from 1980 to
2009 are illustrated in Fig. 2. The climate zones displayed in Fig. 2
2.4. Simulation engine correspond to the criteria listed in Table 1. It can be seen that most
cities do not belong to only one climate zone. For the 30-year per-
EnergyPlus [27] version 7.1, released in June 2012, was used iod, the climates of some cities vary across two zones and some
to perform the building simulations. EnergyPlus is developed even across three or more zones. For example, Fairbanks exhibits
by USDOE as a new generation building energy modeling pro- climatic conditions indicative of the very cold Climate Zone 7
gram that builds upon the most popular features and capabilities and the subarctic Climate Zone 8, while Helena shows conditions
of BLAST [28] and DOE-2 [29]. EnergyPlus has innovative simu- typical of five climate zones: the cool-humid 5A, the cool-dry 5B,
lation capabilities including sub-hourly time steps, an integrated the cool-marine 5C, the cold-humid 6A, and the cold-dry 6B. The
solver for system models with a zone heat balance model, and spread of climate zones for a city based on 30-year AMY weather
user definable and configurable HVAC systems and components. data is a good indicator of weather change year-over-year, which
It calculates space temperature, occupant thermal comfort, cool- cannot be represented by a single-year TMY3 weather data file.
ing and heating loads, HVAC equipment sizes, energy consump- Therefore, running simulations using multi-decade AMY weather
tion, utility cost, air emissions, water usage, renewable energy, data is necessary to evaluate fully the effect of weather on the en-
etc. EnergyPlus is a stand-alone simulation program without a ergy performance of buildings.
‘user friendly’ graphical interface. It reads input and writes out- The variation in annual average global horizontal solar radiation
put as text files. Since the first release in April 2001, EnergyPlus for the 17 cities from 1980 to 2009 is listed in Table 4. In general,
has been evolving with new and enhanced modeling features the highest and lowest levels of annual average global horizontal
and improved usability. EnergyPlus has been validated through solar radiation occur in the hotter and colder climates respectively.
three types of tests, including analytical tests, comparative tests For example, Riyadh has the highest value of 6588 Wh/m2 in 2001,
and empirical tests. while Fairbanks has the lowest value of 2473 Wh/m2 in 1995. Ta-
The EnergyPlus weather file, an epw file, contains 29 weather ble 4 also shows the maximum variations, defined as the maxi-
variables at 1-h intervals (but can be sub-hourly), among which mum of the annual difference between the highest and the
nine important variables were used in the simulations. These key lowest values of all cities across the 30-year period. Among the
variables can be sorted into four groups: (1) outdoor air condi- 17 cities, Chicago has the largest variation of 652 Wh/m2, while
tions: dry-bulb temperature, dew-point temperature, relative Boise has the smallest variation of 360 Wh/m2. The values listed
humidity, and atmospheric pressure; (2) solar radiation: direct in the fifth and sixth columns represent the average global hori-
normal solar radiation and diffuse horizontal solar radiation; zontal solar radiation over the 30 years for the AMY data and
(3) sky radiation: horizontal infrared radiation; and (4) wind TMY3 data respectively. The values listed in the last two columns
conditions: wind direction and wind speed. Another important are the absolute and relative differences between the TMY3 values
weather variable contained in the epw weather file and used and the average values. The largest difference between TMY3 and
by EnergyPlus is the monthly ground temperature at various soil the average AMY is 809 Wh/m2 which occurs in Miami, a hot cli-
depth levels. EnergyPlus is usually run with a time step of 10 or mate. However, compared with the cities in hotter and colder cli-
15 min, and the hourly weather variables are interpolated to the mates, cities in mixed climates tend to have greater differences.
half-hour intervals. There is a noted trend that the AMY data have higher global

Fig. 2. Variations of climate zone based on annual HDD and CDD for 17 cities using AMY weather data from year 1980 to 2009.
 T. Hong et al. / Applied Energy 111 (2013) 333–350 339

Table 4
Statistics of the annual average global horizontal solar radiation of the 17 cities from year 1980 to 2009.

City Annual average global Average TMY3 Variation Variation Variation%

horizontal solar radiation AMYs (Highest  Lowest) (TMY3  Average) (TMY3  Average)/
(Wh/m2), year Average
Highest Medium Lowest
Miami 5825, 5614, 5444, 5612 4803 380 809 14.4
1986 1991 1992
Riyadh 6588, 6329, 5977, 6318 6114 611 204 3.2
2001 1983 1982
Houston 5099, 4765, 4474, 4750 4459 624 291 6.1
1999 1992 1982
Phoenix 6069, 5868, 5531, 5832 5738 538 94 1.6
2002 1980 1983
Memphis 4799, 4560, 4230, 4564 4493 570 71 1.6
1999 1996 1991
EI Paso 6072, 5749, 5519, 5758 5657 554 101 1.8
2003 1997 1986
San 5520, 5352, 4952, 5322 4703 568 619 11.6
Francisco 1988 1981 1998
Baltimore 4435, 4240, 3918, 4223 4078 517 145 3.4
2006 1987 2003
Albuquerque 6054, 5906, 5635, 5881 5426 419 455 7.7
2003 2009 1983
Salem 4110, 3865, 3692, 3881 3701 418 180 4.6
1987 1995 1998
Chicago 4484, 4099, 3832, 4100 3854 652 246 6.0
1988 2008 1993
Boise 5106, 4927, 4746, 4926 4429 360 497 10.1
2002 1990 1982
Vancouver 3946, 3682, 3361, 3674 3369 585 305 8.3
1985 1998 2007
Burlington 3863, 3715, 3465, 3699 3675 398 24 0.6
1995 2002 2000
Helena 4598, 4367, 4082, 4377 3997 517 380 8.7
2001 1981 1998
Duluth 4069, 3746, 3525, 3744 3678 544 66 1.8
1988 1999 1993
Fairbanks 3018, 2860, 2473, 2868 2591 545 277 9.7
1987 1998 1995

Table 5
Statistics of the annual average dry-bulb temperature of the 17 cities from year 1980 to 2009.

City Annual average dry-bulb temperature (°C), year Average AMYs TMY3 Variation (Highest  Lowest) Variation (TMY3  Average)
Highest Medium Lowest
Miami 25.3, 1998 24.8, 2003 23.8, 1984 24.7 24.5 1.5 0.2
Riyadh 27.8, 1999 26.7, 1981 25.0, 1992 26.6 26.2 2.8 0.4
Houston 21.4, 2006 20.6, 2003 19.0, 1983 20.5 20.4 2.4 0.1
Phoenix 24.8, 1989 23.9, 2000 22.5, 1998 23.9 23.8 2.3 0.1
Memphis 18.5, 2007 17.1, 2002 16.2, 1997 17.2 17 2.3 0.2
EI Paso 19.8, 1994 18.5, 2008 16.7, 1987 18.3 18 3.1 0.3
San Francisco 14.7, 1997 13.8, 2009 12.8, 1982 13.8 13.8 1.9 0
Baltimore 14.4, 1990 13.2, 2005 12.2, 2003 13.2 13.2 2.2 0
Albuquerque 15.0, 2003 14.1, 2008 13.0, 1984 14.0 13.7 2.0 0.3
Salem 12.9, 1992 11.6, 2002 10.0, 1985 11.6 11.7 2.9 0.1
Chicago 12.1, 1998 10.1, 2000 8.7, 1985 10.0 10 3.4 0
Boise 12.7, 2003 11.3, 2005 8.1, 1985 11.1 11.2 4.6 0.1
Vancouver 11.6, 2004 10.5, 2002 9.1, 1985 10.5 9.7 2.5 0.8
Burlington 9.2, 1998 7.8, 2007 7.0, 1980 7.9 7.9 2.2 0
Helena 9.1, 2007 7.4, 2005 4.8, 1996 7.1 7.2 4.3 0.1
Duluth 6.3, 1998 4.3, 2000 2.6, 1996 4.3 4 3.7 0.3
Fairbanks 0.3, 1981 1.8, 2001 4.4, 1999 2.0 1.4 4.7 0.6

horizontal solar radiation than the TMY3 data, which can lead to and Duluth all have variations greater than 3.7 °C. In general, the
the AMYs overestimating the cooling energy use and underesti- differences between the TMY3 values and the average AMY are
mating the heating energy use when compared to the TMY3s. Fur- small, except the TMY3 values have a higher average temperature
ther discussion is provided in Section 3.7. by 0.6 °C for Fairbanks and a lower temperature by 0.8 °C for
Table 5 shows the variations in annual average dry-bulb tem- Vancouver.
perature of the 17 cities from 1980 to 2009. The variations are In summary, the variation in weather data year-over-year is sig-
more significant for cold climates. For example Fairbanks, Helena nificant, especially for cold climates. Such variations should not be
340 T. Hong et al. / Applied Energy 111 (2013) 333–350

Fig. 3. Variations of percentage changes of HVAC source EUI between AMY and TMY3: (a) large office, 90.1-2004 models; (b) large office, 90.1-2010 models; (c) medium
office, 90.1-2004 models; (d) medium office, 90.1-2010 models; (e) small office, 90.1-2004 models; (f) small office, 90.1-2010 models. The red bars represent the variations
across the 30-year while the green bars excluding the six percentage changes from the top three and the bottom three extreme weather years. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

ignored and cannot be represented by single-year weather data – source energy use intensity (EUI, kWh/m2) for the three types of of-
either a historical year or a synthetic year such as TMY. fice buildings with two design efficiency levels in the 17 cities are
shown in Fig. 3. The simulation results from using the TMY3
3.2. Weather impact on HVAC source energy use for individual cities weather data are used as the baseline and are represented as 0%
in these figures. The red bars represent the variation of the per-
HVAC energy use is directly affected by weather, because the centage changes across the 30-year period (1980–2009). The green
cooling and heating loads of buildings are dependent upon weather bars show the same results but excluding the top three largest and
conditions such as outdoor air temperature and humidity, wind the bottom three smallest values to filter out the extreme AMY
speed, and solar radiation. The percentage variation of HVAC cases. The left side bars with negative values indicate TMY3 results
 T. Hong et al. / Applied Energy 111 (2013) 333–350 341

are over-estimating the AMY results while the right side bars with and the TMY3 for the large office at both efficiency design levels in
positive values indicate TMY3 results are under-estimating the Tables 6 and 7, it can be seen that the TMY3 results are usually
AMY results. The cities on the vertical axis of the figures from lower than the AMY results, occurring in 13 out of the 17 cities,
the top to the bottom are arranged by climate zone from the very and by as much as 9–9.2% in Riyadh, 5.6–8.7% in Boise, and 5.2–
hot and humid climate zone 1A to the subarctic climate zone 8. 7.7% in San Francisco. Similar trends can be observed for the med-
In general, the AMY results show large differences when com- ium and small offices.
pared to results using the TMY3 weather data. The TMY3 results As an example, detailed variations of the HVAC source EUI of the
can over-estimate AMY results as much as 18% and under-predict large office in Chicago with low and high building efficiency levels
as much as 37%. Three-dimensional comparisons are made to ana- from 1980 to 2009 are illustrated in Fig. 4. The TMY3 results, the
lyze the relative weather impact by climate zone, building type, average of the AMY results, as well as the average results plus
and building design efficiency. First, it can be seen that most large and minus one and two standard deviations are plotted on the
changes occur in colder climates, regardless of the building type same figures. The TMY3 results are fairly close to those of the aver-
(large-, medium-, or small-size office) or building design efficiency age AMY results, within the range of +2.6% and one standard devi-
level (low, 90.1-2004, or high, 90.1-2010). Usually the largest un- ation. Except for 1992, all AMY results fall within one standard
der-estimates occur in Boise, followed by Helena and then San deviation. The variation, in percentage changes, between the max-
Francisco, while the largest over-estimates occur in Fairbanks, fol- imum and minimum AMY results is large, 22.6% for the 90.1-2004
lowed by Chicago and then Duluth. Secondly, the larger changes office and 28% for the 90.1-2010 office.
occur for the medium-size office building, followed by the large- In summary, the weather impacts on the HVAC source energy
size and then the small-size building. The medium office building use are significant, especially for the medium-size office building
has a larger perimeter area than the large office, and has air-side and for all office buildings in cold climates. The impacts are the
economizers, while the small office does not. Thirdly, the larger least for the small-size office among the three office types. The
changes occur for the large and medium offices with the high-effi- medium-size office buildings have air-side economizers, as re-
ciency design level (90.1-2010) than the low-efficiency design level quired by ASHRAE standard 90.1 in appropriate climates, and more
(90.1-2004). The opposite is true for the small office – the low-effi- window area than the small offices, but have less window area and
ciency design level shows larger changes. Fourthly, the differences more perimeter zone area than the large offices. This makes the
between the red and the green bars for each case are compared. medium offices more sensitive to weather variation than the other
The largest differences occur in Boise regardless of building type two.
and building efficiency design level, followed by Helena, Fairbanks, Weather impacts on buildings are about the same across the
and Miami. In general, the differences in the hotter and colder cli- two efficiency design levels. Meanwhile, large differences between
mates are larger than those in the mixed climates. Finally, compar- the simulated results using TMY3 weather data and the AMY
ing the HVAC source EUI between the average of the 30-year AMYs weather data are observed across the 30-year period. The TMY3

Table 6
Statistics of the HVAC source EUI of the large office, 90.1-2004 during the 30-year period.

City HVAC source EUI (kWh/m2), Average TMY3 Variation Variation Variation%
year AMYs (Highest  Lowest) (TMY3  Average) (TMY3  Average)/
Average
Highest Medium Lowest
Miami 250.8, 229.3, 206.1, 228.9 227.6 44.7 1.4 0.6
1998 1999 1984
Riyadh 217.4, 197.9, 181.7, 200.1 182.1 35.7 18.1 9
1998 1980 1992
Houston 206.7, 193.6, 178.3, 193.8 189.2 28.4 4.7 2.4
1980 1986 1984
Phoenix 205.2, 196.5, 185.3, 195.9 189.8 19.9 6.1 3.1
1984 2008 2004
Memphis 165.3, 151.4, 140.5, 152.6 148.8 24.9 3.8 2.5
1985 1996 1992
EI Paso 108.2, 103.1, 96.5, 102.5 97.8 11.7 4.7 4.6
1981 1982 2004
San 74.9, 65.9, 60.9, 66.9 63.5 13.9 3.4 5.2
Francisco 1997 1998 1999
Baltimore 144.9, 134.9, 125.1, 133.6 136.5 19.9 2.8 2.1
1994 2004 1984
Albuquerque 102, 96.8, 91.5, 96.8 93.1 10.5 3.7 3.8
2007 1981 1986
Salem 83.1, 74.7, 71.4, 75.1 75.1 11.7 0.1 0.1
1990 1988 1981
Chicago 138.2, 128, 112.7, 127.6 130.9 25.5 3.3 2.6
1983 1986 1992
Boise 111.3, 92.8, 83.7, 93.2 87.9 27.6 5.2 5.6
1985 1982 1995
Vancouver 74.9, 67.1, 61.1, 66.8 67.5 13.8 0.6 0.9
1990 1989 1983
Burlington 133.3, 118.9, 108.2, 120.1 118.4 25.1 1.6 1.4
1989 2004 2006
Helena 116.6, 99.5, 88.7, 100.1 95.5 27.9 4.6 4.6
1985 1986 1999
Duluth 146.2, 128.6, 117.9, 130.7 133.1 28.2 2.5 1.9
1989 2005 1992
Fairbanks 180.1, 163.6, 135.8, 161.3 157.7 44.2 3.6 2.2
1999 1997 1981
342 T. Hong et al. / Applied Energy 111 (2013) 333–350

Table 7
Statistics of the HVAC source EUI of the large office, 90.1-2010 during the 30-year period.

City HVAC source EUI (kWh/m2), year Average AMYs TMY3 Variation (Highest  Lowest) Variation (TMY3  Average) Variation%
(TMY3  Average)
Highest Medium Lowest
/Average
Miami 167.2, 1998 151.9, 1991 136.2, 1984 151.2 151.4 31.1 0.2 0.2
Riyadh 164.1, 1998 151.2, 1980 137.4, 1992 152 138 26.7 14 9.2
Houston 122.2, 1998 114.2, 2003 106.1, 1984 114.4 111.3 16.1 3.1 2.7
Phoenix 129, 1981 122.9, 2009 116.3, 1982 122.9 119.6 12.7 3.3 2.7
Memphis 99.5, 1985 92, 1991 85.1, 1992 92.5 90 14.4 2.5 2.7
EI Paso 79.2, 1981 75.8, 1990 71.3, 2004 75.5 71.7 7.9 3.8 5
San Francisco 45.5, 1997 38.2, 2008 35.3, 1999 39 36 11.2 2.9 7.7
Baltimore 84.1, 1994 76, 1985 69.7, 1984 76.1 77.3 14.3 1.2 1.6
Albuquerque 74.9, 2007 71.1, 1981 66.9, 1986 71.6 66.9 7.9 4.6 6.4
Salem 53.7, 1990 48.4, 2002 45.3, 1980 48.7 48 8.3 0.7 1.5
Chicago 88.7, 1985 80.8, 1986 69.2, 1992 80.9 83 19.5 2.1 2.6
Boise 75.4, 1985 62.4, 1991 55.3, 1981 62.5 57.1 20.1 5.4 8.7
Vancouver 47.8, 1998 42.1, 2008 38.2, 2001 42.3 39.9 9.6 2.4 5.7
Burlington 85.5, 1989 74.9, 1983 66.9, 2006 75.7 74.5 18.6 1.3 1.7
Helena 78.5, 1985 63.7, 1980 55.8, 1999 64.8 60.1 22.7 4.4 7.1
Duluth 93.9, 1989 79.6, 2005 70.6, 1992 81.7 83.6 23.3 1.8 2.2
Fairbanks 134.2, 1999 116.3, 1988 91.8, 1981 115.9 111.7 42.4 4.2 3.7

Fig. 4. Variations of HVAC source energy of the large office buildings in Chicago from year 1980 to 2009.

results are lower than the AMYs mainly due to the AMYs having Similar but slightly different patterns are observed for the
higher solar irradiance. Further discussion is provided in building total source EUI. In general, the AMY results show notice-
Section 3.7. able differences from those from the TMY3. The TMY3 results over-
estimate the AMY results by as much as 7.8% and under-estimate
3.3. Weather impact on the building total source energy use for by as much as 9.7%. First, it can be seen that most large changes oc-
individual cities cur in colder climates, regardless of the building type or building
efficiency design level. Usually the largest under-estimates occur
Similar results as shown in Fig. 3 are shown in Fig. 5, but for the in four climates: Riyadh, Boise, Helena and Fairbanks, while the
building total source energy use intensity (EUI, kWh/m2). The var- largest over-estimates occur in four climates: Miami, Chicago, Du-
iation of the building total source EUI are about one-third of those luth and Fairbanks. Secondly, the larger changes occur for the med-
of the HVAC source EUI, because weather changes only affect the ium-size office, followed by the large-size and then the small-size.
HVAC source energy use. The percentage changes of the building Thirdly, the slightly larger changes occur for the large and medium
total source energy, although much smaller, represent a significant offices with the high efficiency design level than the low efficiency
amount of the absolute differences in the building total source en- design level. The opposite is true for the small office – the low
ergy use. efficiency design level shows larger changes. Fourthly, the
 T. Hong et al. / Applied Energy 111 (2013) 333–350 343

Fig. 5. Variations of percentage changes of total building source EUI: (a) large office, 90.1-2004 models; (b) large office, 90.1-2010 models; (c) medium office, 90.1-2004
models; (d) medium office, 90.1-2010 models; (e) small office, 90.1-2004 models; (f) small office, 90.1-2010 models. The red bars represent the variations across the 30-year
while the green bars excluding the six percentage changes from the top three and the bottom three extreme weather years. (For interpretation of the references to color in
this figure legend, the reader is referred to the web version of this article.)

differences between the red and the green bars for each case are 3.4. Weather impact on the HVAC and building total source energy use
compared. The largest differences occur in five climates: Miami, aggregated for the US office building stock
Chicago, Boise, Helena, and Fairbanks. This implies that these cli-
mates tend to have more severe weather impacts. Finally, compar- To analyze the variation in the HVAC and building total source
ing the building total source energy use between the TMY3 energy for all office buildings in the US, the source energy use are
weather data and the average of the 30-year AMY weather data, aggregated across the 15 US cities using weighting factors based on
for the large office at both efficiency design levels in Tables 8 the volume of new construction in each of the 15 cities [22]. The
and 9, it can be seen that the TMY3 results are usually lower than percentage changes at the national level are then calculated and
the AMY results, occurring in 13 out of the 17 cities; but except for shown in Fig. 6.
Riyadh, the under-estimates are less than 2% for all other climates. From Fig. 6, the simulated HVAC source energy use using the
TMY3 data can over-estimate and under-estimate the AMY results
344 T. Hong et al. / Applied Energy 111 (2013) 333–350

Table 8
Statistics of the total building total source EUI of the large office, 90.1-2004 during the 30-year period.

City Total building source EUI Average TMY3 Variation Variation Variation%
(kWh/m2), year AMYs (Highest  Lowest) (TMY3  Average) (TMY3  Average)/
Average
Highest Medium Lowest
Miami 533, 511.5, 488.3, 511.2 509.9 44.7 1.4 0.3
1998 1999 1984
Riyadh 499.5, 479.9, 463.8, 482.2 464.1 35.7 18.1 3.7
1998 1980 1992
Houston 489.3, 476.2, 460.9, 476.4 471.7 28.4 4.7 1
1980 1986 1984
Phoenix 487.6, 478.9, 467.7, 478.3 472.1 19.9 6.1 1.3
1984 2008 2004
Memphis 448.2, 434.3, 423.3, 435.5 431.7 24.9 3.8 0.9
1985 1996 1992
EI Paso 391, 385.9, 379.3, 385.3 380.6 11.7 4.7 1.2
1981 1982 2004
San 358.1, 349.2, 344.2, 350.2 346.7 13.9 3.4 1
Francisco 1997 1998 1999
Baltimore 428.2, 418.1, 408.3, 416.9 419.7 19.9 2.8 0.7
1994 1981 1984
Albuquerque 385.2, 380, 374.7, 379.9 376.3 10.5 3.7 1
2007 1981 1986
Salem 366.5, 358.2, 354.8, 358.5 358.4 11.7 0.1 0
1990 1988 1981
Chicago 421.6, 411.5, 396.2, 411.1 414.3 25.5 3.3 0.8
1983 1986 1992
Boise 394.7, 376.3, 367.1, 376.6 371.4 27.6 5.2 1.4
1985 1982 1995
Vancouver 358.4, 350.5, 344.6, 350.3 350.9 13.8 0.6 0.2
1990 1989 1983
Burlington 417, 402.6, 391.9, 403.7 402.1 25.1 1.6 0.4
1989 2004 2006
Helena 400.4, 383.2, 372.5, 383.8 379.2 27.9 4.6 1.2
1985 1986 1999
Duluth 430.2, 412.7, 402, 414.7 417.1 28.2 2.5 0.6
1989 2005 1992
Fairbanks 464.3, 447.9, 420.2, 445.7 442.1 44.2 3.6 0.8
1999 1997 1981

by 4.8% and 6.1% respectively for the large office, by 4.7% and 7.6% tions for the medium office, as shown in Fig. 7c and d, are much
for the medium office, and by 2.5% and 4.8% for the small office. larger than those for the large and small offices.
The corresponding percentage changes for the building total source Additionally, the percentage changes for the small office are
energy use are 1.4% and 1.7%, 1.7% and 2.7%, and 0.8% and 1.7%. In mostly within ±6% except for a few cases as shown in Fig. 7e and f.
general, the weather impacts are about the same for buildings with For a particular city, if only one green bar can be seen, it is because
the two efficiency design levels, with slightly larger impacts for the the red bar is almost the same as the green bar but overlapped by
low-efficiency buildings. The largest impacts are for the medium- the red bar, and thus cannot be seen. This implies that for the small
size office followed by the large and then the small office. office building in this city, the peak demand is not so sensitive to ex-
Compared with the variations shown in Figs. 3 and 5, the vari- treme weather conditions (the top three and bottom three years).
ations in Fig. 6 are much smaller. This implies the weather impacts On the other hand, if only one red bar can be seen, it is because the
across different climates are not uniform and tend to cancel out green bar is too small to be seen. This implies that the peak demand
each other. For example, during a particular year, the TMY3 results is sensitive to extreme weather conditions. When the top three and
may over-estimate the AMY results for some climates but under- the bottom three years are eliminated, peak demands from the
estimate for others, so the overall TMY3 results at the national le- remaining 24-year AMY data and the TMY3 data are very close or
vel are not so different from the AMY results. However, this should equal, thus the differences cannot be seen.
not overshadow the large discrepancies between the TMY3 results As an example, detailed variations of the simulated peak de-
and the AMY results for individual climates, because energy effi- mand of the large office in Chicago with low and high efficiency
ciency technologies are evaluated and applied locally, and energy levels from 1980 to 2009 are illustrated in Fig. 8. The TMY3 results,
policy is made by local jurisdictions. the average of the AMY results, as well as the average results plus
and minus one and two standard deviations are plotted on the
same figures. The TMY3 result is higher than the average AMY re-
3.5. Weather impact on the peak electricity demand of buildings sult by 1.1% (within one standard deviation) for the 90.1-2004 of-
fice, but lower by 6% (outside two standard deviations) for the
The variations of the percentage changes of the building peak 90.1-2010 office. For the 90.1-2004 office, the variation of peak de-
electricity demand are displayed in Fig. 7. The peak demands of mand is relatively small except for 1991, 2004, and 2008 which has
the medium office using the TMY3 weather data can under-esti- lower peak demand by as much as 7.7% compared to the average
mate that from the AMY data by up to 32.4%, and over-estimate value. For the 90.1-2010 office, the variation of peak demand from
by up to 21%. Unlike the variation in the HVAC source energy use individual AMY results is more significant, up to 13.4% between the
mentioned above, there is no clear correlation between the change minimum and maximum values.
in peak demand and the climate/city. Except for the medium office, In summary, the weather impact on the peak electricity
the mixed climates show larger percentage differences. The varia- demand is significant, even greater than the impact on building
 T. Hong et al. / Applied Energy 111 (2013) 333–350 345

Table 9
Statistics of the total building source EUI of the large office, 90.1-2010 during the 30-year period.

City Total building source EUI Average TMY3 Variation Variation Variation%
(kWh/m2), year AMYs (Highest  Lowest) (TMY3  Average) (TMY3  Average)/
Average
Highest Medium Lowest
Miami 401.1, 385.8, 369.9, 385 384.9 31.1 0.1 0
1998 1991 1984
Riyadh 397.9, 385, 371.2, 385.8 371.6 26.6 14.2 3.7
1998 1980 1992
Houston 357.6, 349.6, 341.2, 349.5 345.8 16.4 3.7 1.1
1998 2003 1984
Phoenix 363.3, 357.3, 350.6, 357.2 354 12.7 3.2 0.9
1981 2009 1982
Memphis 333.9, 326.4, 319.4, 326.9 323.9 14.6 3.1 0.9
1985 1983 1992
EI Paso 312.7, 309.3, 304.7, 308.9 305.2 8 3.7 1.2
1981 2005 2004
San 279.5, 272.6, 268.5, 273.2 270.1 11 3.1 1.1
Francisco 1997 1995 1999
Baltimore 318.3, 310.1, 303.8, 310.2 310.9 14.5 0.7 0.2
1994 1985 1984
Albuquerque 308.1, 304.2, 300.1, 304.7 300.3 7.9 4.3 1.4
2007 1981 1986
Salem 288.6, 293.2, 280, 283.5 282.2 8.5 1.3 0.5
1990 1988 1980
Chicago 323.1, 315.2, 303.5, 315.3 317.1 19.6 1.8 0.6
1985 1986 1992
Boise 309.1, 296.2, 289.2, 296.3 291.1 19.9 5.3 1.8
1985 1991 1981
Vancouver 282.9, 277.3, 273.4, 277.4 275.3 9.5 2.1 0.8
1998 1995 2001
Burlington 320.3, 309.8, 302, 310.6 308.9 18.3 1.6 0.5
1989 1983 2006
Helena 312.7, 298.1, 290.1, 299 294.5 22.6 4.5 1.5
1985 1980 1999
Duluth 328.8, 314.8, 305.7, 316.9 318.3 23.1 1.5 0.5
1989 2005 1992
Fairbanks 371.6, 353.8, 329.4, 353.5 349.3 42.2 4.2 1.2
1999 1988 1981

Fig. 6. Variations of percentage changes of HVAC and total source EUIs of the three types of office buildings with low (90.1-2004 standard) and high (90.1-2010 standard)
energy efficiency levels: (a) changes in HVAC source EUI and (b) changes in total source EUI.
346 T. Hong et al. / Applied Energy 111 (2013) 333–350

Fig. 7. Variations of percentage changes of peak electricity demand: (a) large office, 90.1-2004 models; (b) large office, 90.1-2010 models; (c) medium office, 90.1-2004
models; (d) medium office, 90.1-2010 models; (e) small office, 90.1-2004 models; (f) small office, 90.1-2010 models. The red bars represent the variations across the 30-year
while the green bars excluding the six percentage changes from the top three and the bottom three extreme weather years. (For interpretation of the references to color in
this figure legend, the reader is referred to the web version of this article.)

energy use. The simulated peak demands from TMY3 can signifi- peak demand and source energy use of the building with the high
cantly under- or over-estimate those from the AMY. It is necessary energy efficiency level, to those of the same building with the low
to run simulations using multi-decade of AMY weather data to as- energy efficiency level, using the TMY3 and the 30-year AMY
sess accurately demand response strategies. weather data for the three building types across the 17 climates.
The results are shown in Fig. 9, where the green1 bars represent
3.6. Weather impact on peak electricity demand reduction and energy the variation in the demand reduction and source energy savings,
savings of energy conservation measures using the 30-year AMY weather data. The red marks represent the

The peak demand reduction (in%) and the HVAC and building 1
For interpretation of color in Fig. 9, the reader is referred to the web version of
total source energy savings (in%) are calculated by comparing the this article.
 T. Hong et al. / Applied Energy 111 (2013) 333–350 347

Fig. 8. Variations of peak electricity demand of the large office buildings in Chicago from year 1980 to 2009.

corresponding results using the TMY3 weather data. A few key It should be noted that the calculated peak demand reduction
points can be seen from the results in Fig. 9: and source energy savings come from a combination of energy effi-
ciency improvements from ASHRAE standard 90.1-2004 to 90.1-
 Weather impact on peak demand reduction and HVAC source 2010. Whether similar trends apply to an individual energy effi-
energy savings are large. There are no consistent patterns across ciency improvement, such as better wall or roof insulation, better
the building type or climate. windows, high efficiency lighting systems, or high efficiency HVAC
 Generally the weather impact on the peak demand reduction is systems, is an open question worth further studies.
much greater than on the HVAC source energy savings.
 For HVAC source energy savings, larger weather impacts occur
for the mixed to cold climates, from San Francisco to Fairbanks. 3.7. Discrepancies of weather data from different sources and different
The savings based on TMY3 weather files are usually within the time periods
ranges of savings based on the AMY weather files, except for
over-estimates in San Francisco, Albuquerque, Boise, Vancou- Radhi [30] studied the impact of weather data from two dif-
ver, and Helena, where the red marks are usually at the very ferent periods, 1961–1990 and 1961–2005, on the simulated
right end or outside of the green bars. electricity use of a low-rise and a high-rise commercial building
 The peak demand reduction can vary significantly in Bahrain. Significant variations in simulated energy use from
year-over-year for most climates. The differences in demand the two different weather periods were found and weather data
reduction can be as high as 15% for Chicago and Fairbanks covering more recent periods were recommended to be used for
across the 30-year period for the large office, as shown in better prediction of actual energy use in buildings. Bhandari and
Fig. 9a. Shrestha [19] studied the quality of weather data from two dif-
 Generally the peak demand reductions based on the TMY3 data ferent sources by comparing them to actual measured weather
are within the ranges of reductions based on the AMY data, but data, and the associated impact on building cooling and heating
a few cases show the TMY3 results (the red marks) are at the loads and energy consumption for a single year at a specific US
high or low end of, or even outside the AMY results (the green location.
bars). Furthermore, some climates even demonstrate opposing The AMY weather data from Weather Analytics and the TMY3
weather impacts. For example, in Phoenix, the TMY3 demand from NREL were used in the current study, although they are from
reduction is greater than that from the AMY data for the large different sources and cover slightly different time periods. The
office, but less for the small office. El Paso shows the totally AMYs cover 1980–2009, about 4 years ahead of the TMY3s which
opposite situation as Phoenix. cover 1976–2005. Two constraints determined the choice of the
 To assess accurately the peak demand reduction and energy AMYs and TMY3s: (1) both data sources are reliable and available
savings of ECMs, it is necessary and important to run simula- to the public [3,7]; (2) Weather Analytics does not provide TMY3
tions using multi-decade AMY weather data in comparative (based on same selection criteria as the NREL TMY3) weather files
studies of energy conservation measures. Results from TMY3 created from their 30-year AMYs, and the AMYs used to
data can sometimes significantly over- or under-estimate the create NREL TMY3 weather files, although available to the public
actual energy and cost savings. at [Link]
348 T. Hong et al. / Applied Energy 111 (2013) 333–350

Fig. 9. Variations of percentage reduction of peak electricity demand, and percentage savings of HVAC source energy and total source energy of the 90.1-2010 models over
the 90.1-2004 models: (a–c) large office; (d–f) medium office; (g–i) small office.

NCDCStationData/, are not in EnergyPlus weather data epw format ative humidity and cloud cover, is developed to study the relation-
and thus need data mapping and conversion. ships between meteorological variables and electricity demand as
The temperature data from both sources tend to be more con- well as to predict the monthly electricity demand up to 1 month
sistent than the solar radiation data, as seen from Table 4 which ahead. The model demonstrated an accuracy of better than 1% over
shows that the TMY3s have lower global horizontal solar radiation the data covering the period 1994–2009. Lam et al. [35] used prin-
than the average of the AMYs across all the 17 climates. Although cipal component analysis to study prevailing weather conditions in
both sources used similar algorithms, either the original or the en- subtropical Hong Kong. Regression models were developed to cor-
hanced Perez model [31–33] to calculate solar radiation, Weather relate the simulated monthly building cooling loads and total en-
Analytics data sets lack high quality aerosol data which can lead ergy use, for a generic office building, with a developed climatic
to a high bias of modeled solar radiation under certain cloudy/high index Z, which is a function of the dry-bulb temperature, wet-bulb
humidity conditions. This explains that, in Table 4, Miami (a humid temperature and global solar radiation. The regression models
climate) and San Francisco (with frequent morning fog) have the showed an accuracy of 1% for annual and 4% for monthly simulated
greatest deviations in solar radiation between the average AMYs energy use over the period 1979–2008.
and the TMY3. Another source of discrepancies in the solar data In the current study, a regression model is derived to calculate
is the NREL TMY3s do not include data for certain calendar years the HVAC source energy EUI based on the annual cooling degree
due to eruptions of the volcanoes El Chichón and Mount Pinatubo days (CDD10), annual heating degree days (HDD18), and the an-
(1982–1984 and 1992–1994, respectively) that decreased solar nual average daily global horizontal solar radiation (GHSR):
radiation in the US [3]. This explains that, in Tables 4 and 8 out
of the 15 US cities have the lowest solar radiation in those years HVAC Source Energy EUI ¼ c0 þ c1  CDD10 þ c2  HDD18
across the 30-year period. þ c3  GHSR
To quantify what portion of the overestimate of HVAC source
energy by the AMYs in Tables 6 and 7 is attributable to the high where c0 to c3 are regression coefficients.
bias of solar radiation, there is a need to study the correlation be- Table 10 lists the regression results for the large office buildings
tween the key weather variables and the simulated building compliant with ASHRAE Standard 90.1-2004, when the above
performance. Apadula et al. [34] studied the effect of the meteoro- regression was applied to the 30-year AMYs in the four climates,
logical variability on the national monthly electricity demand in Miami, San Francisco, Boise, and Fairbanks. The results show that
Italy. A multiple linear regression model based on calendar and there are more significant discrepancies in solar radiation between
four weather variables, including air temperature, wind speed, rel- the average AMYs and the TMY3s (Table 4). The linear regressions
 T. Hong et al. / Applied Energy 111 (2013) 333–350 349

Table 10 building stock. If more AMY weather data, for example 50–
Regression of HVAC source energy for the 90.1-2004 Large Office during the 30-year 100 years, is available, methods will be developed to define and se-
period.
lect various TMY weather data representing different conditions.
City c0 c1 c2 c3 R2 For example, cool vs. warm years, dry vs. wet years, cloudy vs. sun-
Miami 69.19 0.05367 0.02736 0.0101 0.84 ny years, for various applications including HVAC design, demand
San Francisco 6.57 0.03674 0.01388 0.01795 0.95 response for smart grids, and solar renewable energy systems.
Boise 13.15 0.02675 0.01932 0.02778 0.86
Fairbanks 40.46 0.05493 0.02366 0.00532 0.92
Acknowledgements

are reasonable with R-squared between 0.84 and 0.95. The varia- This work was supported by the US Department of Energy un-
tions of CDD10, HDD18, and GHSR in the AMYs directly contribute der the US–China Clean Energy Research Center on Building Energy
to the variations of the simulated HVAC source energy. AMYs with Efficiency. It is also part of our research activities for the IEA ECBCS
higher CDD10 and HDD18 will lead to higher HVAC source energy Annex 53: Total Energy Use in Buildings – Evaluation and Analysis
use. Except for the cooling dominated climate of Miami, the other Methods. This work was co-sponsored by the Bureau of Energy,
three climates show that higher solar radiation leads to lower Ministry of Economic Affairs, Taiwan, ROC.
HVAC source energy use. The impact of solar radiation on building
performance depends on climate – lower or higher solar radiation
does not necessarily always dominate. References
The regression coefficient c3 represents the sensitivity of the
[1] IEA ECBCS Annex 53: total energy use in buildings – evaluation and analysis
HVAC Source Energy EUI to the annual average daily global horizon- methods.
tal solar radiation, assuming the indirect impact of solar radiation on [2] Ai-Mofeez IA, Numan MY, Alshaibani KA, Ai-Maziad FA. Review of typical vs.
ambient air temperature is considered separately in the sensitivity synthesized energy modeling weather files. J Renew Sustain Energy
2012;4(1):012702.
of CDD10 and HDD18. Based on the regression models, the lower so- [3] Wilcox S, Marion W. Users manual for TMY3 data sets. Technical report. NREL/
lar radiation of the TMY3s in Miami (by 14.4%), San Francisco TP-581-43456; April, 2008.
(11.6%), Boise (10.1%), and Fairbanks (9.7%) would contribute to [4] Hong TZ, Chou SK, Bong TY. A design day for building load and energy
estimation. Build Environ 1999;34(4):469–77.
the underestimate (for Miami) or overestimate (for the other three [5] Hong TZ, Jiang Y. Stochastic weather model for building HVAC systems. Build
climates) of HVAC Source Energy EUI of the TMY3s by 3.6%, 16.6%, Environ 1995;30(4):521–32.
14.8%, and 0.9% respectively. The percentages for San Francisco [6] Jiang Y, Hong TZ. Stochastic analysis of building thermal processed. Build
Environ 1993;28(4):509–18.
and Boise are much higher mainly due to their much lower HVAC
[7] Keller JL, Khuen CA. A new climate & hourly data delivery system providing
source energy EUI compared to those of Miami and Fairbanks. In global support for precision modeling and control. SimBuild 2012.
conclusion, the discrepancy in solar radiation between different [8] Freeman T. Evaluation of the typical meteorological years for solar heating and
cooling system studies. Technical report SERI/TR-8150-1. Solar Energy
weather data sources can have a significant impact on differences
Research Institute; 1979.
in the simulated HVAC source energy. High quality solar radiation [9] Crawley DB. Which weather data should you use for energy simulations of
data is key to improving the accuracy of simulated building commercial buildings? ASHRAE Trans 1998;104(2):498–515.
performance. [10] Barnaby CS, Crawley DB. Weather data for building performance simulation.
Building performance simulation for design and operation. Spon Press; 2011.
It should be noted that the regression model is used to appro- p. 37–55.
priately estimate the effect of the high bias solar data, it is not rec- [11] Chow TT, Chan ALS, Fong KF, Lin Z. Some perceptions on typical weather year-
ommended to replace whole building dynamic simulation for from the observations of Hong Kong and Macau. Sol Energy 2006;80:459–67.
[12] Crawley DB. Estimating the impacts of climate change and urbanization on
calculating the HVAC source energy. building performance. J Build Perform Simul 2008;1(2):91–115.
[13] Long N. Real-time weather data access guide. User’s guide, NREL/BR-550-
4. Conclusions 34303; March, 2006.
[14] Xu P, Huang YJ, Miller N, Schlegel N. Effects of global climate change on
building energy consumption and its implications on building energy codes
Nowadays with the availability of long-term AMY weather data and policy in California. California Energy Commission, CEC-2008-006;
and sufficient computational power of personal computers, it is September, 2008.
[15] Lam JC. Climatic influences on the energy performance of air-conditioned
feasible and necessary to run simulations with AMY weather data
buildings. Energy Convers Manage 1999;40:39–49.
covering multiple decades to fully assess the impact of weather on [16] Yang L, Lam JC, Liu J, Tsang CL. Building energy simulation using multi-years
the long-term performance of buildings, and to evaluate the energy and typical meteorological years in different climates. Energy Convers Manage
savings potential of energy conservation measures for new and 2008;49:1123–34.
[17] Wan KW, Li HW, Liu D, Lam JC. Future trends of building heating and cooling
existing buildings from a life cycle perspective. Main findings from loads and energy consumption in different climates. Build Environ
this study are: (1) annual weather variation has a greater impact 2011;46:223–34.
on the peak electricity demand than on the energy use in build- [18] Wan KW, Li HW, Pan W, Lam JC. Impact of climate change on building energy
use in different climate zones and mitigation and adaptation implications.
ings; (2) simulated building energy use using the TMY3 weather Appl Energy 2012;97:274–82.
data is not necessarily representative of the average energy use [19] Bhandari M, Shrestha S, New J. Evaluation of weather datasets for building
using the AMY data, across the 30-year period. The TMY3 results energy simulation. Energy Build 2012;49:109–18.
[20] US Energy Information Administration. 2003 Commercial buildings energy
can be significantly higher or lower than those from the AMY data; consumption survey; 2012. <[Link]
(3) the weather impact is greater for buildings in cold climates; (4) [21] ANSI/ASHRAE/IES. ASHRAE standard 90.1-2010: energy standard for buildings
the weather has the greatest impact on the medium-size office except low-rise residential buildings.
[22] Thornton BA, Rosenberg MI, Richman EE, Wang W, Xie Y, Zhang J, et al.
building, followed by the large office and then the small office; Achieving the 30% goal: energy and cost savings analysis of ASHRAE standard
and (5) simulated energy savings and peak demand reduction by 90.1-2010. PNNL-20405; May, 2011.
energy conservation measures using the TMY3 weather data can [23] NOAA Satellite and Information Service, National Climatic Data Center; 2012.
<[Link]
be significantly lower or higher when compared to the results
[24] Remund J, Muller SC, Schilter C, Rihm B. The use of Meteonorm weather
using the AMY data. These findings can support energy policy mak- generator for climate change studies. In: EMS annual meeting: EMS2010-417;
ing, energy code development, building technologies evaluation, 2010. p. 7.
and utility incentive programs planning. [25] U.S. Department of Energy. Commercial prototype building models; 2012.
<[Link]
Future work will continue to investigate the weather impact for [26] USDOE. Commercial reference buildings; 2012. <[Link]
other building types, and aggregate the impact across the entire US [Link]/buildings/commercial/ref_buildings.html>.
350 T. Hong et al. / Applied Energy 111 (2013) 333–350

[27] US Department of Energy. EnergyPlus; 2012. <[Link] [32] Perez R, Ineichen P, Seals R, Michalsky J, Stewart R. Modeling daylight
buildings/energyplus/>. availability and irradiance components from direct and global Irradiance. Sol
[28] Hittle D. BLAST – the building loads analysis and system thermodynamics Energy 1990;44:271–89.
program. Technical report. Construction Engineering Research Laboratory, [33] Perez R, Ineichen P, Maxwell EL, Seals R, Zelenka A. Dynamic global to direct
University of Illinois; 1977. irradiance conversion models. ASHRAE Trans 1992;98:354–69.
[29] LBNL, DOE-2; 2012. <[Link] [34] Apadula F, Bassini A, Elli A, Scapin S. Relationships between meteorological
[30] Radhi H. A comparison of the accuracy of building energy analysis in Bahrain variables and monthly electricity demand. Appl Energy 2012;98:346–56.
using data from different weather periods. Renew Energy 2009;34:869–75. [35] Lam JC, Wan K, Lam T, Wong SL. An analysis of future building energy use in
[31] Perez R, Stewart R, Arbogast C, Seals R, Scott J. An anisotropic hourly diffuse subtropical Hong Kong. Energy 2010;35:1482–90.
radiation model for sloping surfaces: description, performance validation, site
dependency evaluation. Sol Energy 1986;36:481–97.
 Applied Energy 111 (2013) 515–528

Contents lists available at SciVerse ScienceDirect

Applied Energy
journal homepage: [Link]/locate/apenergy

On variations of space-heating energy use in office buildings

Hung-Wen Lin a, Tianzhen Hong b,⇑
a
Intelligent Energy-Saving System Division, Green Energy and Environment Research Laboratories, Industrial Technology Research Institute, Hsinchu 31040, Taiwan, ROC
b
Environmental Energy Technologies Division, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA

h i g h l i g h t s

 Space heating is the largest energy end use in the U.S. building sector.
 A key design and operational parameters have the most influence on space heating.
 Simulated results were benchmarked against actual results to analyze discrepancies.
 Yearly weather changes have significant impact on space heating energy use.
 Findings enable stakeholders to make better decisions on energy efficiency.

a r t i c l e i n f o a b s t r a c t

Article history: Space heating is the largest energy end use, consuming more than seven quintillion joules of site energy
Received 21 November 2012 annually in the U.S. building sector. A few recent studies showed discrepancies in simulated space-heat-
Received in revised form 4 March 2013 ing energy use among different building energy modeling programs, and the simulated results are sus-
Accepted 15 May 2013
pected to be underpredicting reality. While various uncertainties are associated with building
simulations, especially when simulations are performed by different modelers using different simulation
programs for buildings with different configurations, it is crucial to identify and evaluate key driving fac-
Keywords:
tors to space-heating energy use in order to support the design and operation of low-energy buildings. In
Building simulation
Design and operation
this study, 10 design and operation parameters for space-heating systems of two prototypical office
EnergyPlus buildings in each of three U.S. heating climates are identified and evaluated, using building simulations
Office buildings with EnergyPlus, to determine the most influential parameters and their impacts on variations of space-
Performance benchmarking heating energy use. The influence of annual weather change on space-heating energy is also investigated
Space heating using 30-year actual weather data. The simulated space-heating energy use is further benchmarked
against those from similar actual office buildings in two U.S. commercial-building databases to better
understand the discrepancies between simulated and actual energy use. In summary, variations of both
the simulated and actual space-heating energy use of office buildings in all three heating climates can be
very large. However these variations are mostly driven by a few influential parameters related to building
design and operation. The findings provide insights for building designers, owners, operators, and energy
policy makers to make better decisions on energy-efficiency technologies to reduce space-heating energy
use for both new and existing buildings.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction suming the most energy in the commercial building sector. In of-
fice buildings, space heating consumes about one-third of total
According to the 2010 United States Department of Energy (US- site energy, according to the CBECS. It is therefore crucial to study
DOE) Building Energy Databook [1], space heating is the largest end the space-heating energy use of such buildings in order to reduce
use in the U.S. building sector. Space heating consumes about 5.2 their energy use and carbon emissions.
and 2.3 quintillion joules of annual site energy for residential and The growth in energy use allocated to the commercial buildings
commercial buildings, respectively. The U.S. Energy Information sector averaged 2.8% annually from 1950 to 2006 [3]. In the past
Administration (EIA) 2003 Commercial Buildings Energy Consump- decade, energy-saving technology improvements in office build-
tion Survey (CBECS) [2] indicates that office buildings are the most ings have received a lot of attention [3–9]. Andrew and Krogmann
common building type, comprising the largest floor area and con- [8] investigated issues affecting the adoption of energy-efficient
heating technology in U.S. office buildings. The factors he studied
⇑ Corresponding author. Tel.: +1 510 4867082; fax: +1 510 4864089. included energy price, building location, floor area, rental, building
E-mail address: thong@[Link] (T. Hong).
vintage, window area, and office equipment. In his study, the mul-

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
[Link]
516 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

tinomial logistic analysis of these factors employed spreadsheet uncertainties are associated with building simulations, especially
manipulations and statistical calculations. Liu et al. [9] describe a when simulations are performed by different modelers using dif-
mathematical modeling framework for energy systems to improve ferent simulation programs for buildings with different configura-
energy efficiency and environmental performance of commercial tions, it is crucial to identify and evaluate key driving factors to
buildings, with the goal of achieving optimal energy designs. How- space-heating energy use to support the design and operation of
ever, a systematic integration approach for truly achieving optimal low-energy buildings. These key driving factors can be categorized
energy-systems design in commercial buildings is still lacking. Re- into six groups: climate conditions, building envelope, space-heat-
cently, more new building designs aim to green buildings or zero ing systems, building operation and maintenance, occupant behav-
net energy buildings, emphasizing the importance of energy-effi- ior, and indoor environmental conditions.
ciency technologies and system designs, building operation and The New Buildings Institute recently published a simulation
maintenance, and occupant behavior. Good operational practice study on total site energy use in midsize office buildings [20] to
and high building design efficiency could lower the energy use of look at key driving factors of building energy use. Twenty-eight
space heating [10,11]. Santin [12] looked at the relationship be- building characteristics were identified and grouped into design
tween user behavior and space-heating energy consumption, and assets, operation practice, and tenant behaviors. Three systems
concluded that behavior patterns could be used in space-heating and equipment-operation practices with respect to building energy
energy calculations, and usage profiles with different behaviors use were identified by using different performance values for each
could be discerned. characteristic parameter. Simulation results showed the key fac-
Pan et al. [13] simulated effects of external wall insulation tors that affect total site energy use in midsize office buildings in
thickness on annual cooling and heating energy uses of an office 16 U.S. climates. Total site energy is a simple sum of electricity
building in three Chinese climates. It was found that, for heating use and gas use – one unit of electricity is valued the same as
dominant climate like Beijing, more insulation reduced the com- one equal unit of natural gas; no generation or transmission or dis-
bined annual cooling and heating energy uses of perimeter offices tribution loss is considered. As the total energy use of a building in-
facing all four cardinal orientations. More insulation reduces an- cludes all end uses such as lighting, space heating, space cooling,
nual energy uses of offices facing North, East, and West, but not service water heating, and plug-loads, the key driving factors of a
necessarily for the south facing office. For cooling dominant cli- building’s total energy use would be very different from those of
mate like Guangzhou more insulation did not reduce annual en- a specific end use like space heating. The use of source or primary
ergy use at all. Yang et al. [14] surveyed envelope designs of energy would be a better indicator of building energy performance.
existing office buildings in five major Chinese climates, and found The objective of the current study is to identify, understand, and
the overall thermal transfer value of envelope was much higher quantify important building design and operation parameters that
than the current local energy code and almost double the ASHRAE can have significant impacts on space-heating energy use in office
Standard 90.1-2001. More insulation of exterior walls and roofs buildings, with different characteristics located in different heating
was recommended to reduce heating energy use for buildings in climates, by computer simulations with EnergyPlus. The impact of
cold climate. Dovjak et al. [15] studied problem of High Heating en- weather data on space-heating energy use is also investigated by
ergy use in Slovenian buildings with exergy and energy analysis. running simulations with multiple decades of historical weather
Their energy analyses showed that less thermal insulation contrib- data. The simulated results are further benchmarked with the
uted the most to the highest heating energy demand especially in space-heating energy use of comparable office buildings selected
colder climate. The results from exergy analysis drew similar con- from the two well-known U.S. commercial building databases to
clusions – insulation has much bigger effect than effect of boiler investigate discrepancies between simulated and actual heating
efficiency. However, the most effective solution is to improve energy use.
building envelope together with boiler efficiency. Yildiz and Gun- It is not the intent of this paper, although the analysis and sim-
gor [16] presented energy and exergy analyses for the whole pro- ulation method can apply, to analyze the total energy use of build-
cess of space heating in buildings in Turkey climates using ings; therefore, this study’s results and findings should not be
simplified steady state heating load and energy calculations. Three directly applied to the whole-building energy use, which includes
heating systems, liquid natural gas (LNG) fired conventional boiler, other end uses. The heating systems discussed in this article are
LNG condensing boiler, and air-to-air heat pump, were compared stand-alone systems powered by natural-gas hot-water boilers or
from the power plant through the building envelope using exergy electric resistance; they are not part of the district heating systems
analysis. Eskin and Turkmen [17] studied the interactions between that are popular in Northern Europe countries and Northern China
different conditions, control strategies and heating/cooling loads in [21].
office buildings in the four major climatic zones in Turkey using This study is part of a bigger effort to study key driving factors
building energy simulation. Calibrated energy models were used of energy performance of buildings under the International Energy
to examine energy conservation opportunities on annual cooling, Agency (IEA) Energy Conservation in Buildings & Community Sys-
heating and total building load at four major cities. The effect of tems (ECBCS) Annex 53 Total Energy Use in Buildings: Analysis &
the parameters like the climatic conditions, insulation and thermal Evaluation Methods.
mass, aspect ratio, color of external surfaces, shading, window sys- The first section of the paper describes analysis methodology,
tems including window area and glazing system, ventilation rates and the second section provides details of the selected building de-
and different outdoor air control strategies on annual building en- sign and operation parameters, together with definitions of simu-
ergy requirements is examined and the results are presented for lation runs. The third section presents and discusses the results.
each city. The conclusion section summarizes key findings and potential fu-
The lack of knowledge about the factors that determine total ture research.
building energy use is a significant barrier to achieving substantial
building energy efficiency. Recently, a few studies [18,19] using
simulations to calculate building performance showed relatively 2. Analysis methodology
low space-heating energy use compared with rules-of-thumb and
large discrepancies in space-heating energy use between different Building simulations and benchmarking with building energy
simulation programs, which raised concerns of whether simulation consumption databases are the two methods we used to study
can be used to predict space-heating energy use. While various the space-heating energy use in office buildings. Two office build-
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 517

ings with different sizes and design configurations – the high-rise HVAC systems, depend on the building’s location or climate zone.
large office and the single-story small office – are studied. To look The internal loads, including interior lighting power and plug
at the influence of climate, three typical climate zones that require loads, occupant density, and operation schedules, stay the same
significant space heating are studied. Based on design and opera- across all climates.
tion practice, a few key parameters for the large- and small-size of-
fice buildings are identified and their impacts on space-heating
energy use are evaluated by energy simulations. The simulated 2.1.1. The large-size office building
space-heating energy uses are benchmarked with two U.S. com- The large office building has 12 stories and a basement, with a
mercial building databases of measured whole-building energy total floor area of 46,320 m2. The building has a rectangular shape
use. Furthermore, 30 years of actual meteorological weather data, with the long axis along the east–west and an aspect ratio of 1.5.
from 1980 to 2009, are used in the simulations to study the impact Each floor has four perimeter zones and one core zone with about
of weather changes year-over-year on space-heating energy use. 30% and 70% of the total floor area, respectively. The window-wall
As defined in this study, space-heating energy use is the site’s ratio (WWR) is about 40%, excluding the basement wall area. The
energy in the form of natural gas consumed by boilers in the large roofs are flat with insulation above deck. The building has central
office building or furnaces in the small office building; it does not built-up VAV systems with hot-water zone reheat. The VAV boxes
include the electricity use of the hot-water pump for the large of- have reverse acting dampers with a maximum supply air temper-
fice building, or the fans of the air-handling units for both office ature of 35 °C. The reverse acting damper in a VAV box can open
buildings during heating operations. For the large office building, wider to meet zone heating loads, which differentiates it from a
the space-heating energy use includes hot-water energy consumed normal acting damper that stays at a fixed minimum position dur-
by the reheat coils in the zone terminal units (i.e., variable air vol- ing heating operations. The supply air temperature leaving the
ume [VAV] boxes) and the central heating coils located in the air- cooling coils is set to 12.8 °C during cooling mode. There is no heat
handling units. For benchmarking purpose, the space-heating en- recovery between outdoor air and exhaust air. There is no humid-
ergy is also presented in energy use intensity (EUI), defined as an- ifier. The central plant has two water-cooled chillers and a hot-
nual site energy in MJ (mega joules) of space heating per building water gas-fired boiler. Fig. 1a illustrates the 3-D and plan views
total floor area in m2. of the building.

2.1. Characteristics of the large- and small-size office buildings 2.1.2. The small-size office building
The small office building has only one floor with an area of
The large- and small-size office buildings were selected from 511 m2. The building has a rectangular shape with the long axis
the USDOE commercial reference buildings (CRBs) [22], which along the east–west and an aspect ratio of 1.5. Four perimeter
comply with the American Society of Heating, Refrigeration, and zones and the core zone have about 70% and 30% of the total floor
Air-conditioning Engineers (ASHRAE) Standard 90.1-2004 [23]. As area, respectively; the perimeter–core ratio is the opposite of that
Standard 90.1-2004 has different efficiency requirements for build- of the large office building. The WWR is about 20%. The building
ings located in different climate zones, the efficiency levels of both has an attic, as shown in Fig. 1b. Each of the five zones is served
office buildings, including envelope insulation, window types, and by a packaged single-zone system: a constant-volume HVAC sys-

(a) The 3-D and plan views of the large-size office building.

(b) The 3-D and plan views of the small-size office building.
Fig. 1. The large- and small-size office buildings from the USDOE commercial reference buildings. (a) The 3-D and plan views of the large-size office building. (b) The 3-D and
plan views of the small-size office building.
518 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

Table 1 (1) Analytical tests compare EnergyPlus simulation results with

Characteristics of selected cities and climate zones. analytical mathematical solutions for simple buildings:
City ASHRAE climate CBECS census region HDD18 CDD10  HVAC tests, based on ASHRAE Research Project 865.
zone  Building fabric tests, based on ASHRAE Research Project 1052.
Chicago Cool-Humid, 5A West North Central, 6176 3251 (2) Comparative tests compare EnergyPlus simulation results
Midwest with those of other simulation engines such as DOE-2, ESP,
Minneapolis Cold-Humid, 6A East North Central, 7981 2680 and Transient System Simulation Tool (TRNSYS):
Midwest
Fairbanks Subarctic, 8 West Pacific 13,940 1040
 ANSI/ASHRAE Standard 140-2007.
 IEA Solar Heating and Cooling Programme (IEA SHC) BESTest
(Building Energy Simulation Test) methods not yet in Standard
140.
tem with heating from a gas furnace and cooling from a direct-  EnergyPlus HVAC component comparative tests.
expansion (DX) unitary system.  EnergyPlus Global Heat Balance tests.
(3) Empirical tests compare EnergyPlus simulation results with
2.2. Climate zones measurements of actual buildings. Although some applica-
tions compare and calibrate EnergyPlus simulation results
Three climates – Chicago, Minneapolis, and Fairbanks – were with measured energy and performance of buildings, much
selected in this study to represent typical climates that require sig- more needs to be done with this type of test for actual build-
nificant space heating in the United States. Based on the climate ings of various complexities of design and operations.
zones used in the ASHRAE Standard 90.1-2010 [24], Chicago be-
longs to the cool and humid zone 5A, Minneapolis belongs to the
2.5. Building databases
cold and humid zone 6A, and Fairbanks belongs to the subarctic
zone 8. Table 1 lists the climate zone information for the three cit-
The simulation results are benchmarked with two databases of
ies. In the table, HDD18 is the heating degree days with a base tem-
commercial buildings in the United States: the 2003 CBECS and the
perature of 18 °C, and CDD10 is the cooling degree days with a base
USDOE high-performance buildings (HPBs) database [28]. The
temperature of 10 °C.
CBECS is a national survey that collects information on energy con-
sumption and expenditures of U.S. commercial buildings. In this
2.3. Weather data
database, commercial buildings include all those in which at least
half of the floor area is used for a purpose that is not residential or
The typical meteorological year, third generation (TMY3)
industrial; they include building types that might not traditionally
[25,26], weather data of the three cities, available at the EnergyPlus
be considered ‘‘commercial,’’ such as schools, correctional institu-
web site, were used in the simulations. The TMY3s are data sets of
tions, and buildings used for religious worship. The HPB database
hourly values of solar radiation and meteorological elements for a
has more than 100 commercial buildings (mostly in the United
one-year period of 12 representative months compiled from 1976
States) that were built recently and have low energy consumption.
to 2005. They are intended to be used for computer simulations of
The database has detailed building descriptions and either mea-
solar energy conversion systems and building systems to facilitate
sured or simulated energy-consumption data. The space-heating
performance comparisons of different system types, configura-
energy use from the HPB database was mostly calculated from cal-
tions, and locations in the United States. Because they represent
ibrated energy models.
typical rather than extreme conditions, they are not suited for
For the CBECS, building location is grouped into four U.S. Census
designing systems to meet the worst-case conditions occurring at
regions that are subdivided into nine divisions. The four regions are
a location.
the West, Midwest, Northeast, and South regions. Fairbanks be-
Historical weather data, generated from actual weather mea-
longs to the Pacific division of the West region; Chicago to the
surements and observations from 1980 to 2009 for Chicago and
West North Central division of the Midwest region; and Minneap-
Fairbanks, are used in the simulations to study the impact of
olis to the East North Central division of the Midwest region.
weather on space-heating energy use for both office buildings.
A few buildings were selected from the two databases in order
Such weather data were not available for Minneapolis during the
to match the simulated buildings as much as possible according to
study, so similar analysis is not done for Minneapolis.
the criteria: (1) building type (office), (2) building size (large or
small), (3) vintage, and (4) location.
2.4. Simulation engine

The EnergyPlus version 7.2, released in October 2012, was used 3. Building design and operation parameters
for the study’s building simulations. USDOE developed it as a new-
generation building energy simulation program that builds on the Based on office-building design and operation practice, 10
most popular features and capabilities of the Building Loads Anal- parameters with potentially significant impacts on space-heating
ysis and System Thermodynamics (BLAST) and DOE-2. EnergyPlus energy use were selected for the study. The parameters were
has innovative simulation capabilities, including time steps of less sorted into two groups – design and operation – as shown in Ta-
than an hour, and modular systems simulation modules that are ble 2, based on whether a parameter is mostly determined during
integrated with a zone heat balance simulation. It calculates space building design or operation. The classification for design and oper-
temperature, occupant thermal comfort, cooling and heating loads, ation parameters for space-heating energy use is listed in Table 2.
HVAC equipment sizes, energy consumption, utility cost, air emis- The selected parameters include envelope insulation, window area,
sions, water usage, renewable energy, etc. EnergyPlus is a stand- window type, internal loads, infiltration (rate and schedule), space-
alone simulation program without a ‘‘user friendly’’ graphical heating temperature setpoint, heating setback during unoccupied
interface. It reads input and writes output as text files. Since the hours, terminal VAV box minimal damper position, and boiler/fur-
first release in April 2001, EnergyPlus has evolved to provide nace efficiency. For each parameter, the reference value is set in the
new and enhanced modeling features and improved usability. basecase models, which are based on ASHRAE Standard 90.1-2004;
EnergyPlus has been validated through three types of tests [27]: a better and a worse performance value are then determined based
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 519

Table 2 cases, the insulation levels are set according to the pre-1980 offices
Selected design and operation parameters for space heating. from the CRBs. Table 3 lists a few key parameters of the office
Design parameters Operation parameters buildings constructed at three different ages.
Window type Air infiltration rate
Boiler/furnace efficiency Air infiltration schedule 3.1.3. Window area
Internal loads (lighting and plug- Space-heating thermostat setting With more windows, space-heating loads tend to increase for
loads)
Envelope insulation Heating setback control
most climates that require heating because windows usually con-
Window area (window-wall-ratio) VAV box minimum damper position tribute more heat loss than walls, even taking into account win-
setting dows’ solar-heat gains. For the basecases, the large office
building has a WWR of 40%, while the small office building has a
WWR of 20%. The High WWR cases double the window area: The
large office building has a WWR of 68% (cannot reach 80% due to
on building design or operation practice, applicable building-en-
the assumption of no windows on the plenum walls), while the
ergy standards, and available measurement or analysis reports.
small office building has a WWR of 40%. The Low WWR cases re-
The selection and determination of these parameters are from
duce window area by 50% from the basecases: The large office
the perspectives of practical building design and operations. This
building has a WWR of 20%, while the small office building has a
differentiates the current study from sensitivity analyses that do
WWR of 10%.
not use or require a high fidelity of parameters, as the sensitivity
is calculated as a ratio between the change in outputs and the
change in inputs. 3.1.4. Window type
Windows with lower U-factor and higher solar heat gain coeffi-
cient (SHGC) reduce space-heating loads. The U-factor is the heat
3.1. Design parameters
transfer rate through the window per unit area and per unit tem-
perature difference. The SHGC represents the fractional amount
3.1.1. Internal loads
of solar energy that strikes the window and ends up warming
Internal loads include heat gains from interior lighting, plug-
the indoor environment. Visible transmittance (VT) is the fraction
loads, and occupants. Internal loads reduce space-heating loads.
of visible light that comes through the glass. This is influenced
For the basecase, interior lighting power density (LPD) is set to
by glass selection as well as the amount of the opening taken up
10.76 W/m2, based on ASHRAE Standard 90.1-2004. The plug-load
by nontransparent components such as the frame. The basecases
(receptacle) equipment power density (EPD) is set to 10.76 W/m2,
have double-pane windows. The worst cases use single-pane win-
based on the CRBs. For the High Internal Loads case, the LPD and
dows, while the better cases use triple-pane windows. Table 4
EPD are set 50% higher than the basecase, while for the Low Inter-
summarizes window-type performance for relevant cases at differ-
nal Loads case, they are set 50% lower. The LPD of the High Internal
ent climate zones.
Loads case is set to 16.14 W/m2, which is based on the prescriptive
requirement of interior lighting for the whole building in ASHRAE
Standard 90.1-1989 [29]. The 50% lower LPD is based on the state- 3.1.5. Boiler and Furnace Efficiency
of-the-art lighting technologies for office buildings. The 50% lower A higher efficiency of heating equipment reduces space-heating
EPD references Fisher’s study [30], which shows plug-load energy energy use. For the basecases, the large office building has a boiler
use could be reduced over 50% by using energy-efficient appli- of 80% efficiency, while the small office building has furnaces of
ances, installing an energy-management system, and most impor- 78% efficiency. The High Boiler/Furnace Efficiency cases, assuming
tant, educating and training occupants on how to save energy. the use of condensing boilers and furnaces, have a boiler of 91%
efficiency for the large office building and furnaces of 88% effi-
ciency for the small office building.
3.1.2. Envelope insulation
Better insulation of a building envelope reduces space-heating
3.2. Operation parameters
loads. For the basecases, the insulation levels of the wall and roof
constructions are based on ASHRAE Standard 90.1-2004. For the
3.2.1. Air infiltration rate
More Envelope Insulation cases, the insulation levels are based
Air infiltration during heating seasons increases space-heating
on ASHRAE Standard 90.1-2010. For the Less Envelope Insulation
loads. Parameters of air infiltration include peak infiltration rate
and infiltration schedule. According to a report by National Insti-
Table 3 tute of Standards and Technology [31,32], peak infiltration rates
Parameters for three building types. measured for typical commercial buildings range from 2.04 to
Vintage Roof construction U-factor Wall construction U-factor 9.14 L/(s m2), based on 75 Pa of pressure difference and per unit
(W/m2 K) (Large/small (W/m2 K) (Large/small of gross exterior wall area. For EnergyPlus simulations, these infil-
office) office)
Less Insulation IEAD/IEAD* Steel/mass
(USDOE CRBs, 5A, 0.358/0.358 5A, 0.698/0.505 Table 4
Pre-1980) 6A, 0.358/0.358 6A, 0.591/0.477 Window type.
8, 0.273/0.273 8, 0.454/0.363
NFRC rated values U-factor (W/ SHGC VT
Basecase (ASHRAE IEAD/attic Mass/mass
m2 K)
90.1-2004) 5A, 0.358/0.193 5A, 0.698/0.698
6A, 0.358/0.153 6A, 0.591/0.591 Base case: double-pane window, low-e
8, 0.273/0.153 8, 0.454/0.454 Chicago 3.24 0.385 0.305
Minneapolis 3.24 0.385 0.305
More Insulation IEAD/attic Mass/mass
Fairbanks 2.62 0.296 0.212
(ASHRAE 90.1- 5A, 0.273/0.153 5A, 0.511/0.511
Single-pane window, clear 5.81 0.822 0.882
2010) 6A, 0.273/0.153 6A, 0.454/0.454
Triple-pane window, spectral selective, 0.87 0.285 0.451
8, 0.273/0.119 8, 0.403/0.403
clear, low-e
*
IEAD: insulation entirely above deck.
520 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

tration rates are adjusted to the 4 Pa of pressure difference. The

Window type
basecase infiltration rate is 2.04 L/(s m2) (equivalent to 0.65 air

90.1-2004
90.1-2004
90.1-2004
90.1-2004

90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004

90.1-2004
90.1-2004
90.1-2004

90.1-2004
90.1-2004
90.1-2004
90.1-2004

90.1-2004
changes per hour), which is also defined in ASHRAE Standard
90.1-2010. For the High Infiltration Rate case, the 7.61 L/(s m2)

Single

Single
Triple

Triple
(2.44 ach) is used in reference to the proposal to ASHRAE Standard
90.1-2013 for buildings without installation of a continuous air

construction
barrier. The High Infiltration Rate is 375% higher than the basecase.

90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004

90.1-2010
90.1-2010

90.1-2010
90.1-2004

90.1-2004
90.1-2004
90.1-2004
90.1-2004
90.1-2004

90.1-2004
90.1-2004

Pre-1980

Pre-1980
Envelope
The 50% lower infiltration rate is used in the Low Infiltration Rate
case for airtight buildings.
The peak air-infiltration rate depends to a great extent on a

Boiler/furnace
building’s airtightness, especially the opening and closing of win-
dows and doors, which are more related to building operation

efficiency

91%/88%

91%/88%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%
78%/80%

78%/80%

78%/80%
78%/80%
78%/80%
78%/80%

78%/80%
78%/80%
78%/80%
and occupant activity. Therefore, air-infiltration rate is categorized
as an operation rather than a design parameter.

Heating setpoint/
3.2.2. Air infiltration schedule
A schedule is used to describe the variation of air infiltration

setback (°C)
during occupied and unoccupied hours. For the basecases, air infil-

21/15
21/21

23/15
21/10
21/10
23/10
18/10
21/10
21/10

21/10
21/10
21/10
21/10
21/10

21/10

18/10
21/10
21/10
21/10
21/10
21/10
21/10
tration during occupied hours is assumed to be 25% of the peak
infiltration rate. Values of 50% and 100% are used in the Medium
and High Infiltration Schedule cases to represent buildings that

damper position
are not airtight or that have poor air balancing during occupied

Min. VAV box

(large office)
hours.

0.15

0.15
3.2.3. Space-heating thermostat setting

0.3

0.3
0.3
0.5

0.3

0.3
0.3
0.3
0.3
0.3
0.3
0.5
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
A higher space-heating temperature setpoint increases space-
Infiltration schedule

heating loads. The basecases set the space-heating temperature

(occupied hours)

for occupied hours to 21 °C, which is typical for office buildings

in the United States. The High Heating Setpoint case raises the
heating thermostat setting to 23 °C, while the Low Heating Set-
point case lowers it to 18 °C.

0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25

0.25
0.25

0.25
0.25
0.25
0.25
0.25
0.25
0.25

0.5
1

1
3.2.4. Space-heating setback control
Infiltration rate

Setback control is usually used to lower space-heating thermo-

(m3/s m2)

0.001133
0.001133

0.000189
0.000189

0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302
0.000302

0.000302
0.000302
0.000302
0.000302
0.000302
0.000302

stats for office buildings during unoccupied hours. The basecases

assume a heating setback to 10 °C during unoccupied hours, which
may be too aggressive. Hence, two heating-setback cases are con-
sidered: One is set back to 15 °C to represent the typical operation
(large/small office)

of most office buildings, and the other is to have no setback at all to

represent the worst-case scenario.
0.68/0.4

0.4/0.2
0.4/0.2

0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2

0.4/0.2

0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2
0.4/0.2

0.1/0.1
0.4/0.2

3.2.5. VAV box minimum damper position

WWR

The large office building is served by central VAV systems,

which use reheat coils at the zone terminal boxes to provide space
EPD (W/m2)

heating. The terminal boxes have reverse acting dampers, which

can open beyond the minimum position during the heating mode
5.38
5.38

10.76
10.76
10.76
10.76
10.76
10.76
10.76
10.76

16.14
10.76
10.76
10.76
10.76

10.76
10.76
16.14

10.76
10.76
10.76
10.76

to meet zone heating loads. The higher the minimum damper posi-
tion, the more reheat energy can be consumed. The basecases have
LPD (W/m2)

the VAV box minimum damper position set to 30%, based on a typ-
ical design that meets recent building energy standards [33]; the
10.76

10.76
10.76
10.76

10.76

10.76
10.76
10.76
10.76
10.76
10.76

10.76
5.38
10.76

10.76
5.38

10.76
10.76
10.76
10.76

16.14
16.14

High Minimum VAV Box Damper Position case sets it to 50%, based
on surveys of office buildings [34]; and the Low Minimum VAV
High minimum VAV box damper position

Box Damper Position case sets it to 15%, based on proposals to up-

Low minimum VAV box damper position

date 2008 California and 2010 ASHRAE building energy-efficiency

standards.
Parametric of the simulation runs.

High boiler/Furnace Efficiency

Medium infiltration schedule

Thermostat setback to 15 °C

4. Simulation runs
High infiltration schedule

More envelope insulation

Less envelope insulation
Thermostat no setback
High Heating Setpoint
Low Heating Setpoint
High infiltration rate

Single-pane window
Triple-pane window
Low infiltration rate

Table 5 lists the parametric of the simulation runs for the two
Description of runs

High internal loads

Low internal loads

office buildings. There are 126 EnergyPlus simulation runs in total,

High Heating

including 22 runs for the large office building and 20 runs for the
Low Heating
High WWR
Low WWR

small office building for each of the three cities. These runs include
Basecase

the basecase, the High and Low Internal Loads cases, the High and
Table 5

Low Infiltration Rate cases, the High and Medium Infiltration Sche-
dule cases, the High and Low Minimum VAV Box Damper Position
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 521

cases for large office only, the High and Low Heating Setpoint cases, envelope insulation for the building design aspect have less impact
the High and Low WWR cases, the Single and Triple Pane Window on space-heating energy use.
cases, the More and Less Envelope Insulation cases, and the High From the operation aspect of the large office building, there are
Boiler/Furnace Efficiency cases. Each run varies only one parameter several significant findings:
from the basecase, except the High Heating case and the Low Heat-
ing case, which combine the worse and better values (based on the  The impact of minimum VAV box damper position setting
influence on space-heating energy) of the selected parameters, ranges from 12% to 128% for Chicago, 7.5% to 78% for Minne-
respectively. In the High Heating and Low Heating cases, the value apolis, and 2.5% to 31% for Fairbanks; the Low Minimum VAV
of the WWR stays the same as that of the basecase for both office Box Damper Position cases show small impact on space-heating
buildings. The High Heating and Low Heating cases aim to capture energy use. Oversizing the VAV box or setting the minimal dam-
the worst case of buildings that consume the most heating and the per position too high can result in huge space-heating penalty.
best case of buildings that consume the least heating.  The cases of Thermostat No Setback during unoccupied hours
In addition to the above-mentioned 126 runs, there are 60 runs increase the space-heating EUIs by 50–102% from the basecases
using the 30 historical years of weather data for Chicago and for the three climates while the cases of Thermostat Setback to
Fairbanks. 15 °C during unoccupied hours have increases from 10% to 12%.
Thermostat setback is an effective operation strategy to reduce
space-heating energy use during unoccupied hours.
5. Results and discussions  The impact of space-heating setpoint ranges from 44% to 65%
for Chicago, 32% to 44% for Minneapolis, and 19% to 22% for
5.1. Impact of design and operation parameters Fairbanks. Lowering the heating thermostat setpoint is an easy
way to save space-heating energy use during occupied hours.
Fig. 2 shows the percentages of change in space-heating EUIs  The cases of High Infiltration Rate show significant increase in
calculated by comparing the space-heating EUI from each para- space-heating energy use by 42%, 38.5%, and 38.9% for Chicago,
metric run to that of the basecase for the large office building in Minneapolis, and Fairbanks, respectively. However, the cases of
the three climates. Fig. 3 shows similar data for the small office Low Infiltration Rate have small impact on space-heating
building. Both figures are sorted by the percent changes for the energy use in all three climates, which may be because the
Chicago climate. low infiltration rates are only 50% lower than the basecases
Looking at results in Fig. 2 for the large office building, it can be while they are 375% higher for the High Infiltration Rate cases.
seen that: (1) based on the relative impact of the building opera-  From the design-efficiency aspect of the large office building, it
tion, the most influencing operation parameters are high minimum can be seen that:
VAV box damper position setting, thermostat without setback dur-  Triple-pane windows can save space-heating energy by 51%,
ing unoccupied hours, High Heating Setpoint, and High Infiltration 43%, and 29% for Chicago, Minneapolis, and Fairbanks, respec-
Rate; (2) for the building design efficiency aspect, the most influ- tively. On the other hand, single-pane windows increase
encing parameters are low and high internal load, window type, space-heating energy by 46% for Chicago, and 34% for both Min-
and window area; and (3) other parameters, including low air infil- neapolis and Fairbanks. High-performance windows with low
tration rate, low minimum VAV box damper position setting, infil- U-factors have great potential to reduce space-heating energy
tration schedule for the operation aspect, and boiler efficiency and use for cold climates.

Fig. 2. Impact of design and operation parameters on space-heating energy use of the large office building.
522 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

Fig. 3. Impact of design and operation parameters on space-heating energy use of the small office building.

 Window area also plays an important role. High WWR cases show very similar patterns to the large office buildings – operation
could increase space-heating energy use by 23–38%; on the parameters have greater impact than design parameters.
other hand, low WWR cases could save space heating by 24% From the operation aspect of the small office building, a few key
to 33%. results are:
 Internal loads can change space heating from 30% to 48%,
compared with the basecase in Chicago; 30% to 40% in Minne-  The impact of the space-heating setpoint ranges from 61% to
apolis; and 28% to 30% in Fairbanks. Internal loads from light- 61% for Chicago, 50% to 46% for Minneapolis, and 33% to
ing, equipment, and occupants directly reduce space-heating 26% for Fairbanks. The heating setpoint could be easily con-
loads. trolled by occupants in small office buildings because in pack-
 Wall insulation has less influence compared with other build- aged single-zone systems, thermostats are usually located in
ing-design parameters mainly due to the basecase has good office spaces. Decreasing the heating setpoint by 3 °C could save
insulation. From the simulated results, the More Envelope Insu- more than 50% in space-heating energy, while increasing the
lation cases used the latest version of the ASHRAE standard heating setpoint by 2 °C would consume more than 50% of
published in 2010, however, this high insulation level reduces space-heating energy in cold climates. Fairbank belongs to the
space-heating energy use by less than 5%. subarctic region; it needs more space-heating energy than other
climates at the same building design and operation conditions.
The relative impacts of the significant parameters on space Thus the percentage changes to space-heating energy use are
heating are consistent across the three climates, with Chicago smaller than other climates by adjusting the same degrees of
showing the largest impact, followed by Minneapolis and Fair- the heating setpoint.
banks. Fairbanks shows the least impact due to its high space-heat-  Not setting back the heating thermostat during unoccupied
ing EUI of the basecase compared with the other two climates. hours can increase space heating from 37% to 42% for the three
Similarly, the results in Fig. 3 for the small office building reveal climates.
that, based on the relative impact for the operation aspect, the  Similar to the results of the infiltration rate cases for the large
most influencing parameters are high and low space-heating set- office buildings, a High Infiltration Rate can significantly
point, thermostat without setback during unoccupied hours, high increase space heating by 41%, 37%, and 30% for Chicago, Minne-
air infiltration rate, and high infiltration schedule. For the build- apolis, and Fairbanks, respectively. The cases of Low Infiltration
ing-design efficiency aspect, the most influencing parameters are Rate in the three cities demonstrate relatively small impact
internal loads, triple-pane windows, and less envelope insulation. compared with other cases with same reason mentioned above.
All the parameters described above except window type and enve-  From the large office building design parameters, it can be seen
lope insulation can be controlled by building occupants or opera- that:
tors. Other parameters, including low WWR, high furnace  Internal loads can change the space heating from 40% to 52%
efficiency, and single-pane windows for the building design for Chicago compared with the basecase, 33% to 39% for Min-
parameters; and low infiltration rate and thermostat setback to neapolis, and 21% to 21% for Fairbanks. The variations of
15 °C during unoccupied hours for the operation aspect have smal- space-heating energy use due to changes to internal load are
ler impact on space-heating energy use. Small office buildings
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 523

Fig. 4. Comparison of space-heating energy use among the High Heating cases, the basecases, and the Low Heating cases for the large and small office buildings.

similar to the large office building in each city. Internal loads ing winter, especially December and January. A relatively small
from lighting, equipment, and occupants directly reduce the amount of heating may occur during summer for the large office
needs of space-heating loads. building, mainly due to heating the basement. There is almost no
 The High WWR cases increase the space heating from 12% to reheat during summer except for the High Minimum VAV
15% in the three climates; on the other hand, the Low WWR Box Damper Position case and the High Heating case. This agrees
cases could save space heating by about 5% to 10%. with common operational practice that setting the VAV box dam-
 Triple-pane windows can save space heating by 27%, 25%, and per wide open is one of the major causes of high reheat energy dur-
16% for Chicago, Minneapolis, and Fairbanks, respectively. ing summer.
 In all three climates, less window area and the use of single- Based on above analysis, it can be seen that space-heating en-
pane windows show relatively small influence on space-heating ergy use can be significantly reduced by more efficient building de-
energy use, which can be due to the tradeoff between the win- sign and even more so by improving the operation of space-heating
dow conduction heat losses and solar heat gains. systems. To improve the accuracy of the prediction of space-heat-
ing energy use by simulations, it is crucial to have proper inputs to
Fig. 4 benchmarks the space-heating EUI of the High and Low the most important design and operation parameters as identified
Heating cases against the basecases for both office buildings across in the study.
the three climates. There are huge differences in heating energy
use between the High Heating (the worst) cases and the Low Heat- 5.2. Impact of weather data
ing (the best) cases – by factors of about 60, 30, and 15 for both of-
fice buildings in Chicago, Minneapolis, and Fairbanks, respectively. To look at the impact of weather data on space-heating energy
The space-heating EUI ranges from 14.3 to 828.5 MJ/m2 (3.97– use, a percentage change of space-heating energy use is calculated
230.1 kWh/m2), 31.0 to 978.3 MJ/m2 (8.61–271.8 kWh/m2) and by comparing the space-heating energy use of a historical year
87.5 to 1315 MJ/m2 (24.3–365.3 kWh/m2) for the large office build- (from 1980 to 2009 for Chicago and Fairbanks) to that of the base-
ings in Chicago, Minneapolis, and Fairbanks, respectively. For the case using the TMY3 weather. Figs. 6 and 7 show that the impact of
small office buildings, the ranges are 6.5–482 MJ/m2 (1.81– weather data on space-heating energy use is significant for the
133.9 kWh/m2), 19.7–671 MJ/m2 (5.47–186.4 kWh/m2), and 93.4– large and small office buildings in both Chicago and Fairbanks.
1185 MJ/m2 (25.9–329.2 kWh/m2) in Chicago, Minneapolis, and For Chicago, the large office building shows that space-heating
Fairbanks, respectively. energy use varies from 18% to +33%; while the small office build-
Compared with the basecases, the High Heating cases signifi- ing shows variations from 24% to +33%. The results indicate that
cantly increase space-heating energy use by a factor of 3–5 for 1985 was the coolest year and 2006 the warmest year for both of-
the large office buildings in these climates; while for the small of- fice buildings across the 30-year period. Most warm years occurred
fice buildings, the increase in space-heating energy use is by a fac- from 1998 to 2006, while most cool years occurred from 1980 to
tor about 3. Similarly, compared with the basecases, the Low 1986.
Heating cases dramatically decrease space-heating energy use to For Fairbanks, the large office building shows that space-heat-
1/8, 1/5, and 1/3 for the large office buildings in Chicago, Minneap- ing energy use varies from 20% to +24%; while the small office
olis, and Fairbanks, respectively; and to 1/17, 1/9, and 1/4 for the building shows variations from 17% to +22%. The results indicate
small office buildings. that 1999 was the coolest year and 1981 was the warmest year for
For the large office building served by VAV systems with zone both office buildings across the 30-year period. Most warm years
reheat, potential reheat during the summer cooling season is a occurred from 2000 to 2003, while most cool years occurred from
waste of energy and thus increases space-heating energy use. 1988 to 1999.
Fig. 5 shows monthly space-heating EUIs of the large office build- The impact of weather on space-heating energy use is very con-
ing in Chicago. It can be seen that most of the heating is used dur- sistent across both office buildings in the same climate, but is very
524 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

Fig. 5. Monthly space-heating energy use of the large office building in Chicago.

Fig. 6. Multiyear space-heating energy use in Chicago.

different across both climates – the coolest and warmest years oc- Comparing the space-heating energy use of the coolest year to
curred differently during the period from 1980 to 2009. the warmest year for Chicago, the increase is 51% and 57% for
It should be noted that space-heating energy use from simula- the large and small office buildings, respectively; while for Fair-
tions using TMY3 weather data can underestimate by up to 33% banks, the increases are 44% and 39%. Thus it is crucial to run sim-
or overestimate by up to 25% compared with using historical ulations with multiple decades of weather data to fully evaluate
weather data. For Fairbanks, simulated results using TMY3 under- the impact of weather on the energy performance of space-heating
estimate space-heating energy for most of the years during the 30- systems in buildings.
year period.
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 525

Fig. 7. Multiyear space-heating energy use in Fairbanks.

Fig. 8. Benchmarking simulation results with the building databases for the large office building in Chicago.

5.3. Benchmarking with building databases Fig. 8 shows the benchmark results for the large office in Chi-
cago. From CBECS, 10 buildings were found with floor area ranging
To form a clear picture of how space-heating energy use varies from 18,580 to 46,450 m2, vintage 1990–2003. The space-heating
in actual buildings, we selected ones from the two databases for EUIs for these buildings vary from 136.7 to 559.72 MJ/m2 (38.0–
Chicago and Minneapolis that were similar to the simulated large 155.5 kWh/m2). Fig. 9 shows the benchmark results for the small
and small office buildings in terms of building type or function, office in Chicago. The selection criteria for the CBECS are set as fol-
size, location, and construction age. Figs. 8–11 show both the sim- lows: (1) floor area from 93 to 9290 m2, (2) vintage 1990–2003,
ulated and the actual space-heating EUIs. Each horizontal line rep- and (3) location in Chicago. Seven such small office buildings were
resents result from a selected building in one of the two databases. found from the CBECS with a space-heating EUI from 249 to
The solid lines represent buildings selected from the CBECS data- 1023 MJ/m2 (69.2–284.2 kWh/m2). Two small office buildings
base, while the dashed lines representing buildings from the HPB were found from the HPB database that are near Chicago and have
database. a floor area of 1390 and 3716 m2. The two offices have space-heat-
In general, space-heating EUIs vary significantly for the selected ing EUIs of 208.8 and 335.2 MJ/m2 (58.0–93.1 kWh/m2).
buildings from both databases and even more across the two For Chicago, the simulated results are always much lower than
databases. the databases except for the High Heating case. Although the High
526 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

Fig. 9. Benchmarking simulation results with the building databases for the small office building in Chicago.

Fig. 10. Benchmarking simulation results with the building databases for the large office building in Minneapolis.

Heating case results overlap with some low-end results from the building. In Fig. 10, three buildings, selected based on the criteria
databases, it is much lower than the high-end results – by more of floor area larger than 9290 m2 and vintage 1990 and 2003, have
than 20%. This implies that there might be other important param- space-heating EUIs ranging from 150.7 to 299.3 MJ/m2 (41.9–
eters that should be considered in simulations; for example, design 83.1 kWh/m2, by a factor of 2). On the other hand, eight small office
and operation problems or faults of the space-heating systems. buildings were selected based on floor area from 93 to 9290 m2
Minneapolis belongs to the East North Central division of the and vintage 1990 to 2003. The space-heating EUI of the eight build-
Midwest region in the CBECS. Fig. 10 shows the benchmark results ings vary from 122.7 to 845 MJ/m2 (34.1–234.7 kWh/m2, by a fac-
for the large office building, and Fig. 11 shows the small office tor of 7). Only one small building was found from the HPB with
 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528 527

Fig. 11. Benchmarking simulation results with the building databases for the small office building in Minneapolis.

floor area of 1104 m2. The space-heating EUI of this office is to run simulations with multiple decades of actual weather data,
75.9 MJ/m2 (21.1 kWh/m2). considering the availability and affordability of such data and
It should be noted that there are uncertainties associated with low extra cost of running such simulations on current PCs with
the two benchmark databases: (1) the space-heating energy uses high computing power. Besides, dynamic analyses should be intro-
are not from actual measurements; rather, they are calculated from duced to study the integrated effect of driving factors to space
statistical analysis (CBECS) or energy modeling (HPB); and (2) the heating energy use.
floor area used to calculate the EUI might not accurately match The actual space-heating energy use for the similar office build-
the actual floor area of the buildings. Furthermore, the buildings ings from the CBECS and HPB databases also vary significantly,
selected from the databases may not exactly match the simulated with wide ranges that well overlap the variation ranges of the sim-
buildings in terms of floor area, vintage, and location. This contrib- ulated results. Based on the study, simulations do not necessarily
utes to discrepancies between the simulated and benchmarked always under- or overpredict space-heating energy use. The simu-
space-heating energy uses. lated space-heating energy use depends on building type, configu-
ration, and climate, with a few special key influential building
design and operation parameters.
6. Conclusions High-efficiency designs and better operation of buildings can
reduce space-heating energy use, but the latter plays a more
The simulated space-heating energy use of the small- and large- important role. For building designers, paying more attention to
size office buildings across the three heating climates can vary sig- the most influential design parameters has significant potential
nificantly, depending on details of a few key building design and to reduce space-heating energy use for new buildings. For building
operation parameters. The most influencing parameters are owners and operators, improving building operations through
space-heating temperature setpoint and setback strategies, air commissioning and retrofits to control key operation parameters
infiltration, VAV terminal box damper minimum position settings is an effective way to reduce space-heating energy use for existing
for the large office, window type, WWR, and internal loads. The rel- buildings. Finally, for energy policy makers, enforcing more strin-
ative impacts of these parameters vary with building type and gent regulations on these design and operation parameters can sig-
climate. nificantly reduce space-heating energy use in new and existing
Compared with the basecase, the High Heating case consumes buildings.
more than double the space-heating energy, while the Low Heating To lower space heating energy use in office buildings, the fol-
case consumes less than half for both office buildings in all the lowing steps are recommended: (1) lower space-heating tempera-
three climates. ture setpoint while maintaining thermal comfort, (2) use heating
For the two climates with the 30-year historical weather data, thermostat setback during unoccupied hours, (3) reduce air infil-
the simulated space-heating energy use for a particular year can tration rate by improving air tightness of the building envelope,
vary dramatically, compared with the average results across the (4) decrease the minimum damper position settings of VAV termi-
30-year period. For the basecase, the simulated space-heating en- nal boxes if applicable, and (5) replace with better insulated
ergy use with the TMY3 weather data can overpredict by 24% windows.
and underpredict by 34%, compared with the results with the his- This study did not look at other influencing factors of space
torical weather data. To understand the long-term impact of heating, such as building occupancy level and operational faults
weather on space-heating energy use, it is critical and necessary of space-heating systems. Building occupancy levels vary case by
528 H.-W. Lin, T. Hong / Applied Energy 111 (2013) 515–528

case, and insufficient data is available for this study. Future re- [19] COMFEN development team of the Lawrence Berkeley National Laboratory.
CBECS/eQUEST/EnergyPlus Comparison, 2010. Personal communication.
search can study the impact of HVAC operational faults on space-
[20] Heller J, Heater M, Frankel M. Sensitivity analysis: comparing the impact of
heating energy use. Other potential causes of high space-heating design, operation, and tenant behavior on building energy performance.
energy in actual buildings might relate to occupants opening win- Report of the New Building Institute; 2011 July.
dows during heating season, space overheating due to lack of tem- [21] Rezaie B, Rosen AM. District heating and cooling: review of technology and
potential enhancements. Appl Energy 2012;93:2–10.
perature controls, heat losses from air ducts, hot-water piping, and [22] Field K, Deru M, Studer D. United States Department of Energy commercial
boilers that might not be counted well or at all in energy-modeling reference building models of the national building stock. Fourth National
programs. Similar analysis can be done for other building types and Conference of IBPSA-USA; 2010. p. 85–93.
[23] Energy standard for buildings except low-rise residential buildings, ANSI/
climates, and to aggregate the impacts at the regional and national ASHRAE/IESNA Standard 90.1-2004; 2004.
levels. [24] Energy standard for buildings except low-rise residential buildings, ANSI/
ASHRAE/IESNA Standard 90.1-2010; 2010.
[25] Typical Meteorological Year 3 (Internet). United States Department of Energy.
Acknowledgement <[Link]
new_construction.html>.
This work was supported by the U.S. Department of Energy un- [26] Wilcox S, Marion W. User’s Manual for TMY3 Data Sets. Report of National
Renewable Energy Laboratory (NREL); 2008.
der the U.S.–China Clean Energy Research Center on Building En- [27] EnergyPlus Version 7.2 (Internet). United States Department of Energy.
ergy Efficiency. It was co-sponsored by the Bureau of Energy, <[Link]
Ministry of Economic Affairs, Taiwan, ROC. [28] High performance buildings database (Internet). United States Department of
Energy. <[Link]
[29] Energy code for commercial and high-rise residential buildings, ANSI/ASHRAE/
References IESNA Standard 90.1-1989; 1989.
[30] Fisher SM, Sultan N, Stromquist RC. Plug load reduction for a net zero energy
[1] United States Department of Energy. 2010 Buildings Energy Data Book; 2011. building, ACEEE summer study on energy efficiency in buildings. California
[2] United States Energy Information Administration. Commercial Buildings (USA): Pacific Grove; 2006. p. 119–23.
Energy Consumption Survey. <[Link] [31] Emmerich SJ, McDowell T, Anis W. Investigation of the impact of commercial
[3] Energy Information Administration (EIA), United States Department of Energy. building envelope airtightness on HVAC energy use, Report of National
Annual Energy Review 2007. Report No. 0384; 2008. Institute of Standards and Technology, NISTIR-7238; 2005.
[4] Pedrini A, Westphal FS, Lamberts R. A methodology for building energy [32] Persily A, Gorfain J. Analysis of ventilation data from the U.S. Environmental
modeling and calibration in warm climates. Build Environ 2002;37:903–12. Protection Agency building assessment survey and evaluation (BASE) study,
[5] Lam JC. Energy analysis of commercial buildings in subtropical climates. Build Report of National Institute of Standards and Technology, NISTIR 7145; 2008.
Environ 2000;35:19–26. [33] Building energy efficiency standards for residential and nonresidential
[6] Dai X, Wu Y, Di Y, Li Q. Government regulation and associated innovations in buildings, California Energy Commission; 2005.
building energy-efficiency supervisory systems for large-scale public buildings [34] Advanced variable air volume system design guide, California Energy
in a market economy. Energy Policy 2009;37:2073–8. Commission; 2005.
[7] Yamaguchi Y, Shimoda Y, Mizuno M. Transition to a sustainable urban energy
system from a long-term perspective: case study in a Japanese business
district. Energy Build 2007;39:1–12.
Glossary
[8] Andrews CJ, Krogmann U. Explaining the adoption of energy-efficient
technologies in U.S. commercial buildings. Energy Build 2009;41:287–94. ASHRAE: American Society of Heating, Refrigeration, and Air-conditioning
[9] Liu P, Pistikopoulos EN, Li Z. An energy systems engineering approach to the Engineers
optimal design of energy systems in commercial buildings. Energy Policy BESTest: Building Energy Simulation Test
2010;38:4224–31. BLAST: Building Loads Analysis and System Thermodynamics
[10] Linden AL, Carlsson-Kanyama A, Eriksson B. Efficient and inefficient aspects of CBECS: Commercial Buildings Energy Consumption Survey
residential energy behaviour: what are the policy instruments for change? CDD: cooling degree day
Energy Policy 2006;34:1918–27. CRB: commercial reference building
[11] Branco G, Lachal B, Gallinelli P, Weber W. Predicted versus observed heat DX: direct expansion
consumption of a low energy multifamily complex in Switzerland based on
EIA: Energy Information Administration
long-term experimental data. Energy Build 2004;36:543–55.
EPD: equipment power density
[12] Santin OG. Behavioural patterns and user profiles related to energy
EUI: energy use intensity
consumption for heating. Energy Build 2011;43:2662–72.
[13] Pan D, Chan M, Deng S, Lin Z. The effects of external wall insulation thickness HDD: heating degree day
on annual cooling and heating energy uses under different climates. Appl HPB: high-performance building
Energy 2012;97:313–8. IEA: International Energy Agency
[14] Yang L, Lam JC, Tsang CL. Energy performance of building envelopes in IEAD: insulation entirely above deck
different climate zones in China. Appl Energy 2008;85:800–17. LPD: lighting power density
[15] Dovjak M, Shukuya M, Olesen BW, Krainer A. Analysis on exergy consumption MJ: mega joules
patterns for space heating in Solvenian buildings. Energy Policy NFRC: National Fenestration Rating Council
2010;38:2998–3007. SHC: Solar Heating and Cooling Programme
[16] Yildiz A, Gungor A. Energy and exergy analyses of space heating in buildings. SHGC: solar heat gain coefficient
Appl Energy 2009;86:1939–48. TRNSYS: Transient System Simulation Tool
[17] Eskin N, Turkmen H. Analysis of annual heating and cooling energy
WWR: window-wall ratio
requirements for office buildings in different climates in Turkey. Energy
VAV: variable air volume
Build 2008;40:763–73.
VT: visible transmittance
[18] Center for the Built Environment, University of California at Berkeley. eQuest
vs EnergyPlus comparison, 2011. Personal communication.
 Modeling and Simulation of HVAC Faulty Operations and

Performance Degradation due to Maintenance Issues

Liping Wang*, Tianzhen Hong

Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720,

USA

ABSTRACT
Almost half of the total energy used in the U.S. buildings is consumed by heating,
ventilation and air conditionings (HVAC) according to EIA statistics. Among various
driving factors to energy performance of building, operations and maintenance play a
significant role. Many researches have been done to look at design efficiencies and
operational controls for improving energy performance of buildings, but very few
study the impacts of HVAC systems maintenance. Different practices of HVAC
system maintenance can result in substantial differences in building energy use. If a
piece of HVAC equipment is not well maintained, its performance will degrade. If
sensors used for control purpose are not calibrated, not only building energy usage
could be dramatically increased, but also mechanical systems may not be able to
satisfy indoor thermal comfort. Properly maintained HVAC systems can operate
efficiently, improve occupant comfort, and prolong equipment service life.

In the paper, maintenance practices for HVAC systems are presented based on
literature reviews and discussions with HVAC engineers, building operators, facility
managers, and commissioning agents. We categorize the maintenance practices into
three levels depending on the maintenance effort and coverage: 1) proactive,
performance-monitored maintenance; 2) preventive, scheduled maintenance; and 3)
reactive, unplanned or no maintenance. A sampled list of maintenance issues,
including cooling tower fouling, boiler/chiller fouling, refrigerant over or under
charge, temperature sensor offset, outdoor air damper leakage, outdoor air screen
blockage, outdoor air damper stuck at fully open position, and dirty filters are
investigated in this study using field survey data and detailed simulation models. The
energy impacts of both individual maintenance issue and combined scenarios for an
office building with central VAV systems and central plant were evaluated by
EnergyPlus simulations using three approaches: 1) direct modeling with EnergyPlus,
2) using the energy management system feature of EnergyPlus, and 3) modifying

*
Corresponding author email: lwang@[Link]
EnergyPlus source code. The results demonstrated the importance of maintenance for
HVAC systems on energy performance of buildings.
The research is intended to provide a guideline to help practitioners and building
operators to gain the knowledge of maintaining HVAC systems in efficient operations,
and prioritize HVAC maintenance work plan. The paper also discusses challenges of
modeling building maintenance issues using energy simulation programs.

KEYWORDS
Building energy use, Building simulation, EnergyPlus, Fault modeling, HVAC
maintenance

INTRODUCTION
Almost half of the total energy used in the U.S. buildings is consumed by heating,
ventilation and air conditioning (HVAC) according to U.S. Energy Information
Administration statistics. For too long, high efficiency design and optimum
operational controls to improve energy performance of buildings have been the focus,
and deficiencies in building operation and maintenance have been neglected. In fact,
among various driving factors to energy performance of building, operation and
maintenance play a decisive role. HVAC maintenance keeps plant and HVAC
equipment in a healthy state in which HVAC system can function properly. This also
includes troubleshooting of defective equipment to perform the intended function in a
cost efficient manner, thus extending life serving time. Mill (2009) identified a wide
diversity of system deficiencies and report frequency of system deficiencies for
existing building and new constructions. The study also found that the most common
problems were in air-handling and distribution systems for existing buildings.

Different practices of HVAC system maintenance can result in substantial differences

in building energy use, maintenance costs, and equipment life. Based on discussions
with HVAC engineers, building operators, facility managers, and commissioning
agents, and literature review on maintenance standards (ASHRAE 2008, 2012; CIBSE
2008), maintenance practices for HVAC systems can be categorized into three levels
depending on the maintenance effort and coverage:
1) proactive maintenance
The performance-monitored maintenance represents the good practice. The system
operation problems can be identified and repaired before a failure occurs. It allows the
maintenance manager has control over maintenance.
2) preventive, scheduled maintenance
This practice represents the average practice (business as usual). In this practice,
maintenance is scheduled over time. For example, a filter in an air handler unit is
replaced every 6 months. Preventive maintenance program may take too long to
demonstrate results or fail to justify its cost.
3) reactive, unplanned maintenance
This maintenance repairs or replaces equipment only when it fails and investigates
system performance issues based on occupants’ complaints. It is often practiced by
facilities that are significantly understaffed and underfunded.
Table 1 summarizes the three practices of HVAC maintenance and their implications
on equipment operating efficiency and energy use, equipment life, short term
maintenance cost, and life cycle cost including maintenance cost, energy cost, and
equipment replacement or repair cost. The good practice will lead to lowest life cycle
cost, while the bad practice seems to save short term maintenance cost, it will result in
the highest life cycle cost.

Table 1. Three types of HVAC maintenance practices

Investigating the impacts of HVAC maintenance issues on building performance is a

complicated research subject. The research requires not only on a good understanding
of common practices on maintenance issues but also modeling techniques to simulate
operation deficiencies. Most building energy simulation programs available today
have limited capabilities of directly modeling HVAC operational faults or
maintenance issues which occur in almost every building. Basarkar et al. (2012)
implemented four types of equipment faults in a development version of EnergyPlus
to simulate common faulty operation in building systems. The purpose of this study,
reported here, is to continue previous research on fault modeling, develop modeling
and simulation methods for maintenance issues and assess the impacts of common
maintenance issues on building performance.

TECHNICAL APPROACH
EnergyPlus is used as the simulation tool in the study for modeling maintenance
issues. EnergyPlus, developed by U.S. Department of Energy, is an open-source
whole-building energy simulation program built upon sub-hourly zone heat balance
and integrated solutions of building loads, HVAC systems, and central plant
equipment. Three different approaches using EnergyPlus, in order of difficulty, are
used to model HVAC maintenance issues:

1) Direct modeling with EnergyPlus (Direct Modeling)

Maintenance issues are directly modeled using existing inputs (either design input
parameters or performance curves) in the current version of EnergyPlus. This
modeling approach can be applied to such maintenance issues as supply air sensor
offset, zone thermostat offset and outdoor air damper leakage. This approach is also
applied to model simplified maintenance issues such as chiller or boiler fouling by
introducing a degradation factor to the chiller or boiler efficiency inputs to the
EnergyPlus models. The advantage of this approach is easy implementation.
2) Using the energy management system (EMS) in EnergyPlus
EMS is an advanced feature of EnergyPlus and designed for users to develop
customized high-level, supervisory control routines to override specified aspects of
EnergyPlus modeling in the EMS program. The EMS feature in EnergyPlus is flexible
to allow users to simulate equipment operating with some maintenance issues by
overwriting or adding algorithms in EnergyPlus within the specified aspects of current
EMS capability. Use of EMS feature may require advanced knowledge of EnergyPlus
and computer programming. EMS is used to model maintenance issues like dirty
filters which increase pressure drop across the filter with operating hours.
3) Modifying EnergyPlus source code (Modified Code)
Modifying the existing EnergyPlus source code, the third modeling approach, is used
when both direct modeling and EMS approaches cannot be applied to simulate any
particular equipment or system deficiencies. This approach requires users to have a
thorough understanding of the existing EnergyPlus source code and to write your own
custom computer program based on existing code. Such HVAC maintenance issues as
cooling coil fouling, outdoor air and return air temperature sensors offset adopt the
third approach.

SAMPLED HVAC MAINTENANCE ISSUES

A list of common HVAC maintenance issues are reviewed and selected for the initial
modeling and simulations. Based on literature reviews and our understanding of the
physics and implications for each maintenance issues, we developed corresponding
models and simulation approaches. Table 2 lists the issues with their potential impacts
and modeling approach according to maintenance types, including sensor calibration,
filter replacement, heat exchanger treatment, mechanical repair and refrigerant charge,
are investigated using detailed simulation models.

Each maintenance issue list in Table 2 was modelled using EnergyPlus. A description
of the implement model for selected maintenance issues is as follows.
Temperature sensor offset
Control sensors such as supply air temperature (SAT) sensors, zone thermostats, and
outdoor air temperature (OAT) sensors may be out of calibration over a long term
operation period. In this study, it is assumed that temperature sensors are offset by ±2˚C.
For example, if a SAT sensor is offset by +2 ˚C and a designed supply air temperature to
control is 13˚C, the actual supply air temperature due to sensor offset is 11 ˚C.
Dirty filter
In terms of filter replacement for reactive maintenance, it is assumed that filters in air handler
units have not been replaced over a year. Therefore, pressure drop for air handler units has
been increased and the maximum additional pressure drop is 500 Pa.
Fouled cooling tower
Cooling towers can become fouled due to unfavourable conditions. The study assumes certain
fouling condition that overall heat transfer coefficient is reduced to 85% of design value.

Table 2. List of sampled HVAC maintenance issues

Maintenance Maintenance Simulated Modeling
Impacts Approach
Types Issues Scenarios
Supply air
temperature Direct modeling,
sensor (SAT) adjust SAT setpoint
offset
controls, heating temperature
Sensor Direct modeling ,
Zone temperature and cooling sensors are offset
Calibration adjust thermostat
sensor offset energy by ±2˚C
settings
Outdoor air Modified Code,
temperature modify the
sensor offset economizer controls
pressure drop, additional 500Pa EMS, adjust fan
Filter
Dirty filter fan energy, of air pressure power for VAV
replacement
airflow drop systems
overall heat
transfer Direct modeling,
Fouled cooling
efficiency coefficient is adjust cooling tower
tower
reduced to 85% UA
of design UA
Direct modeling,
Chiller: fouled chiller COP is
efficiency adjust chiller
Heat exchanger tubes reduced by 10%
efficiency
cleaning/treatme
boiler efficiency Direct modeling,
nt Boiler: hard water
efficiency is reduced by adjust boiler
scale
10% efficiency
overall heat
transfer
Fouled heating efficiency, Modified Code,
coefficient is
/cooling coil comfort adjust coils UA
reduced to 50%
of design UAs
Direct modeling,
Outdoor air heating and 30% OAD
adjust minimum OA
damper leakage cooling energy leakage
flow
OAD is stuck at
Stuck outdoor air heating and EMS, set constant
fully open
damper (OAD) cooling energy OA flow
position
Mechanical
outdoor air flow
repair
is less than 100%
maximum
during
Clogged OA percent of intake Direct modeling, set
economizer
screen fresh air is maximum OA flow
mode thus
reduced to 70%
increasing
cooling energy
Chiller: over or Direct modeling,
Refrigerant chiller COP is
under 10% efficiency adjust chiller
charge reduced by 10%
refrigerant charge efficiency
Fouled Chiller/Boiler/Coils
Fouling on heat transfer surfaces of boiler and chiller increases the thermal resistance and
leads to reduced heat transfer. For the scenario of chiller/boiler fouling, both chiller COP and
boiler efficiency are assumed to be reduced by 10%. For fouled cooling/heating coils, overall
heat transfer coefficients are assumed to be reduced to 50% of design UAs.
Outdoor air damper (OAD) leakage
In the study, it is assumed that OAD leakage level is 30%. When the commanded
outdoor air fraction is smaller than the leakage level, leaky damper cannot effectively
control the air intake.
Stuck outdoor air damper (OAD)
Stuck OAD due to control and mechanical failure is another common fault in field. In this
study, OAD is assumed to get stuck at fully open position. Cooling and heating energy
penalties are introduced when outdoor air is not favourable for free cooling.
Clogged OA screen
Outdoor air intake screens may get clogged due to unfavourable locations or weather
condition. The maximum percent of intake fresh air is assumed to reduce to 70%.

RESULTS AND DISCUSSIONS

Mechanical
Repair

Heat
exchanger
cleaning/treat
ment

Sensor
Calibration

Figure 1. The impacts of poor HVAC maintenance on HVAC source energy

consumption for a large office building in Chicago, USA

The energy penalty introduced by HVAC maintenance issues varies by a few factors
including building and HVAC systems types, vintage (design efficiencies), and
climates. In the study, the commercial building reference model (Anon.) for a
large-size office building in compliance with ASHRAE Standard 90.1-2004 is used as
a baseline representing good maintenance practice. The large-size office building
consists of one basement level and 12 floors above ground served by 4 built-up VAV
systems with 2 water-cooled chillers and one natural gas hot-water boiler.
The results, shown in Figure 1, demonstrated the energy penalty introduced by the
reactive maintenance practice for the built-up VAV system located in Chicago. The
percentages are derived by comparing the total source/primary energy use of HVAC
systems for the reactive maintenance practice to those of the good practice (baseline
reference model). The maintenance issues with significant energy impacts for
Chicago are OA damper stuck at 100% position, blocked OA screen, supply air
temperature offset, boiler/chiller fouling, and chiller refrigerant under/overcharge.
Although there is no significant energy impact due to heating/cooling coil fouling, the
numbers of unmet thermal comfort hours for both heating and cooling are
significantly increased due to reduced system cooling and heating capacities. Two
combined scenarios (#1 and #2) with different temperature sensor offsets were
simulated in the study. The overall energy penalty by combining the sampled
maintenance issues including sensor offset by +2 ˚C can reach 85% of overall HVAC
energy consumption for Chicago climate.

Table 3. The impacts of poor HVAC maintenance on HVAC end-use energy

consumption for a large office building in Chicago, USA
Cooling
Maintenance Issues Cooling/Chiller Heating/Boiler Fans Pumps
Tower
OA Damper Leak -0.04% 8.01% 0.00% 0.67% 0.06%
Clogged OA Screen 13.47% 0.00% -0.01% 14.88% 16.27%
OA Damper Stuck at 100% 0.79% 183.93% 27.20% 11.37% 0.99%
Filter Replacement 0.57% -1.24% 63.69% 0.82% 0.60%
Cooling Coil Fouling 1.52% -0.01% 2.49% 4.14% -0.12%
Heating Coil Fouling 0.00% -0.62% 0.05% 2.48% -0.03%
Cooling Tower Fouling 1.36% - - 0.00% 0.11%
Boiler Fouling - 14.71% - - -
Chiller Fouling 10.24% - - 1.00% 1.58%
Refrigerant
10.24% - - 1.00% 1.58%
Under/Overcharge
OAT Offset (+2˚C) 0.57% 1.52% 0.00% -0.14% -0.59%
OAT Offset (-2˚C) 1.20% -0.01% 2.87% 4.10% -0.18%
SAT Offset (+2˚C) 6.87% 18.08% -5.05% 12.14% 7.74%
SAT Offset (-2˚C) -7.71% -7.37% 13.06% -14.75% -8.74%
Thermostat Offset (+2˚C) 3.83% -29.02% 21.57% 3.77% 5.00%
Thermostat Offset (-2˚C) -3.05% 22.79% -25.10% -5.37% -1.68%

Table 3 shows the impacts of maintenance issues on HVAC end-use energy

consumption for chillers, boilers, fans, pumps and cooling towers. In the table, the
value in each cell represents the percentage change of HVAC end-use energy
consumption relative to that of the baseline model. If OA dampers in the built-up
VAV system get stuck at 100%, heating and fan energy uses increase by 184% and
27%, respectively. As variable speed pumps are used in both chilled and hot water
loops, pump energy uses have been increased for fouled cooling and heating coils.
Cooling tower fouling causes a small increase in cooling energy use (no pump energy
increase as the pump is constant speed in the condenser loop). Outdoor air
temperature sensor offset interferes control thresholds for various operation modes of
air-side economizers and therefore introduces extra energy use for heating and
cooling. Supply air temperature sensor offset by +2 ˚C introduces about 7% cooling
penalty, 18% heating penalty due to increased reheat, 5% less fan energy use due to
the reduction of overall air flow rates, which SAT sensor offset by -2 ˚C reduces
cooling and heating energy use by increasing the actual controlled supply air
temperature.

CONCLUSION
This study applied three different approaches for modeling common HVAC
maintenance issues. Sixteen scenarios on individual maintenance issues and 2
combined scenarios were simulated. The results demonstrated that the combined
impacts of the selected maintenance issues on building energy use for a large office
building in Chicago climate can reach up to 85% of overall HVAC source energy use
due to reactive maintenance practice.
Our on-going research focuses on identifying a broader list of HVAC maintenance
issues for commercial buildings in various climates, and developing modeling
approaches. The research findings can be used to provide a guideline to help
practitioners and building operators to gain the knowledge of maintaining HVAC
systems in efficient operations, and prioritize HVAC maintenance work plan.

ACKNOWLEDGEMENTS
This work was supported by the Assistant Secretary for Energy Efficiency and
Renewable Energy of the U.S. Department of Energy for the US-China Clean Energy
Research Center.

REFERENCES
ASHRAE. 2012. ASHRAE Guideline 32-2012, Sustainable, High-Performance
Operations and Maintenance, Atlanta: American Society of Heating, Refrigerating,
and Air-Conditioning Engineers, Inc.
ASHRAE. 2008. ANSI/ASHRAE 180-200, Standard Practice for Inspection and
Maintenance of Commercial Building HVAC Systems, Atlanta: American Society
of Heating, Refrigerating, and Air-Conditioning Engineers, Inc.
CIBSE. 2008. Guide M, Maintenance engineering and management
Anonymity [Link]
[Link], last accessed on June 2012.
Basarkar M., Pang X. Wang L., Haves P., Hong T. 2012. Modeling and simulation of
HVAC faults in EnergyPlus, Proceedings of Building Simulation 2011:12th
Conference of International Building Performance Simulation Association,
Sydney, 14-16 November.
BUILD SIMUL (2011) 4: 149 – 167
DOI 10.1007/s12273-011-0044-5

A novel approach for building occupancy simulation

Research Article
Chuang Wang, Da Yan ( ), Yi Jiang

Department of Building Science, School of Architecture, Tsinghua University, Beijing 100084, China

Abstract Keywords
Building occupancy is an important basic factor in building energy simulation but it is hard to building occupancy,
represent due to its temporal and spatial stochastic nature. This paper presents a novel approach occupant movement,
for building occupancy simulation based on the Markov chain. In this study, occupancy is handled stochastic process,
as the straightforward result of occupant movement processes which occur among the spaces Markov chain,
inside and outside a building. By using the Markov chain method to simulate this stochastic energy simulation
movement process, the model can generate the location for each occupant and the zone-level
occupancy for the whole building. There is no explicit or implicit constraint to the number of
Article History
Received: 27 April 2011
occupants and the number of zones in the model while maintaining a simple and clear set of
Revised: 26 May 2011
input parameters. From the case study of an office building, it can be seen that the model can
Accepted: 28 May 2011
produce realistic occupancy variations in the office building for a typical workday with key
statistical properties of occupancy such as the time of morning arrival and night departure, lunch
© Tsinghua University Press and
time, periods of intermediate walking-around, etc. Due to simplicity, accuracy and unrestraint, this
Springer-Verlag Berlin Heidelberg
model is sufficient and practical to simulate occupancy for building energy simulations and
2011
stochastic analysis of building heating, ventilation, and air conditioning (HVAC) systems.

1 Introduction other is how they interact with building devices including

windows, doors, blinds, air conditioning terminals, lights,
Building energy simulation tools such as EnergyPlus (Crawley and equipment (TVs, computers, printers, etc.). In most
et al. 2001), ESP-r (ESRU 1999), DeST (Yan et al. 2008; situations, occupants have the right and means to adjust and
Zhang et al. 2008), and TRNSYS (Klein et al. 2004) have control these devices, thus, those interactions are closely
been playing more and more important role in building related to occupancy. For example, lights in a zone might
energy conservation since their conception. In past decades, be turned on by occupants when the zone is occupied and

Building Thermal, Lighting,

and Acoustics Modeling
modeling of heat and mass transfer processes, ambient might be turned off if there is no occupant. In some smart
weather data, as well as HVAC (heating, ventilation, and buildings, occupancy sensors are installed for automatic
air conditioning) systems in buildings have been the main control of devices according to the occupied status of the
focus and have been well established. They support the monitored space to reduce unnecessary energy use while
application of simulation technique to predict and evaluate maintaining the comfort level of the environment. For
the performance of buildings (indoor climate, energy buildings with occupancy controls, occupancy becomes a key
consumption, etc.). Occupant behavior, as a basic factor in driving factor to accurately predict the energy consumption
building performance, still remains a big issue because of of the buildings or the impact of such occupancy-based
its stochastic nature in time and space. control system on building energy performance.
In general, the behavior of building occupants can be The load calculation and performance analysis of building
broken down into two aspects: one is how they occupy the HVAC systems depend on accurate accounting of internal
building (when they occupy the building and how many heat gains due to occupancy (and associated use of lights
people for each zone), which is denoted by occupancy. The and equipment), especially for interior zones in large public

E-mail: yanda@[Link]
150 Wang et al. / Building Simulation / Vol. 4, No. 2

or commercial buildings where the internal heat gains are process model with two different exponential distributions
the most important factor that affects the indoor thermal are used to generate the occupancy sequence in a single
environment. In such situations, the uncertainty or stochastic person office. Based on the examined statistical properties
variation of internal heat gains usually results in overcooling of occupancy, the durations of presence and absence
or overheating during HVAC system operation, and thus during business hours are both exponentially distributed
plays a decisive role in the performance evaluation of a HVAC and the coefficient of each exponential distribution for a
system, such as the comparison between CAV (constant air single office can be treated as a constant over the workday.
volume) and VAV (variable air volume) system (Yan 2005). Meanwhile, in order to combine the clock-time information
Therefore, how the stochastic characteristic of occupancy into the simulation, the morning arrival time, the night
in building can be defined and modeled is an important issue departure time, and the lunch break time are assumed to
that remains to be addressed in building energy simulation. be normally distributed, which are actually not supported
Currently the most commonly used method in simulation by observations. This method is simple and elegant for
tools to represent occupancy is the so-called schedule, diversity single offices, but it is severely limited in describing the
profile or diversity factors (DeST 2008; EnergyPlus 2009; relationship of occupancy in different zones due to occupant
Davis and Nutter 2010), which can describe the time-variation movement in a building, and thus very hard to extend into
of occupancy. A daily schedule consists of 24 hourly values other situations.
while a yearly schedule usually consists of 365 daily schedules, Page et al. (2008) presented an approach based on the
in which the hourly values can be estimated from individual inhomogeneous two-state Markov chain, where the model
experience or onsite survey; the same schedule is usually can generate a time series of the state of presence (in/out,
used for zones that have similar functions. Through this or, present/absent) of occupants within a specific zone of
deterministic approach, the average impact of internal heat a building, and the transition probabilities of the model,
gains (from occupant, lighting and equipment) on energy corresponding to arriving, leaving and staying in the
consumption and cooling load of the building can be respective states, are time-dependent and estimated from
estimated, but it cannot represent the stochastic variations the probability of presence (based on aggregate occupancy
of occupancy in time and space. In addition, the datasets records) at every time step. In addition, an assumed para-
corresponding to occupancy schedule in all its stochastic meter of mobility describes the probability of state changes.
variety from survey and measurement are still scarce, and This model is capable of reproducing the important
even if such datasets were available, since occupancy may characteristic of occupancy in office buildings such as the
vary widely from one building to another depending on the morning arrival time, the night departure time, typical long
type and size of the building, etc., they may not accurately absences and the effective time of presence of the occupant
represent actual occupancy. within the zone by a uniform Markov chain. However, it
Consequently, stochastic models have been proposed to has two major disadvantages: (1) as inputs to the model,
produce synthetic occupancy schedules in a certain way as the profiles of probability of presence and parameters of
inputs to simulation tools. A fundamental stochastic model mobility are too complex to specify in simulation and to
uses the Monte Carlo method (Macdonald and Strachan obtain from survey or measurements because of their time
2001), which generates an estimation of occupancy for each dependency; (2) this model does not simulate the movement
zone in a building based on the probability distribution of of occupants from one zone to another, which is distinct in
occupancy in each zone (normal distribution used in reality and important for occupancy prediction. Even if
practice). It is an easy way to consider occupancy uncertainty, such efforts can be made, extending the model to multiple
regardless of the interactions and relationships of the zones is much more challenging for determining the time
occupancy at different times in a zone and that in different varying entries of high-order transition probability matrix
zones at the same time in a building, which are distinct in and parameters of mobility for state changes.
reality (e.g., an occupant is likely to stay in the same zone at The most recent model is proposed by Liao et al. (2011),
the next time if he stays there at previous time and he cannot an agent-based model of occupancy dynamics in a building.
appear at two places at the same time). Thus, this method is This model regards each occupant as an agent and decides
too rough for occupancy modeling and usually only used for the state of an agent (the location of each occupant) at every
risk analysis. time step through a set of rules specified by four modules,
In recent years, several advanced models have been which can then be collected to generate time-series of zone-
proposed to randomly generate plausible sequences of level occupancy. Compared to Page’s model, Liao’s model
occupancy in buildings. Wang et al. (2005) proposed a maintains a Markov-like property of agent dynamics and is
probabilistic model to predict and simulate occupancy in easily scalable to an arbitrary number of zones and an
single person offices, where non-homogeneous Poisson arbitrary number of occupants. The scheduled activities of
Wang et al. / Building Simulation / Vol. 4, No. 2 151

agents, the zones that each agent can access, the maximum which occur among the spaces inside and outside the building.
occupancy limits of zones, and a secondary agent that Thus there is no explicit or implicit constraint to the number
occupies the building for brief periods of time (like visitors) of occupants and the number of zones. By using the
are also involved. This model’s prediction accuracy was Markov chain method to simulate this stochastic movement
found to be quite good in the single-occupant single-zone process, the model can generate the location for each
scenario and multi-occupant single-zone scenario, but is occupant and the occupancy for each zone of the building
poor in the multi-occupant multi-zone scenario. The most while maintaining a simple, clear set of input parameters.
time-consuming part in constructing the model is specifying From the case study, it can be seen that the model can
the nominal presence probability profile for each agent in produce the realistic occupancy variations in the office
the preliminary state generator module, which consists of building for typical workday with key statistical properties
the probability time series that an agent occupies a zone at of occupancy such as the time of morning arrival and night
every time step. Due to excessive information, these inputs departure, lunch time, periods of intermediate walking-
for the model are hard to obtain especially for multi- around, etc. In the terms of simplicity, accuracy and
occupant multi-zone scenarios. For this reason, although it unrestraint, this model is sufficient and practical to
has no limitation in theory when applied to multiple zones, simulate occupancy for building energy simulations and
some simplifications have to be made in practice when stochastic analysis of building HVAC systems.
specifying the nominal presence profiles, such as the zones
that each occupant can access only include a primary zone, 2 Methodology
a secondary zone, a hallway, and a restroom. The frequency
and average duration of visiting a restroom per day are The goal of the model in this paper is to generate stochastic
assumed to be constant over the workday. Lunch and occupancy schedules with the same statistical characteristic
of building occupancy, which can then be used by a building
dinner breaks are strictly specified by fixed schedule rather
energy simulation tool to further estimate the energy
than random process. These simplifications reduce the
consumption of a building and the performance of a HVAC
capacity of the model and would lead to poor results of the
system.
model in a multi-occupant multi-zone scenario.
The basic idea of the model is that building occupancy
Additional models have tried to generate the stochastic
is a straightforward result of occupant movement processes
occupancy in buildings based on occupant activity simulation
which occur among the spaces inside and outside the building.
(Tabak 2008; Goldstein et al. 2010), where the models
The first-order homogeneous Markov chain technique, a
focused on the chronological sequence of detailed activities,
widely-used well-established stochastic process method (Ross
with attributes of a task (e.g., working, meeting, eating),
1996), was selected for simulating the occupant movement
such as frequency, duration, priority, location, facilities,
process.
interactions with one another, etc. This method is quite
The concept of the first-order homogeneous Markov
sophisticated and can output many details of human behavior
chain (HMC) technique is that any future state is dependent
in a building, from which the occupancy is a natural
only on the present state together with the probabilities of
consequence. However, relying on a huge extensive survey
the state changing (called Markovian property). These pro-
or measurement of such detailed activity, this indirect babilities are held in the transition probability matrix and
method is not as practical for occupancy simulation as the are time-independent (i.e., fixed). Consider a stochastic
aforementioned methods. process { X k , k = 0,1,2, } that takes a set of nonnegative
The above overview indicates that existing models have integers I = {0,1, 2,"} as possible values, HMC (with
demonstrated the capacity to realistically reproduce key discrete states and discrete time steps) can be present by
properties of occupant presence for both the single-occupant
single-zone scenario and the multi-occupant single-zone
P { X k +1 = j | X k = i, X k−1 = ik−1 , , X1 = i1 , X 0 = i0 }
scenario, but have serious limitations when extended to
multiple spaces. Furthermore, the excessive inputs of these = P { X k +1 = j | X k = i } = pij (k) = pij
models due to time dependency or high order of matrix are
complex to handle in simulation. for all states i0 , i1 , , ik−1 , i, j and all time steps k ≥ 0. The
The principal goal of this paper is to present a novel fixed value pij represents the probability that the process
approach for occupancy simulation based on homogeneous will, when in state i, next make a transition into state j. For
Markov chain and demonstrate it in an office building the transition probability matrix (denoted by P matrix in
simulation. In this study, occupancy is handled as the the sequel) that consists of one-step transition probabilities
straightforward result of occupant movement processes pij , we have that
152 Wang et al. / Building Simulation / Vol. 4, No. 2

pij ≥ 0, i, j ∈ I ; ∑p
j∈I
ij = 1, i ∈ I (1) possible values belong to the set (or a subset) of all spaces’
indices {0 = outside, 1 = zone 1, 2 = zone 2, , n = zone n}
that correspond to the occupant’s accessible range. The
HMC has many good statistical analysis features which movement process of each occupant can thus be described
greatly help set up the model. The interested reader can refer by a Markov chain in which the state of the process is
to Ross (1996) for more details on Markov chain. exactly the location of an occupant, and the next location of
the occupant is dependent only on the present location and
2.1 General description the fixed transition probabilities held in the P matrix.
Here the following assumptions have been made: (i) the
The proposed model has a two-level hierarchical structure location of occupant due to movement has a Markovian
consisting of a basic module named movement process and property; (ii) any location change of occupant due to move-
a high-level module named events as shown in Fig. 1. The ment can be finished in one time step; (iii) the movements
module of movement process essentially implements a of each occupant are independent, thus each occupant has
simulation of the Markov chain process and generates the his own transition probability matrix.
locations of occupants step by step, which can then be used The assumption (i) is supported by the experiments of
to calculate the occupancy for each zone in buildings. The Wang et al. (2005) and Dodier et al. (2006). They measured
module of events is used to specify the transition pro- the occupancy states for some single offices by assembling a
babilities of Markov chain in specific periods of time, in order sensor network. If the location of an occupant is described
to represent the occurrences associated with time. by a HMC, the sojourn time of the occupant in any state
2.1.1 Movement process (i.e., presence/in his office, absence/not in his office during
working hours) should be geometrically distributed. From
In this paper, the process of occupant movement covers all their results, it is approximately geometric for both presence
the occurrences that correspond to the location change of and absence durations of an occupant in business hours.
people within a building, such as entering or leaving a Figures 2 and 3 show the fitted and observed probability
specific space, moving around from one space to another, distribution of durations of presence and absence for an office
going outdoors for a while, etc. Such a stochastic process in Wang’s experiment, with a time interval of 15 minutes.
results in the variation of building occupancy in time and Figures 4 and 5 show the histograms of sojourn times and
space. Such processes also may vary widely from one building the fitted probability distribution of durations of presence
to another, depending on the type and size of the building, and absence for two offices in Dodier’s experiment, with
the geographic location and climate, the ethnicity and time intervals of 500 s and 200 s. Both authors concluded
preferences of the occupants, etc. that the presence and absence durations are exponentially
Consider a building with n zones that is occupied by m distributed. However, since the observed data in the
individuals, where a zone is an internal space in the building experiments are analyzed with a discrete time interval, the
and indexed as 1,2, ,n; an individual is denoted by occupant, exponential distributions are indeed geometric distributions,
i.e., residents in case of a residential building or office workers in a discrete form.
in case of an office building, etc. The outside of the building The two experiments, although based on single offices,
is also involved and treated as a specific space indexed by 0, in a way validate the assumption (i) in our model. It further
to form a complete movement graph. ensures that the stochastic occupancy model based on
Regarding the location of an occupant (in which space occupant movement is reliable in other situations, even for
the occupant is) at every time step as a random variable, its multi-occupant multi-zone scenarios.

Fig. 1 The schematic of the model

Wang et al. / Building Simulation / Vol. 4, No. 2 153

the location change of an occupant governed by P matrix

can be realized depends on his movement speed, especially
in a small time step simulation. The transition probabilities
of the occupant from the present state to the next state are
actually influenced by the limit of movement speed. For the
moment, it is not taken into account. Accordingly, to
ensure the temporal resolution of the results and to easily
integrate with building energy simulation tools, the time
intervals for the model can be 5, 10, 12, 15 minutes, or any
other submultiples of 60 minutes (i.e., 1 hour that is the usual
time step for building energy simulation).
With the above hypotheses, the movement patterns of
each occupant can be modeled individually in a simple way
even if they may be quite different, and the realistic Markovian
properties of occupant movement can be reproduced.
Fig. 2 Fitted and observed probability distribution of the presence
durations (Wang et al. 2005) 2.1.2 Event mechanism

It is noted that such a HMC can only produce a pure

random movement process for each occupant among the
spaces (inside and outside of the building), without
considering the time factors for the movement occurrences
that probably happen for occupants in certain periods of
time. Take an office building as an example, in normal
conditions the employees would usually go to the office in
the morning and leave in the evening, rather than random
arrival and departure at any time during the day. Such
occurrences of movement associated with time is common
for a building with specific function and an occupant with
specific career or post, such as in residential buildings, office
buildings, etc., which may be called typical movement
patterns for such types of buildings and occupants. An
event mechanism is proposed to represent and manage the
Fig. 3 Fitted and observed probability distribution of the absence time-triggered occurrences of movement in buildings, in
durations (Wang et al. 2005) which the movement of occupant is driven by a number of
events. In addition, the event mechanism can be also used
As for the hypothesis (ii), since the movement of an to treat the relevance of the movements of occupants (e.g.,
occupant is essentially a continuous-time process, whether a joint movement such as attending a meeting).

Fig. 4 (a) Frequency distribution of duration of presence, both offices aggregated (78 sojourns of presence, Dodier et al. 2006); (b) fitted
and observed probabilities of the presence durations
154 Wang et al. / Building Simulation / Vol. 4, No. 2

Fig. 5 (a) Frequency distribution of duration of absence, both offices aggregated (72 sojourns of absence, Dodier et al. 2006); (b) fitted
and observed probabilities of the absence durations

An event in the model is an object that corresponds to chain look like a HMC during every period split by events.
the specific location change of occupants; for example, the Our method is still labeled as HMC.
event of walking around, regarded as a basic event in all A set of events can be made in chronological order
buildings, corresponds to the location change from one space according to the typical movement patterns of building and
to another (covering general movements, such as going to occupant, and it is easily scalable to describe other events,
the washroom, dropping by another office, etc.). The event such as long absence, meeting, short visit, working at home,
of going to the office in the office building corresponds to working part-time, and even other scheduled events. Once the
an occupant’s location change from outside (space 0) to the list of events is made, the model is completely constructed
office (a space for his/her own office). Each event has a in form.
valid period (with a starting time and an ending time) 2.1.3 Algorithm
during which it takes place, and a range of actions on the
occupants (i.e., it only influences specific occupants). The Due to the simple structure, the implementation of the model
event drives occupant movement through P matrix exactly is relatively simple once the building typology, occupant
by specifying the corresponding elements in the P matrix. and movement parameters (e.g., the number of occupants,
The probability elements associated with the event are also the accessible spaces for each occupant, the set of events with
fixed and time-independent (called “event dependent”). properties, etc.) are determined.
Such transition probabilities can be determined from some The model was implemented as a MATLAB script. The
statistical indices of the event (see Section 2.2). Each event location of each occupant was simulated independently based
on the inputs related to that occupant. For initial states, the
also has a priority that determines the order of the event
occupant is considered to be absent in the case of office
taking effect on P matrix in case that several events are
buildings (or present in the case of residential buildings) at
valid at the same time (i.e., valid periods of events have
the initial time step, 00:00 of January 1st. From then on, the
intersections) and these events have common elements in P
time series of occupant location is generated by the transition
matrix. In this situation, such elements of P matrix would
probability matrix at each time step, which is specified by
be specified with the associated probabilities of the event with
valid events.
the highest priority. Figure 6 shows how the algorithm works: (0) initialize
In summary, an event object usually has six properties: the states of all occupants at time step 0; for each time step,
starting time, ending time, locations (from one space to (1) update the set of active events at present according to
another), participants (taking part in the event), transition the input set of events and their valid periods; (2) update
probabilities (driving the movement of participants), and the P matrices of all occupants according to the set of active
priority (to resolve conflicts). events, the corresponding elements of P matrix are specified
With the event mechanism, the Markov chain of by the active events, and note that the sum of elements in
occupant movement is indeed inhomogeneous since the each row of P matrix should equal 1; (3) for each occupant,
transition probabilities in P matrix could be changed at determine the current state of occupant according to the
certain times. However, the probabilities are fixed for the previous state and the updated P matrix (see Fig. 7 for details,
remaining periods, so most behaviors of such a Markov where the MATLAB function rand generates a pseudorandom
Wang et al. / Building Simulation / Vol. 4, No. 2 155

probabilities for each occupant. Since the probabilities in

the model are time-independent, it greatly reduces the
complexity in the time dimension compared to the pre-
existing models. Actually, all the entries of P matrix for each
occupant can be estimated directly from the information
that is collected by deploying sensors in the building which
track each individual over time as proposed by Tabak
(2008). However, it still seems not trivial in simulation to
directly input all the entries of P matrices of all occupants
in multi-occupant multi-zone scenarios as mentioned by
Liao et al. (2011). This is due to the high order of matrices
(corresponding to the number of spaces) and the number
of matrices (corresponding to the number of occupants).
Therefore, a way to further simplify the specifications for
such matrices is proposed in this paper.
The most important requirement for the simplified
method is the generated P matrix should capture the
specific statistical characteristics of building occupancy (i.e.,
occupant movement). An office building is used as an
Fig. 6 The workflow of algorithm (processing stage) example to demonstrate the procedure how to apply the
present occupancy model to a specific type of building and
how to simplify the inputs based on the key statistical
characteristic of occupant movement.

2.2 Occupancy modeling in an office building

An office building is a common building type which most

occupancy modeling research is focused on. In an office
building, a typical movement pattern for the occupants in a
workday is going to the office in the morning, having a lunch
break at noon, and getting off work in the evening or night
(possibly with overtime), which leads to the phased variation
of the number of occupants of a typical office building over
the workday as shown in Fig. 8. During the working periods,
the occupancy for each zone of the building would change
since the occupants walk around among the spaces inside
and outside of building for a variety of reasons.
In addition to the phased evolution, the occupants’
comings and goings for the morning arrival (t1 to t2), the
Fig. 7 The simulation script of Markov chain at each time step

value drawn from the standard uniform distribution), and

do this for all occupants; (4) calculate the current occupancy
for all zones according to the locations of all occupants.
By repeating this procedure step by step, the time series
of the location of each occupant and the occupancy of each
zone in the building can be generated, without any constraint
on the number of occupants and the number of zones.

2.1.4 Reduction and estimation of transition probabilities

The kernel parameters for the model are the transition Fig. 8 The occupancy variation of an office building over the workday
156 Wang et al. / Building Simulation / Vol. 4, No. 2

lunch break (t3 to t4), and the night departure (t5 to t6) are In our situations, this Markov chain is irreducible and
all random rather than deterministic (or strictly according ergodic. Such a Markov chain has a stationary distribution,
to a schedule), which greatly affects the working time of denoted by π , π = ( π 0 , π1 , , πn ), where πi means the
building devices. The indices representing the morning long-run proportion of time that the Markov chain is in
arrival and the night departure are respectively the time of state i (i.e., with what proportion of time the occupant stays
morning arrival and the time of night departure. Those for a in the space i). And we have (Ross 1996):
lunch break are the time of going out for lunch and the time
n
of coming back to office (it is assumed that the occupants
do not have lunch in their own office).
∑π
i =0
i =1 (3)

Such a typical movement process of the occupant in an

office building can be described by a set of five events: π = πP (4)
walking around, going to the office, going for lunch, coming Based on Eqs. (3) and (4), given the matrix P, the vector
back from lunch, and getting off work. Next, an approach to π can be determined by solving Eqs. (3) and (4) simul-
model such events with a specific statistical characteristic taneously, and expressed as Eq. (5):
and the treatment for events that are not detailed in this
paper are discussed. ⎛ 1 1 … 1 ⎞ ⎛ 1 ⎞⎟
⎜⎜ ⎟⎟ ⎜⎜ ⎟
⎜⎜ p01 p11 … pn1 ⎟⎟⎟ ⎜ 0 ⎟⎟
2.2.1 Walking around A = ⎜⎜ ⎟⎟ b = ⎜⎜⎜ ⎟⎟⎟ π T = A−1b (5)
⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟
⎜⎜ ⎟⎟ ⎜ ⎟⎟
The event of walking around in office buildings corresponds ⎝ p0n p1n … pnn ⎠ ⎟ ⎜⎝ 0 ⎠⎟
to the general location change from one space to another,
among the spaces inside and outside of building (e.g., going where π T is the transpose of π .
to the washroom, walking in the hallway, going outside,
dropping by another office, etc.). In most situations, it The expected sojourn time
means a transient movement of an occupant out of his office. The sojourn time in state i of Markov chain, denoted by STi ,
Its valid period is usually the same as the business time of a is geometrically distributed as expressed in Eq. (6). (Instead,
company, e.g., from 8:00 to 17:00. it is exponentially distributed in the continuous-time Markov
As discussed above, specifying the P matrix for walking chain.)
around during the working period is the most challenging
work for multi-zone scenarios. Since the present stochastic
P { STi = k } = pii k−1 ( 1 − pii ) (6)
model should reproduce the specific statistical characteristics
of occupant movement, a simplified way can be proposed
to determine the probabilities. where k is the number of time steps during which the Markov
From experience, the key statistical characteristic for an chain is out of state i (measured from the time when the
occupant walking around is the long-run proportion of Markov chain is in state i).
time and the expected sojourn time that the occupant stays The expected sojourn time can be expressed by (Sheng
in each space of the building during the working period. et al. 2008):
The simplified way to specify P matrix is based on the ∞ ∞
1
following mathematical formulations. E ( STi ) = ∑ k ⋅ P { STi = k } = ∑ k ⋅ pii k−1 ( 1 − pii ) =
k =1 k =1 1 − pii
The stationary distribution (7)
Let P denote the transition probability matrix that corresponds So that,
to the process of an occupant moving among the spaces (all
zones in the building and outside), which are indexed by 1
pii = 1 − (8)
{0 = outside, 1 = zone 1, 2 = zone 2, , n = zone n}. Suppose E ( STi )
the probabilities are fixed and not influenced by any other
events in normal conditions. Denote the vector of expected sojourn time by Est,
Est = ( Est 0 , Est1 , , Est n ), where Est i means the expected
⎛ p00 p01 … p0n ⎞ sojourn time that the Markov chain is in state i (i.e. how
⎜⎜ ⎟⎟ long the occupant stays in the space i at a time).
⎜⎜ p10 p11 … p1n ⎟⎟⎟
P = ( pij )(n+1)×(n+1) = ⎜⎜ ⎟⎟⎟ (2) Equations (5) and (8) illustrate the relations between
⎜⎜ ⎟⎟
⎜⎜ the P matrix and the long-run proportion of time and the
⎝ pn0 pn1 … pnn ⎠⎟⎟ expected sojourn time.
Wang et al. / Building Simulation / Vol. 4, No. 2 157

Given the long-run proportion of time π and the expected 0 1

sojourn time Est of the occupant in every space, the simplified Pgo_office =0 ⎡ p00 p01 ⎤ (10)
way to specify P matrix can be described as an optimization ⎢ ⎥
1 ⎢⎣ 0 1 ⎥⎦
problem:
where 0, 1 are respectively the indices of outside and the
min ||( A b) − π ||2
−1 T occupant’s office that is known as an absorbing state. The
absorbing state means the occupant will definitely enter his
s.t. pij ≥ 0, ∑p j
ij =1 (9)
office at a certain time, and it means the arrival time of the
1 occupant entering his office when the HMC is in the
pii = 1 − absorbing state. During the valid period of the event “going
Est i
to office”, the elements of the occupant’s P matrix
where π = ( A−1b)T is an estimation of π . This optimization corresponding to outside and his office would be specified
with the probabilities in Eq. (10).
problem can be solved by using the fmincond function in
Measured from the start of the event of going to the
MATLAB.
office, the morning arrival time of the occupant is
Thus, the P matrix with (n + 1) × (n + 1) entries can be
geometrically distributed, and the expected arrival time
specified by using a set of 2 × (n + 1) parameters, which
(denoted by E(FA)) can be expressed by
greatly reduces the space complexity of such inputs. Naturally,
the Markov chain governed by this P matrix will have the 1 1
same statistical characteristic of occupant movement that is E(FA) = p00 = 1 − (11)
1 − p00 E(FA)
defined by the long-run proportions of time and the expected
sojourn times of the occupant in every space. These two vectors If the arrival time is the same as the on-duty time,
of an occupant are then used as the input of the present p00 = 0 and E(FA) = 1.
model, denoted by “movement vectors” in the sequel. Similarly, the event of getting off work (i.e., night
By using the movement vectors, the input information departure) of the occupant corresponds to the location
for the model is easier to collect by deploying tracking change from his own office to outside the building. Its valid
sensors or conducting a questionnaire survey of the period is usually a time span after the business hours, e.g.,
occupants’ behavior. The price of such simplification is that 17:00 − 21:00, corresponding to the earliest and latest night
the produced P matrix might lose some inherent information departure time of office workers.
that relates the movement of occupant in the building com- The night departure process can be expressed by a two-
pared to a directly specified P matrix, since every element state HMC governed by the fixed 2-by-2 P matrix in Eq. (12).
of the P matrix has its own meaning. Whether to choose
the simplified method or specify the P matrix depends on the 0 1
user’s demand. In the most situations of building occupancy Poff_work =0 ⎡ 1 0⎤ (12)
⎢ ⎥
simulation, the detailed specification of P matrix is time- 1 ⎢⎣ p10 p11 ⎥⎦
consuming and not necessary; instead, the movement vectors
of occupant are simple, effective, and accurate enough for where 0, 1 are respectively the indices of outside and the
simulation. occupant’s office. Here, outside the building is the absorbing
state, which means the occupant will definitely leave his
2.2.2 Going to the office and getting off work office at a certain time. And it means the departure time
of the occupant leaving his office when the HMC is in the
The event of going to the office (i.e., morning arrival) of an
absorbing state.
occupant is regarded as the location change from outside
Measured from the start of the event of getting off work,
the building to his own office. Thus it only relates the elements
the night departure time of the occupant is also geometrically
of the occupant’s transition matrix where the row and the
distributed, and the expected departure time (denoted by
column correspond to outside and his office. The valid period
E(LD )) can be expressed by
of this event is usually a time span before the business
hours, e.g. from 7:00 to 8:30, corresponding to the earliest 1 1
and latest morning arrival time of office workers. E(LD) = p11 = 1 − (13)
1 − p11 E(LD)
The morning arrival process can be expressed by a
two-state HMC with an absorbing state, which is governed If the departure time is the same as the off-duty time,
by a fixed 2-by-2 P matrix in Eq. (10). p11 = 0 and E(LD) = 1.
158 Wang et al. / Building Simulation / Vol. 4, No. 2

2.2.3 Lunch break Fig. 9). The interactions of occupants, such as going for
lunch together, can also be defined in the form of events.
The lunch break of an occupant can be divided into two
As for the maximum occupancy limits of spaces, a simple
events. One is the event of going for lunch, corresponding
rule can be added into the determination of occupant location
to the start of lunch break; the other is the event of coming
at each time step, that is, if the space that the occupant
back from lunch, corresponding to the end of lunch break.
would go to has reached its maximum occupancy limit, the
We suppose the location for the occupant’s lunch is out of
occupant would return to his last location rather than enter
his office. The two events can be treated similarly like the
the zone, i.e., the occupant’s location at the current time
event of going to office and getting off work.
step would be the same as the previous time step.
The process of going for lunch can be expressed by the
transition matrix in Eq. (14). Measured with the start of 2.2.5 Summary
the event of going for lunch (i.e., the earliest leaving time),
the expected leaving time for lunch (denoted by (E(LL)) is With the events modeled above, a typical movement pattern
expressed by Eq. (15). of occupants over the workday in an office building, i.e.,
“morning arrival—walking around (working period)— lunch
0 1 break—walking around (working period)—night departure”,
Plunch_out =0 ⎡ 1 0⎤ (14) can be simulated by a unique Markov chain, in which the
⎢ ⎥
1 ⎢⎣ p10 p11 ⎥⎦ transition matrix for each occupant is successively specified
with the associated probabilities of events during different
1 1 periods of the day. Such probabilities for morning arrival,
E(LL) = p11 = 1 − (15)
1 − p11 E(LL) walking around, lunch break, night departure can be
determined based on the statistical indices of each event.
The process of coming back from lunch can be expressed The simulated results of such a Markov chain will have the
by the transition matrix in Eq. (16). Measured with the same statistical characteristic of events.
start of the event of coming back for lunch (i.e., the earliest Besides the basic information of building and occupant
return time), the expected return time from lunch (denoted (building typology, occupant number, working schedule,
by E(LR)) is expressed by Eq. (17). etc.), the particular inputs for the present occupancy model
0 1 of an office building are clarified as follows.
(1) For walking around, the vectors of long-run pro-
Plunch_back = 0 ⎡ p00 p01 ⎤ (16)
⎢ ⎥ portion of time and expected sojourn time for each space
1 ⎢⎣ 0 1 ⎥⎦ are needed.
(2) For morning arrival, the earliest, the latest and the
1 1
E(LR) = p00 = 1 − (17) expected arrival time are needed. For night departure, the
1 − p00 E(LR) earliest, the latest and the expected departure time are
needed.
2.2.4 Other issues

There are many other things in an office building that have

not been discussed in detail in the above paragraphs, such
as meetings that happen during a workday, long period
absence or leaves or working at home that result in the
occupants not entering the office for a whole workday, short
visits or working part-time that the visitors or occupants
only appear in the building for a short period of workday,
tea breaks for some organizations, or the situation that the
occupant has his lunch in his office, etc.
The reasons that those situations are not modeled are,
on one hand, this paper focuses on the modeling of most
typical movement patterns in an office building, it’s not
necessary to involve too many events; on the other hand,
on the basis of our present model, such things can be modeled
easily with an external event generator (random or scheduled),
whose outputs can be used as the inputs for the model (see Fig. 9 The integration of event generators within the present model
Wang et al. / Building Simulation / Vol. 4, No. 2 159

(3) For lunch break, the earliest, the latest and the 3.1 Input settings
expected leaving time for going for lunch, the earliest, the
latest and the expected return time for coming back from 3.1.1 Building typology
lunch are needed.
The 2D plan of the office building is shown in Fig. 10. There
In addition, due to different personalities of occupants,
are 4 office rooms, 1 corridor, and 1 restroom, indexed from
the associated events (or event properties) for each occupant 1 to 6; and the outside is indexed by 0. There are 7 spaces in
can be different and thus need to be specified individually. total. All spaces are connected by doors and corridor.
All the above information needs to be collected and
calibrated before applying the model. 3.1.2 Occupant and movement parameters

There are 3 types of occupants in the building. The occupants

2.3 Discussion on model calibration and validation in offices 1 and 2 are ordinary workers. The occupants in
office 3 are administrative staff (secretary). The occupant in
The validation of the proposed stochastic occupancy model,
office 4 is a manager (head of the organization). They move
theoretically speaking, should be based on probability, which
among the 7 spaces (i.e., all spaces are accessible for each
means the comparison of the probability distribution function
occupant). The number of occupants for each office is
(PDF) of the measured and simulated parameters. This
shown in Table 1.
requires a long period (usually several years) of measured
A fixed working schedule from 8:00 to17:00 is specified.
occupancy data in the real office buildings. But such data Five typical types of events and a scheduled meeting are
are too scarce and not available for most buildings. So the considered in the workday. The schedule and events in the
theoretical probabilistic test cannot be carried out due to workday is shown in Table 2.
too few samples. Thus, from the practical point of view, a The exceeding periods of going to office and returning
simple test approach needs to be proposed. For example, from lunch, compared to the standard schedule, i.e., 7:00 −
the maximum, minimum and average occupancy for each 8:00, 8:00 − 8:30 and 13:00 − 13:30 respectively means the
zone inside the building could be chosen as the test parameters early morning arrival, the late morning arrival and the late
and the criterions to calibrate the model. This will be an back to the office from lunch. The exceeding period of getting
important work and needs more study in the future. off work, 17:00 − 21:00, means the delay or overtime for
For the moment, due to the lack of measured data, the night departure.
proposed model has not been fully validated by comparing The events (except meeting) work for all occupants.
the simulated and the measured data. Nonetheless, an The expected values for the events are shown in Table 3. Note
illustrative case study can be made to check and test the that these values are accounted in the unit of 5-minute time
capacity of the model; whether it can cover the things that step. They are only estimations for the statistical index of each
affect the occupancy variation in a building and how much event, and used to determine the transition probabilities in
it can capture this. P matrix and the variations of occupancy ultimately.

3 Case study

A simple office building is tested to demonstrate the usage

and effect of the proposed model. This case is illustrative
and the input data are taken from experience. The time step
used in the case is 5 min; an occupancy time series of one
day is comprised of 288 points.
Through the case, we will check the capacity of the
model to represent: (1) the trend of “going to the office—
working—lunch break—working—getting off work” in a
typical workday; (2) the random arrival time and the Fig. 10 2D plan of the office building
smooth increase of building occupancy in the morning;
(3) the random departure time (overtime) and the smooth Table 1 Number of occupants in each office
decrease of building occupancy; (4) the random decrease of Number of Number of
Room No. occupants Room No. occupants
building occupancy during the lunch break; (5) random
movement among the inside and outside spaces during Office 1 6 Office 2 6
working time. Office 3 2 Office 4 1
160 Wang et al. / Building Simulation / Vol. 4, No. 2

Table 2 Schedule and events in a working day 3.2 Transition matrix

Schedule Event Valid period
Go to the office 07:00 − 08:30
The transition matrix for each occupant can be determined
by the movement parameters according to the equations in
Leave for lunch 12:00 − 12:30
Working time Section 2.2. The associated probabilities of events for
Return from lunch 12:30 − 13:30
8:00–17:00 occupants in office 1 are shown in Fig. 11. Based on those
Get off work 17:00 − 21:00
Lunch break probabilities, the transition matrix would be specified and
12:00–13:00 Walk around 08:00 − 12:00, changed at certain time steps when the events become valid
13:00 − 17:00
over a workday.
Meeting 10:00 − 11:30

3.3 Results and discussions

The expected value of morning arrival time for going to
the office is 7:45; which means, in the 5 min unit the geometric The simulation for a workday runs 1000 times consecutively,
distribution of morning arrival time has an expected value with different random seeds for each simulation. Figure 12
of 9 time steps (measured from 7:00). shows the generated time series of the locations of four
The values of long-run proportion and mean sojourn occupants in offices 1, 2, 3 and 4. As expected, the four
time for different spaces mean: for the occupants in office 1, occupants stay in their own offices for most of the time,
each of them will spend 90% of business hours in his own with different times of morning arrival and night departure,
office, 3% in the other three offices, 1% in the outside, 1% different times of leaving and returning during lunch break,
in the corridor, and 5% in the restroom; the mean sojourn and occasionally out of the office (into other spaces). The
time steps for each space are 24, 3, 2, 2, 2. scheduled meeting can be seen from that the occupant in
With different movement vectors, the differences between office 1 and the manager in office 4 stay in the office 4
the three types of occupants are specified. A manager may during the period of 10:00 − 11:30. The secretary in office 3
have more meetings out of the office, so he/she only spends seems to move more frequently than others.
60% of business hours in the office and 30% outside, with
3.3.1 Building occupancy
1-hour mean sojourn in both spaces. A secretary may tend
to move more frequently in the spaces, so he/she only spends The change of number of occupants in the whole building
70% of business hours in the office and the remaining 30% over a workday is illustrated in Fig. 13, where (a) shows the
occupying other spaces, with 1-hour mean sojourn in the result of building occupancy for one simulation, (b) shows
office and 10 min for each other space. the results for five successive simulations, (c) shows the
To illustrate the effect of meeting events, a scheduled maximum, minimum, and average occupancy for each
meeting is held in the manager’s office from 10:00 to 11:30, time step from all the simulations, (d) shows the hourly
and it is attended by two occupants from office 1, two occupancy taking the mean of five-minute results. It can be
occupants from office 2 and the manager. All the participants seen that (1) the trend of “going to the office − working −
are determinate. lunch break − working − going off work” in a typical workday

Table 3 Expected values for each occupant

Event Statistical index Expected value
Go to the office Morning arrival time 7:45
Go for lunch Leaving time 12:10
Come back from
Return time 12:50
lunch
Get off work Night departure time 18:00
π = [0.01, 0.9, 0.01, 0.01, 0.01, 0.01, 0.05]
Office 1
Est = [10min, 2h, 15min, 15min, 15min, 10min, 10min]
π = [0.01, 0.01, 0.9, 0.01, 0.01, 0.01, 0.05]
Office 2
Long-run proportion of time and Est = [10min, 15min, 2h,15min, 15min, 10min, 10min]
Walk around
mean sojourn time in each room π = [0.05, 0.05, 0.05, 0.7, 0.05, 0.05, 0.05]
Office 3
Est = [10min, 10min, 10min, 1h, 10min, 10min, 10min]
π = [0.3, 0.01, 0.01, 0.01, 0.6, 0.02, 0.05]
Office 4
Est = [1h, 5min, 5min, 5min, 1h, 10min, 10min]
Wang et al. / Building Simulation / Vol. 4, No. 2 161

0 1 2 3 4 5 6 0 1 2 3 4 5 6
0 ⎡1 0 0 0 0 0 0⎤ 0 ⎡ 0.5000 0.3724 0.0247 0.0247 0.0250 0.0272 ⎤
0.0260
⎢ ⎥ ⎢ ⎥
1 ⎢1 0 0 0 0 0 0 ⎥⎥ 1 ⎢ 0.0042 0.9583 0.0023 0.0023 0.0023 0.0263 ⎥⎥
0.0042
⎢ ⎢
2 ⎢⎢ 1 0 0 0 0 0 0 ⎥⎥ 2 ⎢ 0.0244 0.2120 0.6667 0.0234 0.0237 0.0255 ⎥⎥
0.0243
P= , P= ⎢
3 ⎢⎢ 1 0 0 0 0 0 0 ⎥⎥ 3 ⎢ 0.0242 0.0253 ⎥⎥
⎢ 0.2131 0.0233 0.6667 0.0235 0.0241
4 ⎢⎢ 1 0 0 0 0 0 0⎥
⎥ 4 ⎢
⎢ 0.0247 0.2110 0.0237 0.0236 0.6667 0.0258 ⎥
0.0246
⎥
⎢ ⎥ ⎢ ⎥
5 ⎢1 0 0 0 0 0 0⎥ 5 ⎢ 0.0257 0.3747 0.0242 0.0243 0.0245 0.0267 ⎥
0.5000
⎢ ⎥ ⎢ ⎥
6 ⎢⎣ 1 0 0 0 0 0 0 ⎥⎦ 6 ⎢ 0.0055 0.4734 0.0052 0.0052 0.0053 0.0055 0.5000 ⎥⎦
⎣
Night time (initial state) Walking around

0 1 0 1
Pgo_office = 0 ⎡ 0.8889 0.1111 ⎤ Poff_work = 0 ⎡ 1.0000 0 ⎤
⎢ ⎥ , ⎢ ⎥
1 ⎢⎣ 0 1.0000 ⎥⎦ 1⎢⎣ 0.0833 0.9167 ⎥⎦
Going to the office Going off work

0 1 0 1
Plunch_out = 0 ⎡ 1.0000 0 ⎤ Plunch_back = 0 ⎡ 0.7368 0.2632 ⎤
⎢ ⎥, ⎢ ⎥
1 ⎢⎣ 0.5000 0.5000 ⎥⎦ 1 ⎢⎣ 0 1.0000 ⎥⎦
Leaving for lunch Returning from lunch

Fig. 11 The transition matrices of occupant in office 1

Fig. 12 The time series of occupants’ locations over a workday

162 Wang et al. / Building Simulation / Vol. 4, No. 2

Fig. 13 Building occupancy over a workday

is reproduced; (2) the total building occupancy reaches of occupants from one space to another, the occupancy of
maximum gradually, rather than sharply under a fixed zones inside the building stochastically change in time,
schedule; (3) during the working time, the total occupancy more or less than the design value. Although the absence
varies due to the movement of occupants. Such a workday probability for each occupant out of his office (i.e., the
process changes at every simulation, which is understood probability of staying in other spaces) is as small as 0.1, the
as random in everyday life. resulted occupancy variation is remarkable and should not
be neglected in building energy simulation. Due to the
3.3.2 Zone occupancy
scheduled meeting, the occupancy in offices 1, 2, and 4
Figures 14, 15, 16, 17, 18 and 19 show the changes of change a lot during the meeting period.
number of occupants in office 1, office 2, office 3, office 4, Since the occupancy model is based on the process of
corridor 5, and restroom 6 over a workday, where (a) is the occupant movement, the relationships of stochastic occupancy
result of zone occupancy for one simulation, (b) illustrates in multiple zones are realistically taken into account. This
the maximum, minimum, and average occupancy for each results in the reasonable occupancy distribution in space,
time step from all the simulations, (c) shows the hourly that is, an increase of occupancy in one space usually
occupancy results. It can be seen that due to the movement means a decrease of occupancy in another space while the
Wang et al. / Building Simulation / Vol. 4, No. 2 163

Fig. 14 Office 1’s occupancy over a workday Fig. 15 Office 2’s occupancy over a workday
164 Wang et al. / Building Simulation / Vol. 4, No. 2

Fig. 16 Office 3’s occupancy over a workday Fig. 17 Office 4’s occupancy over a workday
Wang et al. / Building Simulation / Vol. 4, No. 2 165

Fig. 18 Corridor 5’s occupancy over a workday Fig. 19 Restroom 6’s occupancy over a workday
166 Wang et al. / Building Simulation / Vol. 4, No. 2

total occupancy stays the same. From the simulation 4 Conclusions

results, the occupancy of corridor and restroom are strong
coupled with the occupancy of offices and their variations Building occupancy is a key factor to accurately predict
are more significant in time. Even for the spaces with building energy consumption and evaluate the energy saving
similar function type, the transient occupancy schedules from potential of occupancy-based control system and the perfor-
office 1 to office 4 are not synchronous at each time step mance of HVAC systems. However, it is hard to represent
(see Fig. 20). Such an uneven distribution of occupancy in due to its temporal and spatial stochastic nature.
space and nonsynchronous change in time would affect the This paper presents a novel approach for occupancy
performance evaluation of HVAC systems in simulation. simulation based on the homogeneous Markov chain. In this
In general, the stochastic occupancy over a typical study, occupancy is handled as the straightforward result of
workday in an office building can be realistically produced occupant movements among the inside and outside spaces of
by using the proposed model. Further analysis of simulation a building. By using the Markov chain method to simulate
results can be made for exploring the validation approach the stochastic movement process, the model can generate
of the model in the future. the location for each occupant and the occupancy for each

Fig. 20 Four offices’ occupancy over a workday

Wang et al. / Building Simulation / Vol. 4, No. 2 167

zone of the building. Through this approach, the Markovian References

property of the state of occupant location is retained, which
is validated by other experiments, and the relationships of Crawley DB, Lawrie LK, Winkelmann FC, Buhl WF, Huang YJ,
stochastic occupancy in multiple spaces are realistically taken Pedersen CO, Strand RK, Liesen RJ, Fisher DE, Witte MJ, Glazer
into account. By using the event mechanism, this model is J (2001). EnergyPlus: Creating a new-generation building energy
capable of covering most things that affect the occupancy simulation program. Energy and Buildings, 33: 319 − 331.
variation in a building and capturing the movement Davis JA, Nutter DW (2010). Occupancy diversity factors for common
differences of different types of occupants as well. university building types. Energy and Buildings, 42: 1543 − 1551.
From the case study of an office building, it can be seen DeST (2008). DeST-c User Manual. DeST Group, Tsinghua University,
China.
that the model can produce the realistic occupancy variations
Dodier RH, Henze GP, Tiller DK, Guo X (2006). Building occupancy
in the office building for a typical workday with key statistical
detection through sensor belief networks. Energy and Buildings,
properties of occupancy such as the time of morning arrival
38: 1033 − 1043.
and night departure, lunch time, periods of intermediate
EnergyPlus (2009). Input/Ouput Rerference, Version 4.0 Documentation.
walk-around, etc. Especially, it can produce the nonsyn-
University of Illinois and Ernest Orlando Lawrence Berkeley
chronous change of occupancy in time and the uneven
National Laboratory, USA.
distribution of occupancy in space, which can distinctly
ESRU (1999). ESP-r: A building and plant energy simulation environment,
affect the performance evaluation of HVAC systems in a User Guide Version 9 Series. ESRU Publication, University of
simulation. Strathclyde, Glasgow.
The model is simple, clear and has no explicit or Goldstein R, Tessier A, Khan A (2010). Schedule-calibrated occupant
implicit constraint with the number of occupants and the behavior simulation. In: Proceedings of the Symposium on
number of zones. On the strict mathematical basis of Simulation for Architecture and Urban, Orlande, USA.
geometric distribution, the model builds the relations of Klein SA, Beckman WA, Mitchell JW, Duffie JA, Duffie NA, Freeman
the statistical indices of building occupancy such as mean TL, Mitchell JC, Braun JE, Evans BL, Kummer JP, Urban RE,
time of morning arrival and night departure, long-run Fiksel A, Thornton JW, Blair NJ, Williams PM, Bradley DE,
proportion of time and expected sojourn time. Thus it McDowell TP, Kummert M (2004). TRNSYS 16—A transient
overcomes the issue of specifying the transition matrix for system simulation program, User manual. Solar Energy Laboratory,
multi-zone scenarios and maintains a simple, clear set of University of Wisconsin-Madison.
input parameters. In terms of simplicity, accuracy and Liao C, Lin Y, Barooah P (2011). Agent-based and graphical modelling
unrestraint, this model is sufficient and practical to simulate of building occupancy. Journal of Building Performance Simulation,
occupancy for building energy simulations and stochastic doi: 10.1080/19401493.2010.531143.
analysis of building HVAC systems. Macdonald I, Strachan P (2001). Practical application of uncertainty
The occupancy model’s assumption that the location of analysis. Energy and Buildings, 33: 219 − 227.
an occupant due to movement has a Markovian property is Page J, Robinson D, Morel N, Scartezzini JL (2008). A generalized
supported by some experiments for single offices but more stochastic model for the simulation of occupant presence. Energy
validations need to be carried out in the future. More and Buildings, 40, 83 − 98.
Ross SM (1996). Stochastic Processes, 2nd Edn. John Wiley & Sons,
events such as short visits should be taken into account in
Inc.
office building to capture the stochastic occupancy variations.
Sheng Z, Xie SQ, Pan CY (2008). Probability and Mathematical
From a practical point of view, a simple validation and
Statistics, 4th Edn. Beijing: Higher Education Press. (in Chinese)
calibration approach needs to be proposed. The capability
Tabak V (2008). User simulation of space utilisation: system for office
of the model in other types of buildings, such as residential
building usage simulation. PhD Thesis, Eindhoven University of
buildings, should also be tested and calibrated with specific
Technology, Netherlands.
occupant movement patterns. Wang D, Federspiel CC, Rubinstein F (2005). Modeling occupancy in
single person offices. Energy and Buildings, 37: 121 − 126
Acknowledgements Yan D (2005). HVAC system simulation and analysis based on uncertain
inner heat gains. PhD Thesis, Tsinghua University, China.
This project is financially supported by the National Yan D, Xia J, Tang W, Song F, Zhang X, Jiang Y (2008). DeST—An
Natural Science Foundation of China (No. 51008176), the integrated building simulation toolkit, Part I: Fundamentals.
Specialized Research Fund of the Doctoral Program of Building Simulation, 1: 95 − 110.
Higher Education of China (No. 20100002120014), and Zhang X, Xia J, Jiang Z, Huang J, Qin R, Zhang Y, Liu Y, Jiang Y
Tsinghua University Initiative Scientific Research Program (2008). DeST—An integrated building simulation toolkit, Part II:
(No. 2009THZ0). Applications. Building Simulation, 1: 193 − 209
BUILD SIMUL (2012) 5: 85 – 94
DOI 10.1007/s12273-011-0049-0

Quantitative description and simulation of human behavior in

residential buildings

Research Article
Chen Peng, Da Yan (), Ruhong Wu, Chuang Wang, Xin Zhou, Yi Jiang

Department of Building Science, School of Architecture, Tsinghua University, Beijing 100084, China

Abstract Keywords
An in-depth understanding of building energy use requires a thorough understanding of human human behavior,
behavior. This research gives a quantitative description of human behavior in residential buildings. behavioral object,
This quantitative description method can be used to forecast the impact of the human behavior energy use,
on the indoor building environment and energy use. Human behavior influences the energy use lifestyle model,
directly and indirectly by changing window openings, air-conditioner usage, lighting, etc. This time related,
quantitative description method describes these behavioral effects. Behavior can be divided into environmentally related,
random
several types according to the usage with time related, environmentally related and random modes
used to quantitatively describe the behavior. The method is then applied to describe a Beijing
Article History
household with comparison to on-site observations of the resident’s behavior and measurements
Received: 15 April 2011
of energy use to validate the method. The results show that the human behavior in the real world
Revised: 6 August 2011
can be quantified by the quantitative description method. These simulation tools can greatly
Accepted: 6 September 2011
facilitate building energy conservation by describing the influence of human behavior on building
performance and energy use.
© Tsinghua University Press and
Springer-Verlag Berlin Heidelberg
2011

1 Introduction and the local climate affected both household lifestyles and
energy use. Technical progress and social change also
Human behavior greatly influences building energy use by influence human behavior (van Raaij and Verhallen 1983;
the way people move, use equipment, open or close windows, Hitchcock 1993). Further research established that usage
building system (e.g., air-conditioning systems and lighting times of some appliances, such as washing machines and
systems) control by the property management personnel bathroom heaters, change with the seasons. The average

Building Thermal, Lighting,

and automatic controls. These actions directly and indirectly number of household members, their age, and their time

and Acoustics Modeling

affect energy use (e.g., electricity, gas, and water). Li et al. spent at home are all factors closely related to energy use in
(2006) reported that there are significant differences in energy residential buildings (Ouyang et al. 2007). An IEEA study
use among different apartments in the same residential showed that the average operating time of appliances in
building in Beijing. The human behavior such as air- Europe varied in different countries, which is useful for
conditioning operating times, temperature set points, and energy use calculations, but does not describe the human
window opening/closing habits are quite different with these behavior influence on energy use (IEEA 2008). There have
behavioral differences causing large differences in the electricity also been studies of the interactions between human behavior
use by cooling systems in the same residential building, from and building energy use (Mahdavi 2007). The influence of
0 to 14.3 kWh/m2 with average of 2.3 kWh/m2. human behavior on building energy use is now a hot issue at
Scientists began studying the relationship between human the Annex 53(2010).
behavior and energy use around 1980s. Sociological models As computer simulations have improved, these tools
of human behavior were developed, showing that the social- are being used more to analyze human behavior. Recent
cultural environment, the building and design requirements research (Hoes et al. 2009) has shown that human behavior
E-mail: yanda@[Link]
86 Peng et al. / Building Simulation / Vol. 5, No. 2

can be divided into groups based on the complexity and is mostly described by schedule definition, which does not
desired resolution. The information used for building models reflect the actual human behavior complexity. The difference
includes averaged, minimum and maximum values of various between real test data and simulation results is due to human
parameters. Models considering interactions between energy behavior in actual complex situations. Human behavior
use and the environment are more complex. Bourgeois includes system operation and management, occupant
(2005) did a comprehensive study of human behavior in behavior, and indoor environment conditions. These factors
buildings and focused on behavior prediction and advanced work together to influence the building energy use as shown
behavioral modelling. Korjenic and Bednar (2011) found in Fig. 1.
that the lifestyle and occupants’ living standard are the A general relationship between human behavior and
mainly reason for the discrepancy between the calculated building energy use is needed to quantitatively analyze the
and measured energy use from the investigation. influence of human behavior. The relationships between
Some researchers have used simulations to model certain buildings and humans are based on three key points as shown
kinds of behavior in buildings, such as a window opening in Fig. 2 which was developed to describe human behavior
and appliance usage (Rijal et al. 2007; O’Doherty et al. 2008; under study.
Nicol 2001; Reinhart 2004). Yun and Steemers (2008) (1) Ideology to behavioral principles;
developed stochastic models to predict window-opening (2) Human physiological, psychological and economics
behavior patterns as a function of the indoor temperature, feelings;
the time of a day and the previous window state through the (3) Impact of human behavior on building energy use.
field study in offices in UK. Different ways are used to classify human behavior
Existing building energy simulation software packages depending on the field. This study focuses on the impact of
pay more attention to the effects of climate, building envelope, human behavior on building energy use by quantitatively
systems and equipment than the occupant’s actions (Crawley analyzing the influence of human behavior. A human behavior
et al. 2001; EnergyPlus 2009; ESRU 1999; Klein et al. 2004; model is constructed for a building simulation software
Yan et al. 2008; Zhang et al. 2008). Currently, human behavior package. This paper includes a comprehensive definition
of the quantitative description method including the
classification of human behavior and the definition of the
ranges of human behavior in a residential building, with a
detailed case study to help understand the method.

2 Quantitative description method

2.1 Human behavior classification and description

Different kinds of human behavior lead to great differences

in energy use, which is affected by several factors: time,
environment, psychology, economic, etc. A quantitative
description of human behavior requires that the factors be
quantifiable (can be numerically described) and measurable
Fig. 1 Factors influencing building energy use
(can be physically measured).

Fig. 2 Relationship between humans and buildings

Peng et al. / Building Simulation / Vol. 5, No. 2 87

The two kinds of factors that can be used to quantitatively

describe behaviors are time and environment. According
to the investigation and discussion above, human behavior
in buildings can generally be divided into three categories
(Table 1):
(1) time related: actions according to time; Fig. 3 Quantitative description flow chart
(2) environmentally related: actions to achieve a certain
control objective, depending on the environment; Some actions are long life habits, such as opening windows
(3) random: actions that depend on uncertain factors or for ventilation. Some adapt to new environments or con-
something not quantifiable. ditions, such as the use of new electrical equipment. The
three human behavior modes can be used to quantitatively
Table 1 Quantitative description method describe both human actions in residential buildings, as
Type Input Parameters shown in Table 2 to Table 4.
Time related Schedule Set value Human behavior tends to vary periodically in buildings
Actions are based on a specified schedule and the set value, such as the due to the influences of day and night, workdays, weekends
AC set temperature and season. Thus, human behavior models need to pay
Environmentally related Logic form Feedback parameters attention to the time period.
Some parameters such as the temperature, are controlled based on logic
with specified parameters used with feedback to implement the action 2.2 Case study examples
Random Frequency or range
Actions are not regular or may follow unknown or complex rules Typical human behavior was investigated with measurements
in a residential building in Beijing for the floor plan shown
A quantitative human behavior description requires in Fig. 4. The household information is shown in Table 5.
three steps as shown in Fig. 3. First, the object related to the The detailed information is in the Electronic Supplementary
behavior must be identified. Second, investigations are Material of this paper. There were several face-to-face studies
used to classify the behavior and identify which parameters of the behavior of each person in the family. Heat sensors
need to be measured, such as hourly TV power and indoor were put in each room to measure the room temperature
temperatures. Data analysis is then used to describe the with power meters used to measure the power used by the
human behavior. appliances, as shown in Table 6. The measurements lasted

Table 2 Time related mode examples

Behavioral object Schedule Set value

Window On / off
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Example: a housewife, as long as the weather conditions are suitable, opens the window for fresh air for an hour every day from 8: 00 a.m. to 9:00 a.m.

TV On / off
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Example: an office worker turns on the TV after work at about 7:00 p.m., turns it off at 11:00 p.m.

Table 3 Environmentally related mode examples

Behavioral object Logic form Feedback parameters
If FP>29℃, then turn on with a set value of 26℃, otherwise keep closed
Air-conditioner FP: indoor temperature
Action time: 19:00 – 24:00 in the summer
Example: a resident turns on the air-conditioner when he feels hot at a temperature of about 29℃
If FP<100 lx, then turn on, otherwise turn off
Lights FP: interior illumination
Action time: 19:00 – 23:00 annual
Example: a resident turns on the living room lights when he feels dusky, the illumination is about 100 lx
Note: 1. Feedback parameters (FPs) must be measurable: for example, temperature, luminance, and concentration of carbon dioxide.
2. The air-conditioning set value is a kind of behavior parameter that must be measured or investigated.
88 Peng et al. / Building Simulation / Vol. 5, No. 2

Table 4 Random mode examples

Behavioral object Frequency Duration
Washing machine 2 times/week 1 hour/cycle
Example: washing machine is used twice a week and each cycle lasts about 1 hour
Lights 5 times/day 10 min/cycle
Example: lights in the bathroom are turned on about 5 times a day for about 10 minutes each time

2.2.1 Time related mode

The investigation showed that the TV usage in bedroom 1

was time related. The usage schedule was found by surveying
all members of the household which showed that the
occupants of this room usually watched TV three times per
day. The power data for June, 2010 and January, 2011 showed
that the TV in bedroom 1 was used about 4 hours a day as
shown in Fig. 5. The usage times and durations changed a
little each day.
The TV usage probability in Fig. 6 shows that the TV is
generally used in three time periods from morning to noon,
Fig. 4 Household floor plan during the afternoon, and at night. The behavior was
generalized by the behavior model to quantitatively describe
Table 5 Household information this behavior.
Number Note The duration of each period was then calculated from
People 5 One couple, a grandmother, a son and a maid
measured data with the duration T, chosen equal to the
Located in the living room, master bedroom,
mathematical expectation for each time period. The start
TV 4 times were selected so that the model periods spanned the
bedroom 1, bedroom 3
Located in the living room, master bedroom, largest probability times. Thus, the TV usage time periods in
Air-conditioner 4
bedroom 1 and bedroom 2 bedroom 1 were then calculated as
Computer 1 Located in the master bedroom
n
Washing machine 1 Located in the kitchen T = å xi Pi
Refrigerator 1 Located in the kitchen i

where, xi: time step, chosen as 1 minute which was the data
Table 6 Measuring instruments measurement rate; Pi: usage probability for each time step
from the measurements.
Besides the TV usage in bedroom 1, the TV usage in
bedroom 3, the AC usage in bedroom 2 and the refrigerator
operation were all time related while the AC usage in the
master bedroom was environmentally related.
– The refrigerator is on every day, so the probability during
each time step was 1 and
T = å xP = å x ´1 = 24
WZY-1 heat sensor S100 power meter – The AC power usage in bedroom 2 is shown in Fig. 7. The
occupant stated that he would turn on the AC once he
for 10 days during the summer, 2010 and 1 month during the arrived home and would turn it off when he left. The power
winter, 2011. Both instruments were connected to a computer usage in Fig. 7 shows the times the resident was at home
to record and store data. from the investigation as the green line. The red line is
The real life human activities were divided into the three the AC power and the blue line is the indoor temperature.
modes defined in Table 1. The measured data was used to The resident left home at about 7 a.m. and returned at
classify the actions. about 3 p.m. The AC power usage then follow this trend,
Peng et al. / Building Simulation / Vol. 5, No. 2 89

Fig. 5 Daily TV usage in bedroom 1 (time related)

being on when he was at home and off when he left. This The behavior schedule is obtained by investigations or
behavior is also described as T = å xP where P is the measurements. For example, the probability of using the AC
probability the resident is at home. or opening windows can be obtained from measured data
The time related behavior is implemented as while the amount of time a person spends at home can be
(1) the actions take place at fixed intervals (day, week), determined through a survey.
(2) the usage time in each period is the same or nearly
2.2.2 Random mode
the same.
There were four TVs in the household. The electric used by
these TVs was about 1.5 kWh/day, about 20% of the total
household electric use. The people reported that the TV usage
in the master bedroom was random. The survey showed that
the watching TV frequency was approximately once per
day with 1 hour per instance on workdays and 2 hours per
instance on weekends. The measurements in Fig. 8 show
that the occupant does not use the TV every day but the TV
is used for more than 4 hours as some days with no clear
Fig. 6 Usage model for bedroom 1 TV
pattern in the master bedroom.

Fig. 7 AC usage in bedroom 2

90 Peng et al. / Building Simulation / Vol. 5, No. 2

Fig. 8 Daily TV usage in the master bedroom (random action)

A statistical analysis showed that the TV in the master indoor temperature. The occupant survey showed that they
bedroom is used an average for one and a half hour a day. turned on the AC when they felt hot, which they estimate
Analysis of the behavior probability showed that the resident corresponded to an indoor temperature of 29℃ and they
mostly watches TV between 19:00 and 24:00 (Fig. 9). The preferred to set the AC to 26℃.
random behavior was then characterized by a 1.5 hours The feedback and set values (the temperatures) were
period per day. then found by measuring the indoor temperature and the
The Random mode represents complex behavior related power usage. As shown by the orange line in Fig. 10, the
to uncertain factors. This information can be obtained only feedback temperature was 29℃ which means that when the
by observing their actions and the cycle of behavior depends resident are active, once the indoor temperature is over 29℃,
on the kind of object. More study is needed about this kind they will turn on the AC. The measured set temperature was
of human behavior. 27℃, which differed from the survey estimate.
As shown in Fig. 11, the two important parameters for
2.2.3 Environmentally related mode the environmentally related mode are the feedback value and
The investigation showed that the AC usage in the master the set value. When the temperature exceeds to the feedback
bedroom was environmentally related as a function of the value, t1, the occupants turn on the AC and set the tem-
perature to t2. Thus, this kind of action requires two values
to describe the behavior.
The environmentally related mode assumes that:
(1) Behavior is influenced by environmental factors such
as temperature or luminance.
(2) The key factors can be measured.
(3) The feedback logic can be described mathematically.
The feedback logic describes a relationship between the
environmental factors and the behavior that is seen in the
Fig. 9 Random action time scales for the master bedroom TV
measured data; thus, the environmental parameters and the

Fig. 10 Master bedroom AC usage and temperature variations (environmentally related)

Peng et al. / Building Simulation / Vol. 5, No. 2 91

different kinds of human behavior. Since each kind of

behavior is linked to a certain object, the object is used to
describe each kind of behavior. For example, a TV is used
to provide a quantitative description of the watching TV
behavior.
Human behavior in residential buildings is far more
than just interactions with objects. Each family unit in a
residential building is defined as a household. There are
Fig. 11 Environmentally related AC usage behavior then many different objects in each room, such as windows,
air-conditioners and lamps, which are controlled by people
behavior must be measured simultaneously. For example, the with different action modes in the household which can
temperature and AC power usage must be measured together be functions of the types and positions of the objects in the
to analyze the environmentally related AC behavior. building.
A lifestyle describes people’s own behavioral habits at
2.3 Quantitative description of the residential building home, for example, AC use in the summer, opening of
windows for ventilation, and washing clothes frequencies.
2.3.1 Physical description The sum of all the object related behavioral habits in a
residence constitute a lifestyle. A lifestyle model is built
Research on human behavior in buildings seeks to analyze based on the quantitative description to describe the human
the behavioral influence on the building energy use. The
behavior in the household.
behavioral impact on building energy use is both indirectly
and directly related to the building elements such as windows,
3 Energy consumption simulations for different
shades and appliances. Thus, the term “objects” refers to the
lifestyles
building elements and the equipment related to energy use
which can be controlled by the occupants.
A change in an object’s state reflects the human behavior 3.1 Lifestyle model
in the building, with the initial conditions in the simulations
The lifestyles in buildings are affected by many factors.
based on a set of object states. A building-human behavioral
Lifestyles can be divided into the three types listed in Table 8
model is based on description of the quantitative changes
based on differences in life paradigms, professions, economic
of an object’s state. Model simulations are then an effective,
conditions, etc.
scientific way to analyze human behavior in buildings. The
objects included in this study are described in Table 7.
3.2 Quantitative human behavior simulations
2.3.2 Residential building description
Human behavior in residential buildings can be summarized
Objects are chosen to quantitatively describe the three
based on the typical lifestyles described in Table 8. The time
spent at home will affect the AC usage. While there may
Table 7 Residential building objects
be some differences between two families having the same
Type Description
Building envelope Adjustable elements, such as windows, curtains Table 8 Typical lifestyles in China
and shades
Lifestyle Feature Example
Air conditioning Equipment or system to keep the air cool and dry
Energy conscious People who consider energy Older families,
Heating system Equipment or system to warm the air
lifestyle (EC) consumption to be important will low-income
Ventilation system Mechanical system in a building to provide fresh air pay more attention to saving families
Lighting Artificial sources of visible illumination electricity, gas, water, etc. in their
daily lives. Their actions in the
Hot water equipment Supply hot water for bathing, washing, drinking,
building will seek to use as little
cooking, etc.
energy as possible
Office appliances Appliances related to work, such as computers,
Habit related People who are less concerned about Most white collar
printers, projectors
lifestyle (HR) saving energy than living a energy families
Domestic appliances Appliances for household activities conscious lifestyle
Elevator Elevators for transporting people or goods High quality People who have a high demand for Rich and young
Others System services for security, fire control, etc. lifestyle (HQ) indoor environmental quality families
92 Peng et al. / Building Simulation / Vol. 5, No. 2

lifestyle mode, the main features of their behavior patterns conscious lifestyle family is the shortest. The temperature
will be the same. Therefore, one typical family for each lifestyle during the weekend for the high quality lifestyle family is
will be used as examples to build the lifestyle model. controlled at 24℃, which results in the highest energy use
The lifestyle models were built using DeST, a building of the three kinds of lifestyles during these four days.
energy simulation program developed by Tsinghua University The different lifestyles result in the different AC usage
(DeST 2008). A special version for human behavior called times and electrical consumption due to differences in the set
DeST-m is being developed that includes human behavior temperatures as shown in Fig. 13. The household electricity
as an important input parameter along with the building use for all parts of the residence is shown in Fig. 14.
elements and equipment. The different lifestyles result in sharp differences in
The lifestyle behavior model includes the four layers of building performance and energy use. In the case of the
action model, object, room and household. The building energy conscious lifestyle, the residents try to save energy
model in DeST for a residential building in Shanghai chosen and may endure an uncomfortable living environment. They
for simulation is shown in Fig. 12. The detailed inputs of care about how much energy they consume, reducing electric
lifestyles are in the Electronic Supplementary Material of appliance usage by only turning on the AC when they feel
this paper. really hot. Families with habit related lifestyle will operate
The human behavior in the building is then summarized objects in the way they think is suitable. Sometimes they
by the household lifestyle. The analysis considers one will regulate the objects state according to the environment
household in the building. and sometimes they change their habits because of a work
The human behavior greatly impacts the building schedule change. People with high quality lifestyle consume
environment and energy use. For example, the AC usage a great amount of energy to control their living environment.
times and set temperatures are controlled by the resident. They use the AC and lighting all the time when in the house.
The AC usage for different lifestyles is described in Table 9. The electricity consumption of the household appliances and
Simulation results for four typical days (two work days lighting then are also quite different. Since the high quality
and two off days) in the summer of the three kinds of lifestyle lifestyle object usage time is the longest for most appliances
are shown in Table 10 for the AC usage, indoor tem- (such as the TV, PC and washing machine), their energy use
perature and household load. The AC usage for the energy levels are larger than the other two kinds of family lifestyles.

Fig. 12 A six floor residential building model in Shanghai in DeST

Table 9 AC usage and set temperature for the different lifestyles

Lifestyle Summer Winter
If Tin>30℃, then turn on, otherwise off If Tin<16℃, then turn on, otherwise off
EC Action time: 15:00 – 18:00 in August Action time: 15:00 – 18:00 in January
Set temperature: 28℃ Set temperature: 18℃
If Tin>28℃, then turn on, otherwise off If Tin<18℃, then turn on, otherwise off
HR Action time: set by resident Action time: set by resident
Set temperature: 26℃ Set temperature: 20℃

Work day 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Work day 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
HQ Off day Off day
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Set temperature: 24℃ Set temperature: 22℃

Peng et al. / Building Simulation / Vol. 5, No. 2 93

Table 10 Impact of different AC usage modes on the room temperature and household load
Lifestyle AC operation Indoor temperature Household load

lifestyles are being currently studied in real situations for

future work.

4 Conclusions

Human behavior greatly impacts building performance and

energy use. Quantitative descriptions of human behavior are
then a prerequisite for analyzing and forecasting a person’s
impact on the building performance and energy use.
Fig. 13 Annual AC electrical consumption for different lifestyles This research presents a quantitative description method
to define and quantify human behavior. The factors used for
the quantitative description should be able to be numerically
described and physically measured, such as times and
temperatures. Objects usages are divided into a time related
mode, an environmentally related mode and a random mode
to describe individual human behavior in a residential
building by analyzing actual usage. These modes are defined
based on key behavioral factors and can be formulated
in simulation tools for energy usage analyses of residential
buildings.
To make the simulations more workable, human behavior
is classified into a relatively small set of typical lifestyles.
Fig. 14 Annual household electrical consumption for different Each of these lifestyles represents the typical behavior of a
lifestyles certain group of people. Simulations then show that different
lifestyles significantly change the building performance
There are many different kinds of lifestyles besides and energy use. Other lifestyles will be identified based on
those listed in this study. The three lifestyles used in this investigations and measurements of human behavior in a
study are typical lifestyles in China. There may be differences residential building in future work. Much research still
in families within the same lifestyle, but they have key needs to be done to fully understand human behavior in
common behavioral characteristics. This means that people’s buildings. There is an emerging need to combine quantitative
behavior in the building can be divided into several categories, descriptions of human behavior with building simulations
each with their own characteristics. We can then model to help understand and forecast real building energy use
their behavior based on family lifestyles. Additional typical based on on-site measurements and simulations. More
94 Peng et al. / Building Simulation / Vol. 5, No. 2

case studies will be done to fully optimize the quantitative Klein SA, Beckman WA, Mitchell JW, Duffie JA, Duffie NA, Freeman
description method. TL, Mitchell JC, Braun JE, Evans BL, Kummer JP, Urban RE,
Fiksel A, Thornton JW, Blair NJ, Williams PM, Bradley DE,
Acknowledgements Mcdowell TP, Kummert M (2004). TRNSYS 16—A transient
System Simulation Program, User Manual. Solar Energy Laboratory,
This project is financially supported by the National Natural University of Wisconsin-Madison.
Korjenic A, Bednar T (2011). Impact of lifestyle on the energy demand
Science Foundation of China (No. 51008176), the Specialized
of a single family house. Building Simulation, 4: 89  95.
Research Fund of the Doctoral Program of Higher Education
Li Z, Jiang Y (2006). Characteristics of cooling load and energy
of China (No. 20100002120014), and Tsinghua University
consumption of air conditioning in residential buildings in Beijing.
Initiative Scientific Research Program (No. 2009THZ0). Heating Ventilating & Air Conditioning, 36(8): 1  6. (in Chinese)
Electronic Supplementary Material: Supplementary material Mahdavi A, Lambeva L, Mohammadi A, Kabir E, Pröglhöf C (2007).
Two case studies on user interactions with buildings’ environmental
(case study of human behaviors in a household and the
systems. Bauphysik, 29: 72  75.
examples of the three types of lifestyles) is available in the
Nicol JF (2001). Characterising occupant behaviour in buildings:
online version of this article at [Link]
Towards a stochastic model of occupant use of windows, lights,
s12273-011-0049-0. blinds, heaters and fans. In: Proceedings of IBPSA Building
Simulation Conference and Exhibition (pp. 1073  1078), Rio de
References Janeiro, Brazil.
O’Doherty J, Lyons S, Tol RSJ (2008). Energy-using appliances and
Annex 53 (2010). Total energy use in building—Analysis and evaluation energy-saving features: Determinants of ownership in Ireland.
methods. IEA Energy Conservation in Building and Community Applied Energy, 85: 650  662.
Systems Programme, [Link]. Ouyang JL, Gao JL, Luo XY, Ge J, Hokao K (2007). A study on the
Bourgeois DJ (2005). Detailed occupancy prediction, occupancy-sensing relationship between household lifestyles and energy consumption
control and advanced behavioral modelling within whole-building of residential buildings in China. Journal of South China University
energy simulation. Ph D Thesis, l’Université Laval, Québec. of Technology (Natural Science Edition), 35(z1): 171  174. (in
Crawley DB, Lawrie LK, Winkelmann FC, Buhl WF, Huang YJ, Chinese)
Pedersen CO, Strand RK, Liesen RJ, Fisher DE, Witte MJ, Glazer Reinhart CF (2004). Lightswitch-2002: A model for manual and
J (2001). EnergyPlus: Creating a new-generation building energy automated control of electric lighting and blinds. Solar Energy,
simulation program. Energy and Buildings, 33: 319  331. 77: 15  28.
DeST (2008). DeST-c User Manual. DeST Group, Tsinghua University, Rijal HB, Tuohy P, Humphreys MA, Nicol JF, Samuel A, Clarke J
China. (in Chinese) (2007). Using results from field surveys to predict the effect of
EneryPlus (2009). Input/Output Rerference, Version 4.0 Documentation. open windows on thermal comfort and energy use in buildings.
University of Illinois and Ernest Orlando Lawrence Berkeley Energy and Buildings, 39: 823  836.
National Laboratory, USA. van Raaij WF, Verhallen TMM (1983). A behavioral model of residential
ESRU (1999). ESP-r: A Building and Plant Energy Simulation energy use. Journal of Economic Psychology, 3: 119  143.
Environment, User Guide Version 9 Series. ESRU Publication, Yan D, Xia J, Tang W, Song F, Zhang X, Jiang Y (2008). DeST—An
University of Strathclyde, Glasgow. integrated building simulation toolkit Part I: Fundamentals.
Hitchcock G (1993). An integrated framework for energy use and Building Simulation, 1: 95  110.
behavior in the domestic sector. Energy and Buildings, 20: 151  157. Yun GY, Steemers K (2008). Time-dependent occupant behavior models
Hoes P, Hensen JLM, Loomans MGLC, de Vries B, Bourgeois D (2009). of window control in summer. Building and Environment, 43:
User behavior in whole building simulation. Energy and Buildings, 1471  1482.
41: 295  302. Zhang X, Xia J, Jiang Z, Huang J, Qin R, Zhang Y, Liu Y, Jiang Y
IEEA (2008). Residential monitoring to decrease energy use and carbon (2008). DeST—An integrated building simulation toolkit Part II:
emissions in Europe. Applications. Building Simulation, 1: 193  209.
BUILD SIMUL (2013) 6: 23 – 32
DOI 10.1007/s12273-013-0106-y

Statistical analysis and modeling of occupancy patterns in open-plan

offices using measured lighting-switch data

Research Article
Wen-Kuei Chang1, Tianzhen Hong2 ()

1. Green Energy and Environment Laboratories, Industrial Technology Research Institute, Taiwan, China
2. Environmental Energy Technologies Division, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA

Abstract Keywords
Occupancy profile is one of the driving factors behind discrepancies between the measured and building simulation,
simulated energy consumption of buildings. The frequencies of occupants leaving their offices occupancy model,
and the corresponding durations of absences have significant impact on energy use and the occupancy pattern,
operational controls of buildings. This study used statistical methods to analyze the occupancy occupant schedule,
status, based on measured lighting-switch data in five-minute intervals, for a total of 200 office buildings,
open-plan (cubicle) offices. Five typical occupancy patterns were identified based on the average statistical analysis

daily 24-hour profiles of the presence of occupants in their cubicles. These statistical patterns
were represented by a one-square curve, a one-valley curve, a two-valley curve, a variable curve,
Article History
Received: 10 October 2012
and a flat curve. The key parameters that define the occupancy model are the average occupancy
Revised: 19 December 2012
profile together with probability distributions of absence duration, and the number of times an
Accepted: 25 December 2012
occupant is absent from the cubicle. The statistical results also reveal that the number of absence
occurrences decreases as total daily presence hours decrease, and the duration of absence from
© Tsinghua University Press and
the cubicle decreases as the frequency of absence increases. The developed occupancy model
Springer-Verlag Berlin Heidelberg
captures the stochastic nature of occupants moving in and out of cubicles, and can be used to 2013
generate a more realistic occupancy schedule. This is crucial for improving the evaluation of the
energy saving potential of occupancy based technologies and controls using building simulations.
Finally, to demonstrate the use of the occupancy model, weekday occupant schedules were
generated and discussed.

1 Introduction low energy office building design process, which applied

Building Thermal, lighting,

and Acoustics Modeling
a statistical method, found that realistic user behavior
Building energy simulation tools have been widely applied should be incorporated into passive cooling design concepts
in recent years in energy saving proposals for new con- (Pfafferott and Herkel 2007). A methodology that takes
struction designs and existing building retrofits. However, into account the variation in occupant behavior and
simulated results sometimes deviate significantly from schedules was proposed to estimate the cooling demand in
measured data. Such discrepancies can be attributed to residential units (Tanimoto et al. 2008). Its authors concluded
several factors. One of the most important is occupant that occupant behavior is a significant factor in residential
behavior in buildings. Many studies demonstrate that building cooling requirements, though the methodology needs further
occupancy profiles have a significant impact on energy use validation to confirm its plausibility.
and the operational controls of buildings. An investigation Various modeling approaches have been developed for
into the impact of consumer behavior on residential energy use in building energy performance simulations to predict
demand found that consumer behavior is the most important occupancy characteristics in different types of buildings.
issue with respect to energy consumption in households A stochastic user behavior model generates a time series
(Haas et al. 1998). A simulation of user behavior for the of window operations by using Markov chains (Fritsch et
E-mail: thong@[Link]
24 Chang and Hong / Building Simulation / Vol. 6, No. 1

al. 1990). However, the lack of adequate measurements and personality. It is therefore not surprising that simulated
makes computing the Markov matrices impossible. The use energy use deviates from actual consumption in most
of stochastic models to capture human behavior and occupant situations. Although various occupancy models have been
interaction within a building attempts to simulate multiple developed to predict occupancy profiles in buildings, they
influences that occupants can have on a building in terms usually lack validation from adequate field-measured data.
of resource consumption (Page et al. 2008). The results This study uses statistical methods to analyze lighting-
sometimes overestimate and other times underestimate the switch data collected from the open office spaces of an
weekly total energy use and peak demands. A model that office building to identify variations in occupancy patterns.
combines user presence and interaction in a building showed Various occupancy patterns and characteristics are identified,
that improved modeling of user behavior in numerical and a robust occupancy model is being developed to generate
simulations can optimize overall building performance (Hoes more realistic occupant schedules. The results of this study
et al. 2009). A model of activity and location schedules was can be used to understand further and evaluate the impact
developed, using a system of USSU—User Simulation of of occupancy patterns on building energy performance, and
Space Utilization, to generate movement patterns that provide to improve the accuracy of predicting the actual energy use
a representation of human activities in office building spaces of buildings with simulation tools.
(Tabak 2008). However, there were obvious differences
between the observed and predicted human activity behavior 2 Data collection
related to the number of times a workplace was used
during a working day. A model based on Markov chains A total of 200 lighting-switch sensors were installed in
that simulates the movement of occupants inside an office open office cubicles on three floors of an office building.
building can produce more realistic occupancy variations, The numbers of switches installed on each floor are listed
nonsynchronous change of occupancy in time, and an in Table 1. Each cubicle had a single, workstation-specific
uneven distribution in space (Wang et al. 2011). However, suspended fixture with a built-in occupancy sensor. The
more validation and calibration approaches must be carried sensor detected occupant movement and controlled the
out with specific occupant-movement patterns. Behavioral lighting switch for each cubicle. The light was activated
patterns associated with energy spent on heating were (switched on) if the cubicle was occupied, and deactivated
determined statistically, and household and building (switched off) if unoccupied. All occupancy sensors were
characteristics were identified (Santin 2011). It appears calibrated and control systems were commissioned before
difficult to establish relationships between behavioral patterns data were collected. The lighting control system recorded a
and energy consumption. daily log of sensor switch events, including the presence
Recent years have seen the introduction of systems and and absence of occupants, every five minutes. Switch events
devices that can be controlled on a personal basis. These were recorded as 1 or 0, indicating the cubicle was occupied
efforts to improve energy efficiency and increase energy or unoccupied, respectively. In this study, each cubicle was
savings include lighting, office equipment, thermostats for assumed to be unoccupied until the occupant arrived for
heating, ventilation, and air conditioning, windows, and the first time in the morning. After the first occupancy event,
blinds. Accurately estimating the savings and impacts of the data was filled in with 1 or 0, based on the most recent
these systems and technologies requires the accurate event for each cubicle.
prediction of how often and how long occupants stay in This study used data collected for weekdays, weekends,
their offices. Therefore, the impact of occupancy profile on and holidays from May through November in 2011. In a
building energy performance becomes more important. small number of cases there may be some errors in the data
The occupancy pattern defined in the present study is the due to sensor sensitivity and coverage. Switch sensors
frequency of an occupant leaving his/her cubicle and the sometimes are triggered by people walking past cubicles, or
corresponding duration of the absence. It is part of the fail to trigger if occupants remain overly static in their
broader occupant behavior which includes occupant’s
interactions with building envelope and energy systems. A Table 1 Number of lighting switches on three floors of an office
method for obtaining realistic and stochastic occupancy building
is a key concern for building energy simulations, in order Building floor Number of switches
to precisely evaluate the performance of occupancy-based Floor A 104
controls. Currently, most simulation tools apply fixed or Floor B 47
predefined occupancy schedules to represent the time when
Floor C 49
occupants are present. However, occupancy pattern can
Total 200
change significantly according to the season, weather, time,
Chang and Hong / Building Simulation / Vol. 6, No. 1 25

cubicles. Although these cases cannot be excluded in this switch-on profile, respectively. Finally, the variable curve
study, their occurrence is relatively infrequent and should occupancy pattern was defined if the valley occurred twice
not have a noticeable impact on the results. The collected or more. After all occupancy patterns were determined, the
data for weekdays were processed in parallel with data for occurrence percentages of each occupancy pattern could be
weekends and holidays to provide a more accurate view of calculated by counting the frequency of each occupancy
occupancy profiles. The goal was to obtain general occupancy pattern for each floor. By accumulating the probabilities of
trends and patterns for a large number of office cubicles to the five patterns individually, and then dividing by the total
allow for comparisons across each floor. Data were processed number of each occupancy pattern, the average occupancy
for as many valid days as possible, including time periods pattern was determined. Daily working hours were divided
during and after commissioning. Exclusions were made due into four two-hour time periods. The occurrence times of
to missing or incomplete switch data files and insufficient each occupancy pattern for each time period on the three
switch number information. Some days were excluded floors were collected to determine the occurrence percentages
due to the control system going offline temporarily, which of each occupancy pattern, and the relationships between
resulted in incomplete data collection. The final data used occupancy and working time period.
in this study includes 76 weekdays and 34 weekend days The number of daily absences and absence durations
and holidays. of each occupancy pattern were calculated to further
understand the characteristics of each occupancy pattern.
3 Analysis methods Switch-off events tracked when the occupant vacated the
cubicle. Accumulating these events provided time and
Once the collected data were finalized, they were statistically duration information and allowed further understanding of
analyzed to identify occupancy patterns during weekdays their relationship. According to the results, a noticeable
and weekends. The number of daily absences and their valley usually occurred during noon in the occupancy
durations were determined, and the occupancy variations patterns. Therefore, daily working hours were re-divided
were distinguished. into three time periods: 8–11:30 a.m., 11:30 a.m.–1:30 p.m.,
The switch-on events were recorded every minute. and 1:30–6 p.m. The number of daily absences and absence
Therefore, the presence duration of each occupant can be durations in each time period were summarized to investigate
obtained by accumulating the number of switch-on events. when the valley occurred in the occupancy pattern.
The total monthly presence hours were calculated by adding
up the daily presence hours. The average daily presence 4 Results
hours of each occupant were determined by dividing the
total presence hours in each month by the number of data- The profiles of occupant presence hours for each floor are
collection days in that month. Thus, the profiles of occupant shown in Fig. 1. The working time is divided into four periods,
presence hours of the three floors were determined. every two hours. The percentages of occupant presence
Additionally, the daily occupancy profiles of each floor hours for each floor were very different. For Floors A and C,
during weekdays and weekends were obtained by averaging most occupants, 40% and 31% respectively, stayed in their
the probabilities of switch-on events for each cubicle each cubicles for 4 to 6 hours per day. Only a few occupants
month. stayed over 6 hours. On Floor B, occupancy pattern was
A total of 200 occupancy patterns of three floors are significantly different from Floors A and C. Most occupants,
illustrated according to the probabilities of switch-on events. about 66%, on Floor B stayed in their cubicles for around
Different occupant’s behavior results in different occupancy 2 hours per day. There was no one staying for more than
patterns. Based on the variations of each occupancy pattern 6 hours. The average presence hours of Floor B were almost
curve, these 200 occupancy patterns were classified into five half those of Floors A and C. This may indicate that
types: a single-square curve (Fig. 4(a)), a one-valley curve different agencies with different job categories work on
(Fig. 4(b)), a two-valley curve (Fig. 4(c)), a variable curve different floors. The occupants of Floor B may work
(Fig. 4(d)), and a flat curve (Fig. 4(e)). A valley was identified half-time, or work at home or outside the office part of the
when the switch-on profile started to drop and then rise time. Therefore, a working occupant may not always be in
when the difference between the maximum and minimum his or her cubicle. Furthermore, this study observed that
switch-on percentage values exceeded 20%. A single-square occupancy patterns were influenced slightly by the location
curve occupancy pattern was defined if there wasn’t a of the cubicle. Longer occupancy periods occurred in more
valley apparent from the switch-on profile. Similarly, the isolated cubicles that had more privacy, or cubicles that
one-valley curve and two-valley curve occupancy patterns were near windows. However, job category may have more
were defined if the valley occurred once or twice in the impact on occupancy pattern than location of the cubicle.
26 Chang and Hong / Building Simulation / Vol. 6, No. 1

Fig. 1 Profile of presence hours for each floor: (a) Floor A; (b) Floor B; (c) Floor C

Unfortunately, private information like job category for each

occupant was not available for this study.
The average daily weekday switch-on profile of each
floor is shown in Fig. 2. In general, the occupants of each
floor arrived at and departed from the office between 6 a.m.
and 6 p.m. on weekdays. The switch-on percentage of each
floor increased in the morning and reached a peak value at
around 9 a.m. The maximum values of Floors A, B, and C are
about 48%, 16%, and 32%, respectively. A higher switch-on
percentage means higher occupancy. The increase in switch-
on rate of Floor A was greater than that of Floors B and C. Fig. 3 The average daily weekend profile of switch-on events for
Figure 2 also shows that the switch-on percentage of each each floor
floor has an obvious drop at around noon, attributed to
occupants leaving the office for lunch. Also, it can be seen and for Floors B and C was almost equal to 0%. Therefore,
that the switch-off rate of Floor A is greater than that of the this study only focuses on the investigation and analysis of
other two floors. The occupants of each floor began to leave data collected for weekdays.
work approximately between 3 p.m. and 4 p.m. Compared The numbers of lighting-switch sensors installed on
with the decrease in switch-on rate of Floors B and C, the Floors A, B, and C were 104, 47, and 49, respectively. This led
decrease in switch-on rate of Floor A is greater. In addition, to 200 occupancy profiles. The collected occupancy profiles
several spikes occurred after 6 p.m. This can be attributed can be classified into five patterns by occupancy variation,
to the cleaning crews in the evening. The cleaning schedules presence duration in the cubicle, and occupant personality,
of Floors A, B, and C are 6:35 p.m. to 8 p.m., 5:05 p.m. to as shown in Fig. 4. These occupancy patterns are very
6:30 p.m., and 9:25 p.m. to 10:50 p.m., respectively. The different from one another. In Fig. 4(a), the pattern looks
spikes occur within these time periods and the switch-on like a single-square curve. The percentage of occupants
percentage of each floor is about 5%. As for weekends, the stay in the cubicle is more than 60% within daily working
average daily profiles of switch-on events for each floor are hours except two time periods: one from 6 to 8 a.m. when
shown in Fig. 3. Compared with the weekday profiles, the occupants arrive at the office, and the other from 4 to 6 p.m.
weekend switch-on percentages are quite low for all three when occupants get off work. Figure 4(a) indicates that
floors. The switch-on percentage of Floor A was less than 3% occupants leave their cubicles fewer times and with shorter
duration during working hours. Alternatively, this pattern
can be interpreted as the stationary time in which an occupant
does not leave or enter their cubicle frequently. Several
spikes occur after 6 p.m., the reason for which is discussed
in our description of Fig. 3. Figure 4(b) shows an occupancy
pattern similar to Fig. 4(a), except for an observable deep
valley occurring at midday for a period of approximately
1 to 1.5 hours. This can result from the occupant leaving for
lunch. The occupant leaves the cubicle after approximately
11:30 a.m. for lunch and then returns to the cubicle at
approximately 1 p.m. This pattern can be interpreted as the
Fig. 2 The average daily weekday profile of switch-on events for occupant not leaving or entering the cubicle frequently,
each floor but leaving for lunch at midday. Figure 4(c) shows two
Chang and Hong / Building Simulation / Vol. 6, No. 1 27

Fig. 4 The occupancy patterns: (a) single-square curve; (b) one-valley curve; (c) two-valley curve; (d) variable curve; (e) flat curve

noticeable valleys in this pattern. In addition to the valley that for public usage, such as a print station, coffee shop, or office
occurs around noon, another valley appears in the morning. supply room. This kind of pattern will not be discussed
This can be attributed to a longer absence by the occupant, further in this study.
such as attending a meeting or leaving the building. Based on the number of occupants on each floor in
However, the valley observed in this study not only occurs Fig. 2, the occupancy patterns of all occupants were further
in the morning but also in the afternoon (although it is identified. Figure 5 shows occurrence percentages of each
not shown in Fig. 4(c)). Figure 4(d) shows a significant occupancy pattern for the three floors. The designations of
variation in the pattern. There is no regular pattern as with Patterns 1 to 5 shown in this figure correspond to Figs. 4(a)
Figs. 4(a)–(c). This pattern shows the occupant leaving the to (e) as discussed above, and these designations will be
cubicle frequently during work time and being absent for further used in later discussion. Compared with Floor B,
longer amounts of time. Figure 4(e) shows a flat occupancy the occurrence percentages of each pattern are similar for
pattern; the cubicle seldom appears occupied and the occupied Floors A and C. Pattern 2 is the most typical occupancy
duration is short. This can be attributed to a cubicle used pattern, about 45% and 39% for Floors A and C, respectively.
28 Chang and Hong / Building Simulation / Vol. 6, No. 1

Fig. 5 The occurrence percentages of each occupancy pattern: (a) Floor A; (b) Floor B; (c) Floor C

For Floor B, however, the highest occupancy pattern is represents a total of 9 days when the occupant left the
Pattern 5, with an occurrence percentage of about 38%. This cubicle 4 times per day. In this figure, it can be found that
significant difference can be attributed to different agencies
working on different floors, as discussed above.
The occurrence percentages for each occupancy pattern
for the three floors in four time periods are listed in Fig. 6.
Circles displayed in this figure indicate the occurrence times
of the pattern. Larger circles represent higher occurrence
times. Occurrence percentages of Pattern 2 for each floor
were found to be higher than those of other patterns when
occupants stayed in their cubicles for 2 to 8 hours per day.
This indicates that most occupants of each floor left for
lunch during the noon hour. The second highest is Pattern 1,
which represents occupants who did not leave or enter their
cubicles frequently. Additionally, the occurrence percentages
of Pattern 1 for each floor were higher than those of
Patterns 2 to 4 when occupants stayed in their cubicles for
less than 2 hours per day.
The analysis results described above are occupancy
patterns that only represent the overall characteristics of
cubicles occupied on each floor. It is still very approximate
for use as an occupancy schedule in building simulation
tools. For example, the switch-on percentage of Pattern 1
was about 60% during the working hours of 8 a.m. to 6 p.m.
This indicates that the probability of an occupant in the
cubicle was about 60%. However, the number of daily
absences and absence durations still cannot be obtained
via this occupancy pattern. An occupant’s number of daily
absences and absence durations can have significant
impact on energy usage and cause substantial differences
between measured and simulated energy use. To obtain
more accurate simulation results, a more realistic occupancy
schedule—including presence and absence durations of
occupants, and the number of absences in the cubicle—is
required for use in the simulation. Therefore, the number
of daily absences and absence durations of each occupancy
pattern were further identified and detailed, as follows.
Figure 7 represents the accumulated number of daily
absences within the 76-day period for Patterns 1 to 4. The
days when the occupant did not arrive at the office are Fig. 6 The occurrence percentages of each occupancy pattern in
excluded. For example, if the number of daily absences and four time periods for each floor: (a) Floor A; (b) Floor B; (c) Floor
the number of occurrences were 4 and 9, respectively, this C. Larger circles represent more occurrences
Chang and Hong / Building Simulation / Vol. 6, No. 1 29

The outlines of occupancy profiles in Patterns 1 to 3

were similar except for one and two significant valleys in
Patterns 2 and 3. To further understand occupancy patterns,
Pattern 2 was further investigated as follows. The working
time in a day was divided into three periods: 8 to 11:30 a.m.,
11:30 a.m. to 1:30 p.m., and 1:30 to 6:00 p.m. The total
number of absences and average absence durations for
these time periods for Pattern 2 are illustrated in Figs. 9(a)
and (b). The number of absences shown here are the
accumulated numbers within the 76-day period, and the
Fig. 7 The accumulated number of daily absences for each absence durations are the average values. The number of
occupancy pattern in a 76-day period for Patterns 1–4 absences increased as the day progressed. The occupant left
the cubicle more often and with a shorter duration in the
the maximum number of occurrences of each occupancy afternoon. This may be due to the dwindling concentration of
pattern shifted and decreased with the number of daily an occupant or increasing fatigue, resulting in the occupant
absences. The most typical numbers of daily absences of walking around or going to the restroom more often. The
each pattern are 1, 4, 5, and 9. For Pattern 1, there are a average absence duration from 11:30 a.m. to 1:30 p.m. was
total of 5 days when the occupant never left the cubicle. significantly longer than the others, as this is the lunch
Although the peaks of the other patterns were less than period. However, the average absence durations from 8 to
that of Pattern 1, each of the total number of absences of 11:30 a.m. and 1:30 to 6:00 p.m. were almost the same.
Patterns 2, 3, and 4 was almost greater than those of Pattern The accumulated numbers of absence minutes of each
1, except the cases with none or one daily absence. More time period for Pattern 2 is illustrated in Fig. 10. The most
daily absences indicate that the occupant entered or left the typical absence duration for all three time periods was 10 to
cubicle more frequently. 19 minutes. The percentages for each time period were
The accumulated numbers of durations of each absence 43%, 25%, and 45%. Figure 10 shows that occurrence times
within the 76-day period for each pattern is shown in Fig. 8. decrease with longer absence minutes. This corresponds to
The absence durations concentrate in a 10-to-29-minute span the result mentioned before. However, the occurrence times
for all occupancy patterns except Pattern 4. For Patterns 1
and 2, most absence periods were 10 to 19 minutes in
duration, and their occurrence percentages were 41% and
37%, respectively. For Pattern 3, most absences lasted 20 to
29 minutes, and the percentage was 56%. For Pattern 4,
most absences were 0 to 9 minutes, and the percentage
was 62%. For all patterns, occurrence times decreased with
longer absence minutes. It can be deduced that for shorter-
duration absences, the occupant leaves the cubicle to take a
break, go to the restroom, or walk around. Longer periods
can be attributed to meetings, lunch, or outside business.

Fig. 9 Total number of absences and average absence duration of

Fig. 8 The accumulated number of occurrences of absence duration three time periods for Pattern 2 over a 76-day period: (a) number
for each occupancy pattern over a 76-day period of absences; (b) absence duration in minutes
30 Chang and Hong / Building Simulation / Vol. 6, No. 1

occupants leave their cubicles for lunch around noon, in

addition to other longer events, such as attending meetings
or going outside. The second most popular occupancy is
Pattern 1, in which occupants do not leave or enter their
cubicles frequently. Additionally, the occurrence percentages
of Pattern 1 for each floor are higher than those of other
patterns when occupants stayed in their cubicles for less
than 2 hours per day.
The number of absences and absence duration for
each occupancy pattern are identified. More daily absences
Fig. 10 The accumulated number of occurrences of each absence mean an occupant moves in and out of the cubicle more
duration for occupancy Pattern 2 for three time periods over a frequently. The most typical numbers of daily absences for
76-day period
Patterns 1 to 4 are 1, 4, 5, and 9, respectively. Additionally,
the absence durations for each absence are mostly from 10
of absence minutes for different time periods can be further to 29 minutes for all occupancy patterns. The number of
distinguished. For the time periods of 8 to 11:30 a.m. and absences decreases with the longer absence duration for all
1:30 to 6:00 p.m., the curves dropped drastically after a peak patterns. For a short absence duration, it can be deduced
and then descended slowly. Compared with the period of that an occupant leaves the cubicle to take a break, go to
1:30 to 6 p.m., more absences of longer duration occurred the restroom, or walk around. On the other hand, a longer
from 8 to 11:30 a.m. It can be deduced that there were more period can be attributed to an occupant attending a meeting,
meetings or longer events in the morning. For the time having lunch, or going outside.
period 11:30 a.m. to 1:30 p.m., the curve declined more Finally, the working time in a day is divided into three
smoothly after the peak and the times of longer absences periods for further analysis of occupancy patterns. Occupants
were higher than the other two time periods. This figure leave the cubicle more often in the afternoon but for shorter
indicates that the occupant may spend over 10 minutes and durations. In other words, the occupants leave the cubicle
sometimes almost 2 hours for lunch. less often but for longer. The average absence at midday is
longer due to lunch. However, the average absence durations
5 Discussion in the morning and afternoon are almost the same.
This study also observes that occupancy patterns are
Cubicle occupancy for a typical 8-hour weekday for the slightly influenced by cubicle location. Longer occupancy
three floors mostly begins between 8 and 9 a.m., with a dip periods occur in more isolated cubicles that have more privacy
around noon, and then begins to decrease from 4 to 6 p.m. or are near windows. However, job category may have more
Spikes, caused by the late-night cleaning crews after most influence on occupancy pattern than cubicle location. Due
occupants have left in the evening, are also observed. to privacy concerns, no data is available to further relate job
Weekend occupancy levels for cubicles on all three floors characteristics to occupancy patterns.
are fairly low and can be neglected. Furthermore, weekday
occupancy levels for Floor B are very different from the other 6 Occupancy model and schedule generation
two floors, which can be attributed to different agencies
working on different floors, with occupants on Floor B Based on the results, a stochastic occupancy model of
working part time, going out for business more often, each pattern is developed with three key elements: (1) the
or working from home part of time. Due to privacy and cumulative distribution function (CDF) of the number of
security concerns, no further data is available to allow daily absences, (2) the CDF of each absence duration, and
further verification. (3) the probability distribution function (PDF) of the start
200 occupancy patterns for the three floors were collected time of each absence.
in this study. These collected patterns can be classified into For an open-plan office with a certain number of cubicles,
five patterns according to occupancy variation, appearance assuming one occupant per cubicle, a profile of occupancy
duration in the cubicle, and occupant personality. The five patterns must first be determined by energy modeling. Then
identified occupancy patterns are: Pattern 1 (single-square occupancy schedules for a weekday can be generated by the
curve), Pattern 2 (one-valley curve), Pattern 3 (two-valley following steps, using Patterns 1 and 2 as examples. First, a
curve), Pattern 4 (variable curve), and Pattern 5 (flat curve). uniform-distribution random number between 0 and 1 is
Statistical results show that the most common occupancy generated, and it is used as an input to the inverse function
among all occupants is Pattern 2, which indicates that most of the CDF of the number of daily absences in Fig. 11(a) to
Chang and Hong / Building Simulation / Vol. 6, No. 1 31

find the corresponding number of daily absences. For each there is mostly one daily absence, lasting 10 to 30 minutes.
absence, a uniform-distribution random number between 0 Three generated weekday occupant schedules of Pattern 2
and 1 is generated, and it is used as an input to the inverse are shown in Fig. 14. As with the occupant schedules of
function of the CDF of the daily absence duration in
Fig. 11(b) to find the corresponding daily absence duration
in minutes. Finally, for each absence, a uniform-distribution
random number is generated and used to calculate the start
time of each absence. After that, the end time of each
absence can be determined by adding the absence durations
previously calculated. For Pattern 1, according to Fig. 4(a),
the absence start time can be assumed to be uniformly
distributed between 8 a.m. and 4 p.m. For Pattern 2, the
absence start time is not uniformly distributed, as there is a
deep valley at around noon, as shown in Fig. 4(b). Therefore,
the distribution of number of absences is determined by
the relative probability of occurrence in the three time
periods: morning, noon, and afternoon, based on Fig. 12.
For each absence in either of the three time periods, the Fig. 12 The curve of cumulative distribution function (CDF) of
same procedure as Pattern 1 is used to determine the absence daily absence section for Pattern 2
start time.
Three generated weekday occupant schedules of Pattern 1
are shown in Fig. 13. The value 1 in the figure indicates the
occupant is in the cubicle, while 0 indicates the occupant is
away from the cubicle. It can be seen that for Pattern 1,

Fig. 13 Three generated weekday occupant schedules for Pattern 1

Fig. 11 The curves of occurrences, probability distribution function

(PDF), and cumulative distribution function (CDF) of Pattern 1:
(a) number of daily absences; (b) absence duration Fig. 14 Three generated weekday occupant schedules for Pattern 2
32 Chang and Hong / Building Simulation / Vol. 6, No. 1

Pattern 1, most absence durations last 10 to 30 minutes, but Acknowledgements

the number of daily absences are increased to 3 or 4, and
one absence occurs during noon. The authors thank Joy Wei and Abby Enscoe for providing
the lighting-switch data and answering our questions. This
7 Conclusions work was supported by the U.S. Department of Energy
under the U.S.-China Clean Energy Research Center for
This study statistically analyses information collected from Building Energy Efficiency, and it was co-sponsored by the
200 cubicle offices on three floors of a commercial office Bureau of Energy, “Ministry of Economic Affairs, Taiwan”.
building. It used measured lighting-switch data to represent
the occupancy status of cubicles. Occupancy levels were References
identified and occupancy profiles were classified into five
patterns as displayed in Figs. 4(a)–(e). The number of daily Fritsch R, Kohler A, Nygard-Fergusono M, Scartezzini JL (1990). A
absences and absence durations for each occupancy pattern stochastic model of user behavior regarding ventilation. Building
were further calculated and analyzed. Based on these results, and Environment, 25: 173  181.
a mathematical model to describe the occupancy patterns, Haas R, Auer H, Biermayr P (1998). The impact of consumer behavior
including the probability distributions of the number of on residential energy demand for space heating. Energy and
absences and absence duration, was developed. The Buildings, 27: 195  205.
Hoes P, Hensen JLM, Loomans MGLC, de Vries B, Bourgeois D (2009).
occupancy model can be used to generate more realistic
User behavior in whole building simulation. Energy and Buildings,
occupant schedules for open-plan cubicle offices, for use in
41: 295  302.
building energy simulations. In addition to lunch breaks,
Page J, Robinson D, Scartezzini J (2008). A generalized stochastic model
more occupancy events such as meetings, short visits, walking
for the simulation of occupant presence. Energy and Buildings,
around, and late-night cleaning can be taken into account
40: 83  98.
in the model to better capture the stochastic nature of actual
Pfafferott J, Herkel S (2007). Statistical simulation of user behavior in
occupancy variations in the building. These more detailed
low-energy office buildings. Solar Energy, 81: 676  682.
occupancy schedules can replace the fixed or predefined Santin OG (2011). Behavioral patterns and user profiles related to energy
ones currently used in building energy simulations to better consumption for heating. Energy and Buildings, 43: 2662  2672.
assess the impact of occupancy patterns on building energy Tabak V (2008). User simulation of space utilisation: System for office
performance, and to improve the accuracy of simulated building usage simulation. PhD thesis, Eindhoven University of
results. This method can also be used to validate and enhance Technology, The Netherlands.
other building occupancy models. However, more case studies Tanimoto J, Hagishima A, Sagara H (2008). A methodology for peak
and measured data analyses are needed. The analysis energy requirement considering actual variation of occupant’s
methods used in this study can also be adapted to study behavior schedules. Building and Environment, 43: 610  619.
the occupancy patterns of private offices and other building Wang C, Yan D, Jiang Y (2011). A novel approach for building occupancy
types, such as residential. simulation. Building Simulation, 4: 149  167.
 Occupant Behavior: Impact on Energy Use of Private Offices

Tianzhen Hong1,*, Hung-Wen Lin 2

1
Environmental Energy Technologies Division, Lawrence Berkeley National
Laboratory, Berkeley, USA
2 Green Energy and Environment Research Laboratories, Industrial Technology

Research Institute, Taiwan

ABSTRACT
Measured energy use of buildings demonstrated large discrepancies even between
buildings with same function and located in similar climates. Among various factors
contributing to the discrepancies, occupant behavior is a driving factor. Occupant
behavior is also one of the most significant sources of uncertainty in the prediction of
building energy use by simulation programs. How occupants set the comfort criteria
(including thermal, visual, and acoustic), interact with building energy and services
systems, and response to environmental discomfort directly affect the operation of
buildings and thus their energy use. This study employs building simulations to
evaluate the impact of occupant behavior on energy use of private offices with single
occupancy. Typical occupant behavior we studied includes how an occupant sets
comfort criteria, operates lights, office equipment, space thermostat, and HVAC
systems. The behaviour is categorized into three workstyles: 1) austerity – occupants
are proactive in saving energy, 2) standard – average occupants, and 3) wasteful –
occupants do not care about energy use. The simulation results demonstrate the
impact of occupant behavior on building energy use is significant, and even so at the
energy end use levels such as lighting, space cooling and heating. For a typical
single-occupancy office room, compared to the standard or reference workstyle, the
austerity workstyle consumes up to 50% less energy, while the wasteful workstyle
consumes up to 90% more energy.

Three methods are proposed to model occupant behavior depending upon the
complexity: 1) use EnergyPlus directly, 2) use the advanced feature of EnergyPlus -
Energy Management System, and 3) use modified code of EnergyPlus. Our study
provides a method to evaluate energy impact of occupant behavior, which can be a
good tool for decision makers of behavioral programs that target energy savings in
buildings.

KEYWORDS
Building simulation, Energy use, EnergyPlus, Occupant behavior, Office buildings

____________________________________________
*Corresponding author email: thong@[Link]
INTRODUCTION
Occupant behavior affects the building energy use directly and indirectly by
opening/closing windows, turning on/off or dimming lights, turning on/off office
equipment, turning on/off heating, ventilation, and air-conditioning (HVAC) systems,
and setting indoor thermal, acoustic, and visual comfort criteria. Measured energy use
of buildings demonstrated large discrepancies even between buildings with same
function and located in similar climates. Among various factors contributing to the
discrepancies, occupant behavior is a driving factor. Occupant behavior is also one of
the most significant sources of uncertainty in the prediction of building energy use by
simulation programs due to the complexity and inherent uncertainty of occupant
behavior. With the trend towards low energy buildings that reduce fossil fuel use and
carbon emissions, getting occupants actively involved during the design and operation
of buildings is a key to achieving high energy performance without scarifying
occupant comfort or productivity. Pilot projects demonstrated that low energy systems,
such as natural ventilation, shading to control solar heat gains and glare, daylighting
to dim lights, and demand controlled ventilation, especially need the interactions and
collaborations of occupants. Energy savings from 5 to 30% were achieved by
behavioral studies that motivate changes to occupant behavior.

In the last decade, new designs target net-zero energy buildings which emphasize the
importance of energy efficiency technologies, integrated design, building operation
and maintenance, and occupant behavior. Good operation practice and high design
efficiency in buildings could lower the energy use (Mahdavi et al. 2008, Linden et al.
2006). Santin (2011) looked at the relationship between user behavior and space
heating energy use, and concluded that behavior patterns could be used in building
energy calculations and usage profiles with different behavior could be discerned.

Mahdavi (2008) described an effort to observe control-oriented occupant behavior in

a few office buildings in Austria. His results imply the possibility of identifying
certain patterns of user control behavior as a function of indoor and outdoor
environmental parameters such as illuminance and irradiance. However, his
observations also underscore the need for typologically differentiated occupancy and
control action models for different buildings. Parys (2009) evaluated various lighting
and blind control systems in combined with four types of user behavior in office
buildings in Belgium. His simulation results demonstrated that the energy savings of a
daylight dimming system in an individual office decrease by about 10% when the
occupant behavior is accounted for.

On windows operating, Haldi (2008) and Rijal (2008)’s study are based on the
presumption that the main driver of occupant window intervention is occupant
discomfort. The adaptive thermal comfort model by Humphrey and Nicol (1998),
proposed that the occupants’ comfort temperature changes with the monthly outdoor
air mean temperature from a number of surveys conducted world-wide for natural
ventilated buildings. Although the adaptive comfort model was original obtained for
naturally ventilated buildings, it can be adapted for mechanically cooled spaces. Rijal
(2008) proposed a method of implementing Humphrey’s observations of occupant
window opening behaviour in a building simulation model, assuming that occupants
only interact with windows when they are thermally uncomfortable, defined as 2°C
above the upper bound or below the lower bound of the adaptive comfort temperature.

Peng et al (2012) presents a quantitative description method of human behavior in

residential buildings. The method can be used to predict the impact of the human
behavior on the indoor environment and energy use. It was applied to a household in
Beijing with comparisons to on-site observations of the occupants’ behavior and
measurements of energy use for validation.

The objective of this study is to identify, understand, and categorize occupant

behavior that can have significant impact on energy use of private offices, and
evaluate how different types of occupant behavior affect the energy use by building
simulations. The study applies to private offices with single occupancy, assuming the
occupant has freedom to interact and change his indoor environment. Open offices or
private offices with multiple occupants involve the complexity of group behavior,
which is not covered in the study.

RESEARCH METHODS
First, occupant behavior in private offices is categorized into three different
workstyles according to the level of energy is used to provide comfort for the
occupants: 1) the Austerity workstyle with occupants being proactive in saving energy;
2) the Standard workstyle representing most occupants in terms of average energy use
behavior; and 3) the Wasteful workstyle with occupants consuming energy at will,
lacking motivation to reduce energy use. The three types of occupant behavior is
based on literature review and occupants surveys like the post occupancy survey done
by Center of the Built Environment, University of California at Berkeley; they aim to
represent general situation. Then building simulations using EnergyPlus (USDOE
2012) version 7.0 are employed to quantify and evaluate the impact of the three
workstyles on energy use of private offices. To look at the influence of climate, three
U.S. typical climates are studied.

The energy metric used in the study is the source or primary energy use by the
individual office, which includes the source energy of the natural gas for heating, and
the source energy of electricity for cooling, ventilating, lighting, and office equipment
(plug-load).

Characteristics of the private offices

Three adjacent and equal size private offices, located on the south facade of a middle
story of a medium size office building, are selected for the study. Each office has only
one exterior wall (with a window) facing south, and has a rectangular shape with a
floor area of 15 m2. The private office is occupied by only one person, and is served
by a constant air volume HVAC system with heating from a gas furnace and cooling
from a direct-expansion unitary system. The efficiency levels of the building envelope,
lighting, and HVAC are set to meet the minimum requirements of ASHRAE Standard
90.1 (2004). The internal loads, including the interior lighting power and plug loads
(both at 10.76 W/m 2), and operation schedules (Figure 1) stay the same across
climates. The building operates 6am to 10pm, while the typical private office is
occupied 8am to 5pm. Cooling and heating thermostat of the private offices are set to
24°C and 21°C respectively. The occupant is assumed to take three breaks: half hour
in the morning, one hour lunch, and half hour in the afternoon. The middle office is
occupied by one person with standard workstyle, while the adjacent two offices are
each occupied by one person with austerity and wasteful workstyle respectively. The
interior walls of the three offices are insulated well to ignore heat transfer from
adjacent offices.

Figure 1. Schedules of lighting, plug-load, and people

Climate zones
Three climates, Miami (Hot and Humid), San Francisco (Coastal, Mild), and Chicago
(Cool Summer, Cold Winter), are selected in this study to represent typical climates in
the U.S. Table 1 lists the climate zone information for the three representing cities
based on ASHRAE Standard 90.1-2010. In the table, HDD18 is the Heating Degree
Days with a base temperature of 18°C, and CDD10 is the Cooling Degree Days with a
base temperature of 10°C.

Table 1. Characteristics of selected cities and climate zones

Cities ASHRAE Climate Zones HDD18 CDD10
Miami Hot –Humid, 1A 200 9474
San
Francisco Warm-Marine, 3C 3016 2883
Chicago Cool-Humid, 5A 6176 3251

The TMY3 weather data was used in the EnergyPlus simulations. The TMY3 weather
data represented typical weather conditions during 1991 to 2005 and was available for
download at EnergyPlus web site (USDOE 2012).
Occupant behavior
Typical occupant behavior related to energy use is studied and summarized in Table 2,
including:
• Cooling setpoint
The Standard occupant prefers a room air temperature of 24°C during cooling.
The Austerity occupant prefers a warmer temperature of 26°C, while the
Wasteful occupant likes a cooler temperature of 22°C. The lower the cooling
setpoint, the higher the cooling energy use.
• Heating setpoint
The Standard occupant prefers a room air temperature of 21°C during heating.
The Austerity occupant prefers a lower temperature of 18°C, while the
Wasteful occupant likes a warmer temperature of 23°C. The higher the heating
setpoint, the higher the heating energy use. Note that the heating setpoint of
the Wasteful occupant is actually higher than the cooling setpoint, which is not
unusual for people with such workstyle.
• Adaptive comfort
Adaptive comfort theory allows the indoor cooling comfort temperature to be
adjusted upward based on the monthly average outdoor air temperature. Hot
climates with higher monthly average outdoor air temperatures would have
higher indoor comfort temperatures. The Austerity occupant adjusts the
cooling setpoint based on the adaptive comfort model, while the Standard or
Wasteful occupant does not. As shown in Figure 2, for Miami climate, the
cooling setpoint in July and August can be adjusted as high as 26.5°C, which
is 2.5°C higher than the constant setpoint 24°C. This reduces the cooling
energy use.

Figure 2. Adjusted cooling setpoints based on the ASHRAE adaptive comfort model
• Occupancy controls
For the Austerity occupant, he turns off lights and HVAC, and turns down
plug-load 30% when he leaves office for break. The Standard occupant
operates lights, HVAC, and office equipment according to schedules (Figure
1). The Wasteful occupant leaves everything 100% on during breaks.
• Daylighting controls
The Austerity occupant dims lights to 50% or completely turns them off if
adequate daylight meets the visual comfort. The other two occupants do not
response to daylight.
 • HVAC operation time
Compared to the standard HVAC operation schedule, the Austerity occupant
turns on HVAC one hour late at 9am and turns off one hour early at 4pm. The
Wasteful occupant sets the HVAC operation the same as the whole building -
from 6am to 10pm.
• Cooling startup control
The Austerity occupant turns on cooling only when he feels warm, which
usually occurs when the space air temperature reaches 28°C. When the cooling
is turned on, cooling setpoint temperature of 24°C is maintained. This is
demonstrated in Figure 3 for a hot summer day. The other two occupants set
the startup temperature the same as the cooling setpoint.

Figure 3. Cooling startup control

Table 2. Occupant behavior categorized into three workstyles

Occupant behavior Austerity Standard Wasteful
workstyle workstyle workstyle
Cooling setpoint (°C) 26 24 22
Heating setpoint (°C) 18 21 23
Adaptive comfort Yes None None
If unoccupied, turn
off lights and Leave everything
Occupancy controls Scheduled on: lights, HVAC,
HVAC, turn down and plug-load
plug-load 30%
Daylighting Control 3 Steps Dimming None None
Turn on 1 hour late Scheduled on: Same as the whole
HVAC operation time and turn off 1 hour building schedule:
early: 9am to 4pm 8am to 5pm 6am to 10pm
Cooling turns on
when space air Follow HVAC Follow HVAC
operation
Cooling startup temperature schedule (8am operation schedule
reaches 28°C, then (6am to 10pm) to
control maintains at 24°C. to 5pm) to maintain 24°C.
maintain 24°C.
Cooling turns off Same as above. Same as above.
when unoccupied.
Combined All above All above All above
behavior behavior behavior

Modeling approaches
Three different approaches using EnergyPlus, in order of difficulty, are used in the
study to model the occupant behavior discussed before:

1) Direct modeling with EnergyPlus

Occupant behavior, including cooling setpoint, heating setpoint, daylighting control,
and HVAC operation time, is modelled directly with EnergyPlus by changing
corresponding inputs from the base cases for the Standard occupant. The advantage of
this approach is easy implementation.

2) Using the energy management system (EMS) in EnergyPlus

EMS is an advanced feature of EnergyPlus and designed for users to develop
customized high-level, supervisory control routines to override specified aspects of
EnergyPlus modeling in the EMS program. EMS has certain limitations and its use
requires advanced knowledge of EnergyPlus and computer programming. EMS is
used to model occupant behavior of adaptive comfort and Occupancy control. The
Occupancy control can also be modelled directly by pre-calculating the new schedules
for lights, HVAC, and office equipment, but the Direct Modeling approach would not
work if the occupant schedule is stochastic.

3) Modifying EnergyPlus source code

Modifying the existing EnergyPlus source code, the third modeling approach, is used
when both the Direct Modeling and EMS approaches cannot be applied. This
approach requires users to have a thorough understanding of the EnergyPlus data
structure and existing source code before being able to modify code. This is the most
difficult approach but offers the most flexibility to model complex occupant behavior.
This approach is used to model the Cooling start up control.

RESULTS AND DISCUSSIONS

The simulation results are presented as percentage changes of source energy of the
Austerity workstyle and Wasteful workstyle compared to the Standard workstyle for
individual occupant behavior as well as the combined behavior. Figures 4 to 6 show
the results for the three climates.
 Figure 4. Changes of source energy in San Francisco climate

Figure 5. Changes of source energy in Chicago climate

Figure 6. Changes of source energy in Miami climate

From these results, it can be seen that:

• The combined Austerity workstyle and Wasteful workstyle have significant impact on
energy use of the private office. Compared to the Standard workstyle, the Austerity
workstyle can save 42%, 50%, and 48% of source energy in San Francisco, Chicago, and
 Miami respectively; while the Wasteful workstyle consumes 89%, 81%, and 74% more
energy for the three climates respectively.
• For the Austerity workstyle, the Cooling startup control, the Occupancy control, and the
Cooling setpoint have the most energy savings. While for the Wasteful workstyle, the
Cooling startup control is the same as the HVAC operation time, and the Cooling setpoint
cause the most increase of energy use.
• The impact of Heating setpoint is relatively small because the heating source is natural
gas which is valued much less in source energy compared to other end uses in electricity.
• The adaptive comfort model based Austerity occupant behavior can save 30% of source
energy for the hot climate of Miami.
• Occupant behavior that leads to longer HVAC operation time and lower cooling setpoint
increase of energy use significantly.
• Occupant behavior that leads to delay the cooling (Startup control), higher cooling
setpoint (including Adaptive comfort), and turning off or down equipment when
unoccupied reduce energy use significantly.

CONCLUSION
This study identified and evaluated a few typical occupant behavior related to
operation and control of energy service systems of private offices. The behavior is
categorized into three workstyles – Austerity, Standard, and Wasteful – according to
the potential impact on energy use. The simulation results demonstrate that occupant
behavior has significant impact on energy use of private offices – the combined
Austerity workstyle can save up to 50% of source energy, while the combined
Wasteful workstyle can increase energy use by 89% compared to the Standard
workstyle.

Three approaches to modelling occupant behavior using EnergyPlus are discussed.

Our on-going research focuses on occupant behavior in operating windows and
shading devices, and implementing our behavior models in EnergyPlus for public use.

It is a different topic, well worth exploring but outside our expertise, on how to
motivate occupants to change from Standard workstyle to Austerity workstyle to save
energy. There have been many pilot behavioral programs presented in the Behavior,
Energy, and Climate Change conference (Anon.).

ACKNOWLEDGEMENTS
This work was supported by the U.S. Department of Energy under the US-China
Clean Energy Research Center on Building Energy Efficiency. It is also part of our
research activities for the IEA ECBCS Annex 53: Total Energy Use in Buildings –
Evaluation and Analysis Methods. This work was co-sponsored by the Bureau of
Energy, Ministry of Economic Affairs, Taiwan.
REFERENCES
Anonymity [Link] last accessed on 30 July, 2012.
USDOE. 2012. EnergyPlus, [Link] last accessed on 31 July,
2012.
Mahdavi A, Mohammadi A, Kabir E, and Lambeva L. 2008. Shading and lighting
operation in office buildings in Austria: a study of user control behavior, Building
Simulation. 1: 111-117.
Linden AL, Carlsson-Kanyama A, and Eriksson B. 2006. Efficient and inefficient
aspects of residential energy behaviour: what are the policy instruments for
change? Energy Policy. 34: 1918-1927.
Santin OG. 2011. Behavioural patterns and user profiles related to energy
consumption for heating, Energy and Buildings. 43: 2662-2672.
Haldi F and Robinson D. 2008. On the behaviour and adaptation of office occupants,
Building and Environment. 43(12): 2163–2177.
Rijal HB, Tuohy P, Nicol JF, Humphreys MA, Samuel A, and Clarke J. 2008.
Development of an adaptive window-opening algorithm to predict the thermal
comfort, energy use and overheating in buildings, Journal of Building
Performance Simulation. 1(1): 17–30.
Humphreys MA and Nicol JF. 1998. Understanding the adaptive approach to thermal
comfort, ASHRAE Transactions. 104 (1): 991 – 1004.
ASHRAE. 2004. ANSI/ASHRAE Standard 90.1-2004, Energy standard for buildings
except low-rise residential buildings, Atlanta: American Society of Heating,
Refrigerating, and Air-Conditioning Engineers, Inc.
ASHRAE. 2010. ANSI/ASHRAE Standard 55-2010, Thermal environmental
conditions for human occupancy, Atlanta: American Society of Heating,
Refrigerating, and Air-Conditioning Engineers, Inc.
Parys W, Saelens D, and Hens H. 2009. Impact of occupant behavior on lighting
energy use, Proceedings of Building Simulation 2009, Glasgow, UK, pp
1143-1150.
Peng C, Yan D, Wu RH, Wang C, Zhou X, and Jiang Y. 2012. Quantitative
description and simulation of human behavior in residential buildings, Building
Simulation, 5(2): 85-94.
 DATA ANALYSIS AND MODELING OF LIGHTING ENERGY USE IN

LARGE OFFICE BUILDINGS

Xin Zhou1, Da Yan1,, Xiaoxin Ren1, Tianzhen Hong2

1
Department of Building Science, School of Architecture, Tsinghua University, Beijing, China
2
Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA

ABSTRACT
Lighting consumes about 20 to 40% of total electricity use in large office buildings in the U.S. and
China. In order to develop better lighting simulation models it is crucial to understand the
characteristics of lighting energy use. This paper analyzes the main characteristics of lighting
energy use over various time scales, based on the statistical analysis of measured lighting energy
use of 17 large office buildings in Beijing and Hong Kong. It was found that the daily 24-hour
variations of lighting energy use were mainly driven by the schedule of the building occupants.
Outdoor illumination levels have little impact on lighting energy use in large office buildings due to
the lack of automatic daylighting controls and relatively small perimeter areas. A stochastic lighting
energy use model was developed based on different occupant activities during six time periods
throughout a day, and the annual distribution of lighting power across those periods. The model was
verified using measured lighting energy use of one selected building. This study demonstrates how
statistical analysis and stochastic modeling can be applied to lighting energy use. The developed
lighting model can be adopted by building energy modeling programs to improve the simulation
accuracy of lighting energy use.

KEYWORDS
building simulation, energy use, lighting, modeling, occupant behavior, office buildings, Poisson
distribution

INTRODUCTION
Lighting energy use in large office buildings is as high as 20% to 40% of the building total in both
China and the U.S. This has caught the attention of practitioners, researchers, and policy makers.
Studies have shown that the two main factors affecting lighting energy use are outdoor illumination
and occupant behavior. From other researchers’ field studies and simulations, it was concluded that
lighting energy use has a correlation with outdoor illumination. When the outdoor illumination is
above a certain level, people around perimeter zones with access to natural light are less likely to
use artificial electrical lights, and the artificial illumination needed to meet design illuminance
levels is lower (Reinhart and Voss 2003, Li et al 2006, Galasiu and Atif 2002, Li and Lam 2001,
Maitreya 1997). However, other researchers also found that occupants have a crucial influence on
the lighting energy use. Through case studies of actual buildings, Yun et al (2012a) found that in
open-plan offices, the usage of lighting was not influenced by outdoor illumination. Instead it had a
close relationship with the indoor activities of occupants. Meanwhile, Yong et al (2012b) formulated

Corresponding author. E-mail address: yanda@[Link]
the concept that outdoor illumination has no statistical significance with lighting energy use, and the
operation of lighting was strongly correlated to the time of day. Other studies also found that
operation of lights by occupants only depended on whether the room is occupied, and is
independent of outdoor illumination (Love 1998, Lindelof and Morel 2006).

Currently, most research on lighting energy use is focused on small office and residential buildings.
The analysis methods and conclusions from this research provide some hints to help understand the
lighting energy use in large office buildings. In China, a common method to predict lighting energy
use involves combining lighting power density information with lighting schedules. However, the
generation of lighting schedules is too simplified and lacks verification against measured data (Yun
and Steemers 2008). This leads to a large discrepancy between simulated and measured lighting
energy use (Bluyssen 2009, Norford et al 1994). Furthermore, the annual variation of actual lighting
energy use is not [Link] complex lighting energy use models have been reviewed. Hunt
(1979) introduced a stochastic model to calculate the probability of turning on lights after the arrival
of occupants. He concluded that the probability of occupants turning on artificial lights increases
only when the illumination of the working surface is below 100 lux. Newsham (1995) developed
the Lightswitch model that followed a stochastic approach and simulated user occupancy at the
workplace based on measured field data in an office building in Ottawa, Canada. Reinhart (2004)
improved the Lightswitch model to Lightswitch-2002 to calculate the probability of occupants
arriving and leaving offices, and the related probability of turning on and off lights. Meanwhile, in
Reinhart‘s study, based on the model, the amount of energy savings under different lighting control
strategies was evaluated. Joakim Wide´n et al (2009) used Markov chains to estimate the probability
of occupant movement. Then the probability of turning on lights was modeled as a decision based
on the lighting level and occupant movement. Since these studies were mainly based on small office
buildings (Hunt 1979, Reinhart 2004) and residential buildings (Wide’n et al 2009), there is a strong
need to conduct more research on lighting energy use in large office buildings if energy use targets
are to be met.

Based on large quantities of measured data from several large office buildings, this paper analyzes
the characteristics of total amount and distributions of lighting energy use in large office buildings.
Due to the lack of detailed information on the physical characteristics of lighting systems in these
buildings, this paper focuses on daily and seasonal lighting energy use patterns. Daily 24-hour
lighting curve, annual distribution of lighting power, and main influencing factors of lighting energy
use are first identified and discussed through statistical analysis of hourly data. Then a stochastic
model is developed which effectively capture random characteristics of lighting energy use. The
model accounts for the time-varying nature of lighting energy use, including peaks in usage at
certain times of the day. In this study, the general lighting energy use features of large office
buildings are analyzed and discussed in depth, and the main influencing factors and distributions of
lighting energy use are clarified more distinctly.

RESEARCH METHODS
The research method in this paper is shown in Figure 1. First, the two main factors influencing
lighting energy use - outdoor illumination and occupant behavior - are analyzed. To determine the
influence of outdoor illumination on lighting energy use, the lighting energy use between different
seasons and different building levels (above-grade areas and basements) is compared. The effect of
occupant behavior is analyzed by comparing lighting energy used on different types of day
(workdays, weekends, holidays), and by comparing lighting energy use under different occupancy
schedules. Then, based on the understanding of main influencing factors of lighting energy use,
further discussion about the feature of lighting energy use curve in large office buildings can be
gained through the analysis of measured lighting energy use from dozens of large office buildings
with energy sub-metering systems. The analysis is mainly focusing on four aspects: 1) annual total
energy use; 2) monthly distribution; 3) daily feature; and 4) annual distribution. More in-depth
analysis is conducted to decode the annual distribution feature and the time-relevant properties
between different time periods. A whole-building lighting energy use model is developed based on
the results from the analysis of lighting energy use and lighting profiles at various time scales. The
model is then applied to a case study to simulate the lighting energy use, and the simulated results
are compared with measured data to verify the model.
Lighting
model

Comparison and verification

Main Case study Lighting

influencing energy use
factors feature

Daily
Outdoor Occupant Annual total Monthly Annual
average
illumination movement energy use distribution distribution
feature

Figure 1 Research route

RESULTS
1 Analysis of influencing factors
1.1 The influence of outdoor illumination
1.1.1 Comparison of lighting energy use between the basement and the above-grade
floors
The lighting energy use is shown in Figure 2. The red line in the figure represents the daily mean
lighting energy use. The edges of the blue boxes are placed at the 25% and the 75% quartiles. The
maximum and minimum data points are also shown. It reveals that the outdoor illumination has no
obvious effect on the shape of the average lighting power curves in large office buildings.
Above-grade Floors Basement Floors
120 50
Lighting power(kW)
Lighting power(kW)

100 40
80 30
60
20
40
20 10

0 0

Time Time

Figure 2 Comparison of lighting energy use between the basements and the above-grade areas

1.1.2 Comparison of lighting energy use between seasons

To further study the influence of outdoor illumination, the lighting energy use between different
seasons is compared as shown in Figure 3. It can be concluded that outdoor illumination has no
noticeable influence on the total lighting energy use.
 250
Spring
200

Lighting power/kW
Summer
150
Autumn
Winter
100

50

Time

Figure 3 Weekly profiles of average lighting energy use for the four seasons
1.2 The influence of occupant behavior
To assess the influence of occupant behavior on lighting energy use, the power draw between
workdays and weekends for the same lighting branch is compared and shown in Figure 4 and Figure
5. As there are many more occupants in the building on workdays, the lighting power on workdays
is higher than weekends. Different occupancy events such as arriving at work, going out for lunch,
and leaving work can be detected from the workday lighting power curve. While during weekends,
the discrete range of lighting energy use is much larger, and a homogeneous lighting schedule
cannot be detected due to the uncertainty of overtime work and other events.
1000
1000
900
900
800 Lighting power(kW) 800
700 700
Lighting power(kW)

600 600
500 500
400 400
300 300
200 200
100 100
0 0

Time Time

Figure 4 Average lighting power draw on workdays Figure 5 Average lighting power draw on
weekends
Note: the red lines are the hourly averages; the green lines represent data at 5% and 95% probability;
the blue boxes show the first and third quartiles; the vertical dashed blue lines show the range
2 General characteristics of lighting energy use
Based on measured data of 17 large office buildings in Beijing and Hong Kong, general
characteristics of the lighting energy use are analyzed.

2.1 Annual lighting energy use

The electricity end-use intensities are calculated and shown in Figure 6. It can be seen that the
offices in Hong Kong have greater electricity use intensity.
200 Hong Kong
180
Electricity use intensity (kWh/m2a)

160
140
120
Lighting&Socket&Power
100
Special Function
80
Power
60
Lighting&Socket
40 AC
20
0
A B C D E F G H I J K L M N O P Q

Government Teaching Commercial

Office Buildings Office Buildings Office Buildings

Figure 6 Electricity use intensity of large office buildings in Beijing and Hong Kong
Note：Power refers to utilities equipment like elevators; AC refers to HVAC equipment.

2.2 Monthly distribution of lighting energy use

Table 1 shows the months with the maximum and minimum daily lighting energy use for the 17
buildings. It cannot be judged statistically from the results which months have the greatest or least
daily lighting energy use.
Table 1 Months with maximum and minimum daily average lighting energy use
The month with The month with
The month with the The month with the
the maximum the minimum
Buildings maximum daily minimum daily lighting Buildings
daily lighting daily lighting
lighting energy use energy use
energy use energy use
A 7 4 H 3 8
B 7 9 I 12 1
C 12 5 J 11 8
D 4 5 K 12 1
E 11 5 M 2 3
F 7 10 N 7 12
G 12 10 O 8 2
R 9 5

2.3 Daily distribution of lighting energy use

The hourly lighting energy use for a typical workday for Building F is shown in Figure 7. The curve
has dual peaks and can be divided into six time periods: 1)Night Period; 2)Going-to-work Period:;
3)Morning Period; 4)Noon-Break Period; 5)Afternoon Periodl; 6)Off-Work Period.
60.00
Llighting power(kW)

Going-to-
50.00
work Period
40.00 Morning Afternoon
Night Off work Night
30.00 Period Period Noon- Period Period Period
20.00 break
10.00 Period
0.00

Time

Figure 7 Curve of lighting power for a typical workday

The six periods can be divided into two categories:

1. Constant Power
Morning Period, Noon-break Period, Afternoon Period and Night Period are periods that can be
represented by a flat curve with a constant lighting power. Table 2 lists the maximum coefficient of
variation for each of the four periods. It indicates that the variation can be ignored.
Table 2 Maximum coefficient of variation for the four constant power time periods
Morning Noon-break Afternoon
Night Period
Period Period Period
Coefficient of
0.22 0.13 0.25 0.31
variation Vσ
Note: Vσ = σ/x, where Vσ is the coefficient of variation, σ is the mean square deviation, and x is the average
value
2. Variable power
The daily distribution of lighting energy use during the Going-to-Work and Off-Work periods
satisfies an exponential curve. Wang et al (2005) proved, with measured data, that the probability of
a certain number of people (represented by k) arriving during a certain time period fits a Poisson
distribution:
λk
P{X = k} = e−λ
k!
1
  , where T is the average time before k people arrive the office. So the probability of some
T
people arriving during a certain time period fits P = P{k > 0}, which is an exponential distribution.
And the probability of lighting turning on is related to the probability of people arriving. During the
Off-Work Period, the probability of people leaving the office can be represented by an exponential
distribution, which means that the probability of people in the office is calculated as P = 1 −
P{k > 0}.

Taking Going-to-Work period as an example, using the least square regression model, confidence
levelα= 0.05 is assumed, and the functional form is set to the exponential distribution. The results
are shown in Figure 8. From the regression curve, almost all the data is within the confidence
interval, which proves that the curve fitting is good.
Schedule

Time

Figure 8 The daily regression curve during the Going-to-Work Period [λ=0.89]

3 Annual distribution of lighting energy use

From the spread of hourly lighting use from a single lighting branch shown in Figure 9, it can be
seen that during one year the lighting energy use during these periods is not constant.

120.00
Mon. Tue. Wed. Thu. Fri. Sat. Sun.
Lighting power/kW

100.00
80.00
60.00
40.00
20.00
0.00

Time

Figure 9 Annual hourly lighting energy use

Regression analysis is used to verify the distribution characteristic, as shown in Figure 10. Most
data are within the confidence interval, which proves that the annual variations of the lighting
power during each of four periods can be represented with a normal distribution.

Morning Afternoon
Frequency

Frequency

Morning Period Afternoon Period

stage stage

Group of lighting power interval Group of lighting power interval

 Night Lunch

Frequency
Night Period

Frequency
stage Noon-break Period
stage

Group of lighting power interval Group of lighting power interval

Figure 10 The normal distribution of the lighting power

DISCUSSION
A model of the whole-building lighting energy use was developed based on the daily lighting curves
and the annual distribution properties. This model is applicable to large office buildings, where the
lighting energy use has almost no relationship with outdoor illumination, but has a close
relationship with the occupancy schedule. Only the lighting energy use on a typical workday is
simulated here. The simulated results of a lighting branch in Building A are shown in Figure 11.
With the consideration that simulation aims to represent the most typical scenarios in reality, the
data edges of this quartile graph are the data points at the probabilities of 95% and 5%. It can be
seen that the simulated daily lighting curve agrees quite well with the curve from the measured data.
The annual distributions of each period are also described.
Measured data of work day Simulation data of work day
60 60
Measured lighting power(kW)

Simulated lighting power(kW)

50 50
40 40
30 30
20 20
10 10
0 0

Time Time

Figure 11 Comparison between the simulated and measured lighting energy use
CONCLUSION AND IMPLICATIONS
This paper analyzed the main characteristics and major influencing factors of lighting energy use in
large office buildings, based on measured lighting energy use of 17 large office buildings in Beijing
and Hong Kong. It is important to describe the daily lighting profiles accurately in order to
represent the various characteristics of lighting energy use in large office buildings. A stochastic
lighting model was developed to quantify the uncertainty of occupant behavior. This paper focused
on the description of lighting energy use curves.

Main findings in this study include:

1、In large office buildings, the lighting energy use is mainly affected by the occupant schedule,
and the influence of outdoor illumination is very limited.
2、In large offices, the time when lights are turned on is closely correlated with the time when most
occupants arrive. While turning off lights is related to the time most occupants leave. Accurate
prediction of the presence of occupants in offices is crucial to predict lighting energy use.
3、Lighting is a major electric end-use in large office buildings. The annual lighting energy use per
square meter is similar for large offices in Beijing and Hong Kong.
4、Poisson and normal distributions can accurately describe the stochastic properties of daily
lighting power curves and annual variations.
5、A whole-building lighting energy use model is developed based on daily lighting curves and
annual distribution of lighting power levels. The model is verified using measured lighting
energy use from an actual building.
Future work can be done to improve the simulation accuracy of the annual distribution of lighting
power levels for the Going-to-Work and Off-Work periods. A lighting model for weekends can also
be developed and verified.

REFERENCES
Bluyssen, P.M. (2009) The Indoor Environment Handbook: How to Make Buildings Healthy and
Comfortable, Earthscan, London.
California Energy Commission (2006) California Commercial End-Use Survey. PIER report,
CEC-400-2006-005.
Galasiu, A.D. and Atif, M.R. (2002) Applicability of daylighting computer modeling in real case
studies: comparison between measured and simulated daylight availability and lighting
consumption. Building and Environment, 37, 363–377.
Hunt, D. (1979) The use of artificial lighting in relation to daylight levels and occupancy. Building
and Environment, 14, 21-33.
Joakim Wide´n, Nilsson, A.M., Wa¨ckelga˚rd, E. (2009) A combined Markov-chain and bottom-up
approach to modelling of domestic lighting demand. Energy and Buildings, 41(10), 1001–1012.
Li, D.H.W., Lam, T.N.T., and Wong, S.L. (2006) Lighting and energy performance for an office
using high frequency dimming controls. Energy Conversion and Management, 47, 1133-1145.
Li, D.H.W. and Lam, J.C. (2001) Evaluation of lighting performance in office buildings with
daylighting controls. Energy and Buildings, 33, 793–803.
Lindelöf, D. and Morel, N. (2006) A field investigation of the intermediate light switching by users.
Energy and Buildings, 38, 790–801.
Love, J.A. (1998) Manual switching patterns observed in private offices. Lighting Research and
Technology, 30, 45–50.
Maitreya, V. K. (1997) Daytime artificial lighting in office buildings in India. Building and
Environment, 12, 159–163.
Newsham, G.R., 1995. Lightswitch: A stochastic model for predicting office lighting energy
consumption. In: Proceedings of Right Light Three, 3rd European Conference on Energy
Efficient Lighting, Newcastle, UK, pp. 59–66.
Norford, L.K., Socolow, R.H., Hsieh, E.S., and Spadaro, G.V. (1994) Two-to-one discrepancy
between measured and predicted performance of a ‘low-energy’ office building: insight from a
reconcillation based on the DOE-2 model. Energy and Buildings, 21, 121–131.
Reinhart, C.F. (2004) Lightswitch 2002: a model for manual control of electric lighting and blinds.
Solar Energy, 77 (1), 15–28.
Reinhart, C.F. and Voss, K. (2003) Monitoring manual control of electric lighting and blinds.
Lighting Research and Technology, 35, 243–260.
Wang, D., Federspiel, C.C., and Rubinstein F. (2005) Modeling occupancy in single person offices.
Energy and Buildings, 37, 121-126.
Yun, G.Y., Kong, H.J., Kim, H., and Kim, J.T. (2012a) A field survey of visual comfort and lighting
energy consumption in open plan offices. Energy and Buildings, 46, 146–151.
Yun, G.Y., Kong, H.J., Kim, H., and Kim, J.T. (2012b) Effects of occupancy and lighting use
patterns on lighting energy consumption. Energy and Buildings, 46, 152–158.
Yun, G.Y. and Steemers, K. (2008) Time-dependent occupant behaviour models of window control
in summer. Building and Environment, 43, 1471–1482.
BUILD SIMUL
DOI 10.1007/s12273-013-0126-7

A detailed loads comparison of three building energy modeling

programs: EnergyPlus, DeST and DOE-2.1E

Research Article
Dandan Zhu1, Tianzhen Hong2, Da Yan1 (), Chuang Wang1

1. Department of Building Science, School of Architecture, Tsinghua University, Beijing 100084, China
2. Environmental Energy Technologies Division, Lawrence Berkeley National Laboratory, One Cyclotron Road, Berkeley, CA 94720, USA

Abstract Keywords
Building energy simulation is widely used to help design energy efficient building envelopes and building energy modeling program,
HVAC systems, develop and demonstrate compliance of building energy codes, and implement building thermal loads,
building energy rating programs. However, large discrepancies exist between simulation results comparison,
from different building energy modeling programs (BEMPs). This leads many users and stakeholders DeST,
to lack confidence in the results from BEMPs and building simulation methods. This paper compared DOE-2.1E,
the building thermal load modeling capabilities and simulation results of three BEMPs: EnergyPlus, EnergyPlus

DeST and DOE-2.1E. Test cases, based upon the ASHRAE Standard 140 tests, were designed to
isolate and evaluate the key influencing factors responsible for the discrepancies in results between
Article History
Received: 15 November 2012
EnergyPlus and DeST. This included the load algorithms and some of the default input parameters.
Revised: 6 February 2013
It was concluded that there is little difference between the results from EnergyPlus and DeST if the
Accepted: 18 February 2013
input values are the same or equivalent despite there being many discrepancies between the heat
balance algorithms. DOE-2.1E can produce large errors for cases when adjacent zones have very
© Tsinghua University Press and
different conditions, or if a zone is conditioned part-time while adjacent zones are unconditioned.
Springer-Verlag Berlin Heidelberg
This was due to the lack of a strict zonal heat balance routine in DOE-2.1E, and the steady state 2013
handling of heat flow through interior walls and partitions. This comparison study did not produce
another test suite, but rather a methodology to design tests that can be used to identify and
isolate key influencing factors that drive the building thermal loads, and a process with which to
carry them out.

1 Introduction interfaces (GUI) such as VisualDOE (VisualDOE 2004),

Building Thermal, Lighting,

EnergyPro (EnergyPro 2011), eQuest (eQuest 2009), and

and Acoustics Modeling

Computer simulation is an important and proven method EnergyGauge (EnergyGauge 2012). EnergyPlus is a next
to help understand and analyze the thermal performance of generation BEMP developed, supported and maintained by
buildings, and predict their operational energy consumption. a team led and funded by USDOE since 1996. EnergyPlus
Since the 1960s, many building energy modeling programs is based on the load algorithms of BLAST and the system
(BEMPs) have been developed to perform building energy algorithms of DOE-2. New features and enhancements were
simulation, for instance, the widely-used DOE-2 (DOE-2 added to support innovative, low-energy building designs
1980), EnergyPlus (Crawley et al. 2001), ESP (ESRU 1999), and operational controls. Development of ESP-r started in
and DeST (Yan et al. 2008; Zhang et al. 2008). DOE-2 was 1974 at the University of Strathclyde and is primarily used
developed at the Lawrence Berkeley National Laboratory in Europe. DeST (Designer’s Simulation Toolkits) is a BEMP
with funding from the U.S. Department of Energy (USDOE) developed at Tsinghua University since the late 1980s with
after the energy crisis in the late 1970s and is still the the aim of aiding teaching, research and the practical use of
most widely-used BEMP in the U.S. This includes its use as building energy analysis and simulation in China.
a stand-alone calculation engine and with graphical user BEMPs play a significant role in the design of energy

E-mail: yanda@[Link]
2 Zhu et al. / Building Simulation

efficient envelopes and HVAC (heating, ventilation, and the simulation engine that usually simplifies, hides or hard-
air-conditioning) systems for new buildings and retrofitting wires some inputs that can be important; third is the fact
existing buildings, the development of building energy codes that users may model the building or system inaccurately
and standards, and defining and implementing building as they are not familiar with the chosen BEMP, or input
energy rating/labeling programs. However, the issue that poor data due to constraints of budget and resources. In
large discrepancies exist in simulation results between order to address the issue of large discrepancies between
different BEMPs, even for the same building modeled by different BEMPs, the impact of the above three factors must
the same person, leads many users and stakeholders to lack be identified and quantified.
confidence in building simulation methods and the results This paper mainly discusses why and how different
from BEMPs. This is a major barrier for the wider adoption BEMPs produce different simulation results. As the building
and effective application of building energy simulation, and load calculation forms the basis of building energy and
represents a challenge to the industry. The large discrepancies thermal performance simulations, this paper focuses on
of simulation results between different BEMPs mainly come detailed comparisons of loads calculation between the three
from three factors as Fig. 1 illustrated: first is the simulation BEMPs: EnergyPlus, DeST, and DOE-2.1E, with the goal to
engine that is the unchangeable core; second is the GUI to identify and quantify the influences of the simulation engines
and input values or algorithms. EnergyPlus was chosen
because it is widely used and continuously being developed
and supported by USDOE. DOE-2.1E was chosen as it is still
widely used in the U.S. DeST was chosen due to its emerging
use in China and a few Asian countries and regions. Top-
level key features of DOE-2.1E, DeST and EnergyPlus are
summarized in Table 1.
Our findings can be a valuable reference for decision
makers to determine which BEMP to use for various ap-
plications including development and compliance calculations
for building energy codes and standards. Another separate
paper will discuss the methodologies and findings from a
Fig. 1 The top three influencing factors for the discrepancies in detailed comparison of the same three BEMPs in HVAC
simulation results between different BEMPs systems and central plant modeling.

Table 1 Comparison of top-level key features of DOE-2.1E, DeST and EnergyPlus

Feature DOE-2.1E DeST EnergyPlus
Developer LBNL/USDOE Tsinghua University, China USDOE
Development and support No more development or support On-going On-going
User Worldwide Mostly in China Worldwide
Input Text, BDL Database, Microsoft Access Text, IDF
Output Summary & hourly reports Summary & hourly reports Extensive summary & detailed reports with user
specified time steps
GUI Simulation engine only; 3rd party Coupled with AutoCAD Simulation engine only; 3rd party GUIs available
GUIs available
Algorithm Surface heat transfer: CTF; zone Zone heat balance: state space Surface heat balance: CTF; zone heat balance
weighting factors method
Time step 1 hour, fixed 1 hour, fixed 1 to 60 minutes (15 minutes is used in this paper)
Weather data Hourly Hourly Hourly or sub-hourly
HVAC 28 pre-defined systems A few pre-defined systems User configurable with some limitations
User customization User functions N/A Energy management systems
Co-simulation N/A N/A External interface
Language Fortran 77 C++ Fortran 2003
Limitation Lack zone air heat balance, linear Limited user customization, linear Potentially long run-time for large models
systems systems
Licensing Free download; source code available Free download; source code not Free download; open source
open to public
Zhu et al. / Building Simulation 3

2 Methodology Previous building simulation comparisons have generally

resulted in standard test suites or case studies, which
The comparisons of the features and capabilities of twenty present the discrepancies between the simulation results,
major BEMPs were summarized to help understand their but do not address specific reasons for the discrepancies
functions, advantages and disadvantages from a simple from the view of calculation algorithms or model inputs.
overview (Crawley et al. 2008). Many efforts to test simulation In this paper, new sets of in-depth tests were designed and
results from EnergyPlus (Henninger and Witte 2011a, b, c), carried out as complement to the ASHRAE Standard 140
DeST (2006), and DOE-2.1E (IEA 1995; Sullivan 1998; tests. To further identify and understand the differences
Henninger and Witte 2006; Meldem and Winkelmann 1995) and their effects on simulation results, new test cases were
have been undertaken by their development teams. These designed by modifying inputs of the ASHRAE Standard
tests employed three testing approaches: analytical tests, 140 tests, based on deep understandings of load calculation
comparative (inter-program) tests, and empirical tests. These algorithms, modeling assumptions, and defaults of inputs
tests demonstrated that the three BEMPs had the basic of the three BEMPs. It should be noted that all test cases in
capability to simulate dynamic thermal processes and the Standard 140 were set to be continuously (24 hours per day)
energy performance of buildings. LBNL performed com- conditioned by mechanical cooling and heating systems.
parative tests between DOE-2.1E and EnergyPlus based on However, it is very common that heating or cooling is only
the test cases defined in the Alternate Calculation Method used during some specific hours, or only in specific zones
Manual of California Building Energy Efficiency Standards of a building. This means that the heat transfer between
Title 24 in order to evaluate the possibility of using EnergyPlus unconditioned and conditioned spaces needs to be accurately
as the reference simulation engine for development and accounted for in the heat balance calculation. Therefore, test
compliance calculations of future revisions of Title 24 (Huang cases with two adjacent spaces were added, and extreme
et al. 2006). A key finding was that the simulation results were thermal conditions and practical engineering conditions
highly sensitive to seemingly minor differences in inputs were applied to test possible limitations of the three BEMPs.
and model algorithms. Another comparison (Andolsun and Figure 2 summarizes the method used to develop the tests
Culp 2010) of EnergyPlus and DOE-2.1E was undertaken and perform the comparisons. For all test cases, EnergyPlus
using case studies that ranged from a sealed box to a detailed Version 7.0, DeST Version 2011-11-23 and DOE-2.1E114 were
residential building. Andolsun and Culp demonstrated that used. For all EnergyPlus tests, the CTF (conduction transfer
EnergyPlus under-estimated total building loads by 16%– function) method was used with a simulation time-step of
17% compared to DOE-2.1E as incremental loads were 15 minutes.
added, and air infiltration reduced the differences of loads
calculation results between the two BEMPs. Another study 3 Results from the ASHRAE Standard 140 tests
(Waddell and Kaserekar 2010) compared the results of
solar gains, cooling load calculations, and the transition ASHRAE Standard 140-2007 (ASHRAE 2007), Standard
from the solar gains to cooling loads, from a few BEMPs Method of Test for the Evaluation of Building Energy Analysis
including EnergyPlus, eQuest, IES, and TRACE 700. These Computer Programs, is based on the work previously per-
comparisons uncovered that large discrepancies existed formed by the International Energy Agency (IEA) under the
in the results from DOE-2.1E and EnergyPlus which were Building Energy Simulation Test (BESTEST) and Diagnostic
not well understood, nor did they explain what the key Method (IEA 1995). ASHRAE Standard 140 defines a
influencing factors could be. standard method of tests that can be used for identifying

Fig. 2 Methodology to build the tests and perform the comparisons

4 Zhu et al. / Building Simulation

and diagnosing predictive differences from whole building remaining are additional test cases (Cases 195 to 320, Cases
BEMPs that may possibly be caused by algorithmic differences, 395 to 440 and Cases 800 to 810). The outputs from these test
modeling limitations, input differences, or coding errors. So, cases were annual heating and sensible cooling loads, peak
the load calculation comparison based on ASHRAE Standard heating and cooling sensible loads, and zone air temperatures
140 tests is carried out first in our study. The results for for the cases without mechanical heating or cooling systems.
DOE-2.1E were obtained from ASHRAE Standard 140; results The annual heating loads for the low-mass building are
for EnergyPlus 7.0 were obtained from the EnergyPlus showed in Fig. 3, where each bar represents a different
development team; while results for DeST were produced BEMP, from DOE-2, BLAST (BLAST 1991), ESP, SRES/SUN,
during this study as earlier tests were limited and outdated. SERIRES, S3PAS, TRNSYS ([Link]
All test cases use single-zone models except Case 960. TASE (Aittomäki and Kalema 1976), ENERGYPLUS, or
Inputs including weather data, building construction, en- DEST. BEMPs participating in the ASHRAE Standard 140
velope materials, infiltration, internal loads, and mechanical comparison are listed in Table 2.
system are controlled in each of the three BEMPs according to One method to see how well DOE-2.1E (the output of
ASHRAE Standard 140-2007. Case 600 is the base test, Cases DOE2.1E-RevWindow), DeST, and EnergyPlus predict
610 to 650 are low-mass tests, Cases 900 to 960 are high- building loads is to see if their results fall within the range
mass tests, Cases 600FF to 950FF are free flow tests and the of spread of results from other BEMPs. Tables 3 to 7 show

Fig. 3 Annual heating loads for the 600 series low-mass tests from the ASHRAE Standard 140-2007

Table 2 BEMPs participating in the ASHRAE Standard 140 comparison (Henninger and Witte 2011b)
Code name Computer program Developer Implemented by
BLAST BLAST-3.0 level 193 v.1 CERL, U.S. NREL, U.S.; Politecnico, Torino, Italy
DOE-2.1D DOE-2.1D 14 LANL/LBNL,U.S. NREL, U.S.
ESP ESP-RV8 Strathclyde University, U.K. De Montfort University, U.K.
SRES/SUN SERIRES/SUNCODE 5.7 NREL/Ecotope, U.S. NREL, U.S.
SERIRES SERIRES 1.2 NREL, U.S. and BRE, U.K. BRE, U.K.
S3PAS S3PAS University of Sevilla, Spain University of Sevilla, Spain
TASE TASE Tampere University, Finland Tampere University, Finland
TRNSYS TRNSYS 13.1 University of Wisconsin, U.S. BRE, U.K.; Vrije Universiteit, Brussels, Belgium
DOE-2.1E DOE-2.1E LANL/LBNL, U.S. GARD Analytics, U.S., using NREL input files
DOE2.1E-RevWindow DOE2.1E-RevWindow LANL/LBNL, U.S. GARD Analytics, U.S., using Window 4 data file
which more closely matches specification
BLAST3.0-334 BLAST3.0 level 334 CERL, U.S. GARD Analytics, U.S., using NREL input files
ENERGYPLUS EnergyPlus ver.[Link],Nov 2011 U.S. Dept. of Energy GARD Analytics, U.S.
DEST DeST 2011-11-23 Tsinghua University, China Tsinghua University, China
Zhu et al. / Building Simulation 5

Table 3 Comparison of annual heating loads for various ASHRAE Table 5 Comparison of peak heating loads for various ASHRAE
Standard 140 tests Standard 140 tests
Annual heating load (MWh) Peak heating load (kW)
Case Description Min. Max. DOE-2.1E DeST EnergyPlus Case Description Min. Max. DOE-2.1E DeST EnergyPlus

600 Base case 4.296 5.709 4.994 y 5.007 y 4.364 y 600 Base case 3.437 4.354 3.767 y 3.986 y 3.732 y
610 South shading 4.355 5.786 5.042 y 5.042 y 4.398 y 610 South shading 3.437 4.354 3.755 y 3.954 y 3.720 y
620 East/west window 4.613 5.944 5.144 y 5.292 y 4.512 y1 620 East/west window
orientation orientation 3.591 4.379 3.785 y 3.962 y 3.726 y
630 East/west shading 5.050 6.469 5.508 y 5.570 y 4.813 y1 630 East/west shading 3.592 4.280 3.762 y 3.963 y 3.703 y
640 Thermostat setback 2.751 3.803 2.995 y 3.127 y 2.667 y1 640 Thermostat setback 5.232 6.954 5.656 y 5.991 y 6.265 y
650 Night ventilation 0.000 0.000 0.000 y 0.000 y 0.000 y 650 Night ventilation 0.000 0.000 0.000 y 0.000 y 0.000 y
900 High-mass base case 1.170 2.041 1.301 y 1.894 y 1.163 y1 900 High-mass base case 2.850 3.797 3.248 y 3.600 y 3.140 y
910 High-mass south 1.512 2.282 1.559 y 2.266 y 1.427 n 910 High-mass south
shading shading 2.858 3.801 3.256 y 3.612 y 3.139 y
920 High-mass east/west 3.261 4.300 3.312 y 4.025 y 3.087 n 920 High-mass east/west
window orientation window orientation 3.308 4.061 3.508 y 3.776 y 3.453 y
930 High-mass east/west 4.143 5.335 4.249 y 4.485 y 3.785 n 930 High-mass east/west
shading shading 3.355 4.064 3.536 y 3.801 y 3.475 y
940 High-mass thermostat 0.793 1.411 0.838 y 1.270 y 0.727 n 940 High-mass thermostat
setback setback 3.980 6.428 5.322 y 5.723 y 4.785 y
950 High-mass night 0.000 0.000 0.000 y 0.000 y 0.000 y 950 High-mass night
ventilation ventilation 0.000 0.000 0.000 y 0.000 y 0.000 y
960 Sunspace 2.144 3.373 2.216 y 2.835 y 2.322 y 960 Sunspace 2.410 2.863 2.603 y 2.601 y 2.691 y

Table 4 Comparison of annual cooling loads for various ASHRAE Table 6 Comparison of peak cooling loads for various ASHRAE
Standard 140 tests Standard 140 tests
Annual cooling load (MWh) Peak cooling load (kW)
Case Description Min. Max. DOE-2.1E DeST EnergyPlus Case Description Min. Max. DOE-2.1E DeST EnergyPlus

600 Base case 6.137 8.448 8.054 y 5.924 y1 7.006 y 600 Base case 5.965 7.188 6.965 y 6.151 y 6.678 y
610 South shading 3.915 6.139 5.874 y 4.873 y 4.976 y 610 South shading 5.669 6.673 6.482 y 5.964 y 6.274 y
620 East/west window 3.417 5.482 5.256 y 3.847 y 4.384 y 620 East/west window
orientation orientation 3.634 5.096 4.679 y 3.819 y 4.005 y
630 East/west shading 2.129 3.701 3.235 y 2.879 y 2.952 y 630 East/west shading 3.072 4.116 3.834 y 3.270 y 3.446 y
640 Thermostat setback 5.952 8.097 7.713 y 5.759 y1 6.710 y 640 Thermostat setback 5.884 7.126 6.903 y 6.116 y 6.614 y
650 Night ventilation 4.816 7.064 6.678 y 4.625 y1 5.538 y 650 Night ventilation 5.831 7.068 6.843 y 5.973 y 6.479 y
900 High-mass base case 2.132 3.669 3.390 y 2.296 y 2.683 y 900 High-mass base case 2.888 3.932 3.778 y 3.469 y 3.320 y
910 High-mass south 0.821 1.883 1.738 y 1.202 y 1.350 y 910 High-mass south
shading shading 1.896 3.277 2.703 y 2.953 y 2.640 y
920 High-mass east/west 1.840 3.313 3.169 y 2.401 y 2.683 y 920 High-mass east/west
window orientation window orientation 2.385 3.505 3.342 y 2.844 y 2.835 y
930 High-mass east/west 1.039 2.238 1.823 y 1.696 y 1.745 y 930 High-mass east/west
shading shading 1.873 3.080 2.638 y 2.527 y 2.332 y
940 High-mass thermostat 2.079 3.546 3.272 y 2.262 y 2.606 y 940 High-mass thermostat
setback setback 2.888 3.932 3.778 y 3.497 y 3.320 y
950 High-mass night 0.387 0.921 0.749 y 0.455 y 0.571 y 950 High-mass night
ventilation ventilation 2.033 3.170 2.917 y 2.586 y 2.451 y
960 Sunspace 0.411 0.895 0.855 y 0.537 y 0.732 y 960 Sunspace 0.953 1.422 1.048 y 1.085 y 1.213 y
6 Zhu et al. / Building Simulation

Table 7 Comparison of hourly zone temperatures for various ASHRAE Standard 140 tests
Maximum annual hourly zone temperature (℃)
Min. Max. DOE-2.1E DeST EnergyPlus
600FF Base case 64.90 75.10 73.40 y 65.49 y 66.03 y
650FF Night ventilation 41.81 46.40 45.50 y 42.39 y 43.65 y
900FF High-mass base case 63.24 73.50 71.70 y 63.67 y 64.31 y
950FF High-mass base case 35.54 38.50 37.10 y 35.67 y 36.90 y
960FF Sunspace 48.88 55.34 51.60 y 55.54 y1 52.93 y
Minimum annual hourly zone temperature (℃)
Min. Max. DOE-2.1E DeST EnergyPlus
600FF Base case 18.80 15.57 17.70 y 18.60 y 17.51 y
650FF Night ventilation 6.38 1.65 2.00 y 4.50 y 2.39 y
900FF High-mass base case 23.00 21.10 21.00 y 22.91 y 23.08 y1
950FF High-mass base case 20.20 17.80 17.80 y 19.97 y 20.34 y1
960FF Sunspace 2.82 5.80 6.00 y1 0.48 y 2.44 y
Average annual hourly zone temperature (℃)
Min. Max. DOE-2.1E DeST EnergyPlus
600FF Base case 24.22 27.40 24.43 y 26.19 y
650FF Night ventilation 24.45 27.50 24.45 y 26.40 y
900FF High-mass base case 17.99 20.80 17.81 y 18.87 y
950FF High-mass base case 14.00 15.30 13.88 y1 14.62 y
960FF Sunspace 26.43 30.50 29.92 y 29.51 y

the results of the three BEMPs with an extra column for each difference, even though the absolute difference is not the
of the three BEMPs to indicate whether its results fall within largest (2.130 MWh in Case 600). For peak heating and
the ranges. The columns of the Min. and Max. represent cooling loads in Tables 3 and 4, the results from the three
respectively the minimum and maximum results from all of BEMPs all fall within the ranges, but there are still large
the tested BEMPs except DOE-2.1E, DeST and EnergyPlus. difference of 16.4% in Case 940 peak heating loads and 18.4%
An indicator of “y” means that the test results are within the in Case 620 peak cooling loads.
[Min., Max.] range, a “y1” means the results are not within Even though test cases from ASHRAE Standard 140 are
the [Min., Max.] range but are within the 5% relaxed range very simple, there are still large discrepancies in the results
[Min./1.05, Max. × 1.05], while an “n” means the results are from the three BEMPs, especially the annual heating loads
outside of the relaxed range [Min./1.05, Max. × 1.05]. for the high-mass test cases. The ASHRAE Standard 140 tests
From the comparison of the ASHRAE Standard 140 test did not provide adequate details to explain or isolate the
results, we can conclude that the simulation results from influencing factors that drive the discrepancies.
the three BEMPs mostly fall within the ranges except the
heating loads results for the high-mass cases 910, 920, 930, 4 In-depth tests
and 940, where EnergyPlus calculated smaller annual heating
loads than DOE-2.1E and DeST. EnergyPlus results for In this part, new sets of in-depth tests were carried out as
these four cases are about 10% lower than the minimum of complement to the ASHRAE Standard 140 tests. Firstly,
the ranges. The largest percent differences are in the annual influencing factors that drive the differences between
heating loads from test case 940 high-mass with thermostat EnergyPlus and DeST are isolated by modifying inputs of
setback, where EnergyPlus gave the lowest annual heating the ASHRAE Standard 140 tests. DOE-2.1E (DOE-2 1982)
loads of 0.727 MWh while DeST gave the highest result of lacks strict heat balance calculations of zone and surfaces.
1.27 MWh, a 42.7% difference, even though the absolute It calculates the interior surface heat exchange by using the
difference is not the largest (0.938 MWh in Case 920). The combined convective and radiative heat transfer coefficients,
largest percent differences are in the annual cooling loads rather than calculating the convection from surfaces to
from test case 950 high-mass night ventilation, where DOE- zone air, and long-wave radiation between interior surfaces
2.1E gave the highest annual cooling loads of 0.945 MWh separately. Thus, input values for DOE-2.1E cannot be
while DeST gave the lowest result of 0.455 MWh, a 39.3% matched well with EnergyPlus and DeST, and in Section
Zhu et al. / Building Simulation 7

4.1 EnergyPlus and DeST are compared by in-depth tests.

Secondly, double-zone cases were added up to test the ability
of the three BEMPs calculating the heat balance calculation
accurately when heating or cooling is only used during some
specific hours, or only in specific zones of a building in
Sections 4.2 and 4.3. As DOE-2.1E uses the adjacent space
temperature from the previous time step to calculate the heat
transfer from adjacent spaces, some errors may come up if
two adjacent spaces are not both conditioned, or if there is
a large temperature difference between two adjacent spaces.
Fig. 4 Hourly exterior surface convection coefficients of the south
Thus, these comparisons are very useful for the three BEMP’s
wall from Case 195 in ASHRAE Standard 140
applications in practical engineering conditions.

4.1 Control input values based on ASHRAE Standard

140 tests

4.1.1 Specification of test cases

A series of tests were designed to identify the influence

of different modeling assumptions between DeST and
EnergyPlus, as shown in Table 8. When using default values
or algorithms in EnergyPlus and DeST, there are large
discrepancies in the surface convection coefficients, as shown
Fig. 5 Hourly interior surface convection coefficients of the south
in Figs. 4 and 5. DeST assumes constant values of convection
wall from Case 195 in ASHRAE Standard 140
coefficients for both the inside and outside surfaces (DeST
2006), while EnergyPlus calculates these coefficients using
correlations with indoor and outdoor air temperatures and exchange between interior surfaces, and constant outdoor
wind speeds (EnergyPlus 2011). In all test cases except C10, air temperature. C3 is the same as Case 195 except the surface
the exterior and interior surfaces convection coefficients convection coefficients are set to the constant value used in
for EnergyPlus were set to be the same constant values as DeST. Based on C3, other influencing factors such as internal
those used in DeST. gains, air infiltration, thermostat control strategy, and a
C1 is the simplest case and can be calculated analytically. south-facing window were added step by step to build the
It assumes no solar absorption, no long-wave radiant final C10 test case.

4.1.2 Results and discussions

Table 8 In-depth test cases
Case Description Annual heating and cooling loads for the ten test cases are
Case 195 + suspended in outdoor air + solar/visible absorptance and
summarized in Figs. 6 and 7.
C1
infrared emissivity equal zero + the outdoor air temperature was
always kept at 10℃
C2 C1+the outdoor air temperature was exactly the same as case 195
C3 Case 195 (surfaces convection coefficients for EnergyPlus were set
as DeST defaults)
C4 C3 + solar absorptance equals 0.6
C5 C4 + infrared emissivity equals 0.9
C6 C5 + internal gains
C7 C6 + infiltration
C8 C7 + the thermostat control strategy of Case 600
C9 C8 + south windows of Case 600
C10 C9 + default algorithms for surfaces convection coefficients in Energy-
Fig. 6 Annual heating loads of test Cases C1 to C10 from EnergyPlus
Plus (exactly the same as Case 600 in ASHRAE Standard 140)
and DeST
8 Zhu et al. / Building Simulation

Fig. 7 Annual cooling loads of test Cases C1 to C10 from EnergyPlus

and DeST Fig. 9 Solar transmittance at various incident angles

The hourly heating load of test case C1 can be calculated

analytically to obtain the result of 0.50 kW, as Table 9
illustrates. The hourly results from EnergyPlus and DeST
are both 0.50 kW, matching the analytical one well.
Case C2 considers the hourly variation of outdoor air tem-
perature and the simulation results indicate little differences
between DeST and EnergyPlus, as shown in Fig. 8.
Comparing the simulation results from EnergyPlus and
DeST for Cases C1 to C9, we find that the largest discrepancy
occurred after south-facing windows were added. The solar
transmittance at different incident angles for the double-
pane window are almost the same in DeST and EnergyPlus, Fig. 10 Annual solar radiation on the south wall and the annual
which are also close to the values provided in ASHRAE window-transmitted solar
Standard 140, as shown in Fig. 9. This indicates that the
window algorithm should cause little difference between that the annual solar radiation on the south wall from DeST
the results from DeST and EnergyPlus. However, when com- is about 5.1% smaller than from EnergyPlus, while the
paring the solar radiation on the southern exterior surface window-transmitted solar radiation is 7.0% smaller. This
and window-transmitted solar as shown in Fig. 10, we find helps explain why cooling loads from DeST are lower than
EnergyPlus as less solar radiation enters the room.
Table 9 The UA calculation of Case C1 The main reason for this discrepancy is the time point
h_out R h_in Area U-factor UA used for the solar position calculation and the sky diffuse
(W/(m2·K)) ((m2·K) /W) (W/(m2·K)) (m2) (W/(m2·K)) (W/K) solar radiation model. DeST (2006) uses the beginning of
Light wall 23.3 1.789 3.5 75.6 0.47 35.70 the hour for solar calculations during the hourly time step
Light floor 0 25.254 4 48 0.04 1.88 simulations, while EnergyPlus (2011) performs interpolation
Light roof 23.3 2.993 1 48 0.25 11.89 of solar radiation data from weather input files, and uses the
Total 49.48 end of each sub-hourly time step for the solar calculations.
For the sky diffuse solar radiation model, EnergyPlus takes
into account the anisotropic radiance distribution of the
sky, while DeST uses an isotropic model.
Comparing the results between C9 and C10, we can
find that another main influencing factor is the surface
convection coefficients. EnergyPlus calculates the coefficients
considering the variations of indoor and outdoor conditions,
while DeST uses constant values as required by the linear
system structure of its state-space heat balance method. As
both exterior and interior surface convection coefficients are
smaller in EnergyPlus than in DeST, EnergyPlus produces
lower annual heating loads and higher annual cooling loads.
Fig. 8 Hourly heating loads for Case C2 In conclusion, the differences between annual heating or
Zhu et al. / Building Simulation 9

cooling loads from EnergyPlus and DeST can be controlled

below 10% if inputs including default values or algorithms are
matched, although EnergyPlus and DeST have difference in
their load calculation algorithms and modeling assumptions.
Therefore, matching inputs are the key when using different
BEMPs.

4.2 Double-zone cases under extreme thermal conditions

4.2.1 Specification of test cases

DOE-2.1E has limitation in the heat balance calculation, Fig. 12 Temperature settings of Zone 1
especially multiple zones under different thermal conditions.
To see how this limitation affects DOE-2.1E simulation results,
Case EC1 was designed for the three BEMPs. A building has
two adjacent rectangular spaces each with dimensions 10 m
wide × 10 m long × 3 m high, as shown in Fig. 11. Zone 1 is
conditioned with a special thermostat setting while Zone 2
is un-conditioned.
The construction of the exterior walls, roof and floor
are the same as Case 600 and the interior walls are the same
as Case 960 in ASHRAE Standard 140 tests. All solar/visible
absorptance and thermal emissivity coefficients are set to
zero, so only convective heat transfer between the outdoor air
and the two indoor zones is considered. Surface convection Fig. 13 Simulated temperature of Zone 2
coefficients are specified as the same constant values for
DOE-2.1E, EnergyPlus, and DeST. The outdoor air tem- zone temperature from the previous time step to calculate
perature and ground temperature are always kept at 10℃. the heat transfer between adjacent zones, and the heat flow
Each zone has no internal gains or air infiltration. through the interior walls and partitions is treated as
steady-state.
4.2.2 Results and discussion

The air temperature of Zone1 varied periodically (switched 4.3 Double-zone cases under practical engineering
between 29.8℃ and 16.2℃ hourly) all year round controlled conditions
by a scheduled air-conditioning system, as shown in Fig. 12.
The air temperature of Zone 2 was then calculated, as The results from Case EC1 implied that DOE-2.1E has
shown in Fig. 13, which shows that the results from DeST limitations in accurately calculating heat transfer between
and EnergyPlus were always constant, but DOE-2.1E gave adjacent zones. Further tests were designed to ascertain the
fluctuating air temperatures between 13.8 and 14.4℃. This influence of this under practical engineering conditions.
is mainly due to the fact that DOE-2.1E uses the adjacent
4.3.1 Specification of test cases

Three new tests were created. Case SC1 is the base test and
its results are used as the baseline for Cases SC2 and SC3
(Table 10). The building model of all the three cases includes
two zones and each zone has dimensions 10 m wide × 10 m
long × 3 m high (Fig. 14), with a window area of 12 m2 on
the south facade.
Weather data was the same as for Case 600. The con-
structions of the exterior walls, roof and floor, as well as the
properties of the double-pane window, were the same as for
Case 600. The interior wall was the same as for Case 960 in
ASHRAE Standard 140 tests. Infiltration was always 0.5 air
Fig. 11 Building model of Case EC1 changes per hour, DeST and EnergyPlus use constant values,
10 Zhu et al. / Building Simulation

Table 10 Test cases under practical engineering conditions Table 11 Internal heat gains for Case SC1
Case Description DOE-2.1E DeST EnergyPlus
2 2
SC1 Base test, both zones are conditioned 24 hours a day 0.1 person/m 0.1 person/m 0.1 person/m2
SC2 Zone 1 with office daytime occupancy (Fig. 15); Zone 2 empty Sensible heat: 66 W Sensible heat: 66 W Active level: 137 W
People
and unconditioned Latent heat: 71 W Latent heat: Sensible heat
0.102 kg/h fraction: 0.48
SC3 Zone 1 with bedroom nighttime occupancy (Fig. 16); Zone 2
empty and unconditioned People heat Default weighting Default distribution Same as DeST
distribution factors
Lighting 10 W/m2 10 W/m2 10 W/m2
Lighting heat Default lighting type Default distribution Same as DeST
distribution
Equipment 5 W/m2 5 W/m2 5 W/m2
Equipment heat Default weighting Default distribution Same as DeST
distribution factors

and time step were set to the default values or algorithms in

each BEMP.
DeST and EnergyPlus use ideal air systems and DOE-
2.1E uses a “two pipe fan coil” system. The system used in
Fig. 14 Building model of Case SC1
DOE-2.1E was set to be a 100% convective air system, 100%
efficient with no duct losses, and with adequate cooling and
heating capacities, which is very close to an ideal air system.
The air systems were always on and the zone thermostat
set-point for Case SC1 was always 21.1℃. In Cases SC2 and
SC3, the set-point for heating was 20℃ and for cooling was
27℃. SC2 was exactly the same as SC1 except that Zone 1
used an office daytime schedule and Zone 2 was empty and
unconditioned (no internal gains).
SC3 is exactly the same as SC2 except that Zone 1 used
Fig. 15 Office daytime schedule used in SC2 a bedroom nighttime schedule.

4.3.2 Results and discussion

In DOE-2.1E, hourly heating or cooling loads are calculated

by the LOADS subprogram, first assuming constant space
temperature, then adjusted by the SYSTEMS subprogram
to consider the thermostat settings, outdoor air ventilation,
etc. The loads from the SYSTEMS subprogram are used in
our study.
Comparing the results of different BEMPs from SC1 to
SC3, heating loads were always close but cooling loads were
Fig. 16 Bedroom nighttime schedule used in SC3 not. From the results of SC1 (Fig. 17), the monthly cooling
loads of the three BEMPs were very close. Comparing SC1
while DOE-2.1E uses the AIR-CHANGES/HR method which with SC2 and SC3 (Figs. 18 and 19), we can see that the
calculates the infiltration based on the wind speed. Internal monthly cooling loads of DeST and EnergyPlus were always
heat gains from people, lighting, and equipment were close, but the results of DOE-2.1E deviate more significantly.
assumed as constant. Table 11 lists the input assumptions for In SC1 there was no heat transfer between adjacent zones,
the internal heat gains. The inputs for “people heat level” so its results can be used as the benchmark where the biggest
were equivalent for the three BEMPs though their input difference between annual cooling loads from DOE-2.1E
values were not the same. and EnergyPlus was 10.3%. In SC2 the annual cooling load
Solar and visible absorptances were set to 0.6, thermal of DOE-2.1E was 35.0% higher than DeST and 18.2% higher
emissivity to 0.9, and ground reflectance to 0.2 in the three than EnergyPlus. In SC3 the annual cooling loads were very
BEMPs. Surface convection coefficients, solar distribution, small compared to SC1 and SC2 (Fig. 20), so the percentage
Zhu et al. / Building Simulation 11

differences, even though very high, are not very meaningful.

These results reveal that DOE-2.1E has serious limitations
in accurately accounting for multi-zone heat balance and
part-time operation of HVAC systems, especially for the
nighttime air-conditioning case SC3.
From the monthly heating results (Figs. 21 to 23) from
the three BEMPs, DeST and EnergyPlus always have close
simulation results across all three test cases, but DOE-2.1E’s
results are always lower, mainly due to the differences in
default values and algorithms used. For all three cases SC1
to SC3, the annual heating loads (Fig. 24) from DOE-2.1E
Fig. 17 Monthly sensible cooling loads of SC1 were about 20% lower than those from EnergyPlus or DeST.

Fig. 18 Monthly sensible cooling loads of SC2 Fig. 21 Monthly heating loads of SC1

Fig. 19 Monthly sensible cooling loads of SC3 Fig. 22 Monthly heating loads of SC2

Fig. 20 Annual sensible cooling loads from SC1 to SC3 Fig. 23 Monthly heating loads of SC3
12 Zhu et al. / Building Simulation

for the solar calculations. Comparing the hourly result of

different methods with the standard value, we finally chose
the middle of the hour and the hourly time step considering
the accuracy and computation speed for DeST in version
2013-01-15. The differences of load results between DeST
and EnergyPlus will be reduced as the solar radiation on
exterior surfaces has smaller discrepancy now. One major
change in EnergyPlus from version 7.0 to 7.2, directly related
to loads calculation, is the correction of exterior surface
convection coefficients for windows, including the use of
Fig. 24 Annual heating loads from SC1 to SC3 near-window wind speeds rather than the weather station
wind speeds and the use of new coefficients for the empirical
5 Discussions and conclusions correlation (Booten et. al 2012). This change to EnergyPlus
increases heating loads, which will reduce the discrepancy
This study compared the capability and simulation results in heating loads between EnergyPlus and DeST.
of EnergyPlus, DeST and DOE-2.1E for performing building Although the study provides new methods and results in
thermal load calculations, to evaluate the impact of the comparing simulation programs, it serves as a good starting
different simulation engines. Algorithms used to calculate point for a much bigger study that is needed to fully justify
the thermal loads play an important role, but keeping inputs which program is better, or under which situation, users
to energy models exactly the same or equivalent is even more should use which programs.
important to guarantee consistent results from different This study covers the loads comparison between
BEMPs. All the three BEMPs have the basic capability for EnergyPlus, DeST and DOE-2.1E. A future paper will discuss
performing building load calculations, and the discrepancy in the methods and findings from comparing the HVAC models
load results from DeST and EnergyPlus was reduced to less across the three programs.
than 10% if the surface convection coefficients were set exactly
the same as those in ASHRAE Standard 140 test cases. Acknowledgements
To differentiate this study from previous work, it did
not produce another test suite, but rather a methodology to This study was a joint effort between Tsinghua University,
design tests and a process to carry out these tests. A few China and Lawrence Berkeley National Laboratory, USA
in-depth tests, built upon the ASHRAE Standard 140 tests, under the US-China Clean Energy Research Center on
were designed and performed to identify and quantify Building Energy Efficiency. It was co-sponsored by the Energy
the key influencing factors that drive the discrepancies in Foundation under the China Sustainable Energy Program.
results from EnergyPlus, DeST and DOE-2.1E. It was verified
that DOE-2.1E has limitations in handling heat transfer References
between adjacent zones, with large errors emerging for
cases when adjacent zones have very different conditions, Aittomäki A, Kalema T (1976). TASE—A Computer Program for
or if a zone is part-time conditioned while adjacent zones Energy Analysis of Buildings. Technical Research Centre of Finland
are unconditioned. (VTT), Laboratory of Heating and Ventilating. (in Finnish)
Andolsun A, Culp CH (2010). A comparison of EnergyPlus to DOE-
The simulations done in the study were based on
2.1E: Multiple cases ranging from a sealed box to a residential
EnergyPlus version 7.0 and DeST version 2011-11-23. The
building. In: Proceedings of SimBuild, New York, USA.
latest available is EnergyPlus 7.2 and DeST 2013-01-15. The
ASHRAE (2007). ANSI/ASHRAE Standard 140-2007, Standard
solar position algorithm in DeST is improved in the new
Method of Test for the Evaluation of Building Energy Analysis
version, which is an important feedback from this com-
Computer Programs. Atlanta, USA: American Society of Heating,
parison study. DeST V2011-11-23 used the beginning of Refrigerating and Air-Conditioning Engineers.
the hour for solar calculations during the hourly time step BLAST (1991). BLAST User Reference, Vols. 1, 2. Urbana, USA:
simulation. We got the measured direct normal solar University of Illinois.
radiation per minute from the Beijing Weather Station and Booten C, Kruis N, Christensen C (2012). Identifying and Resolving
calculated the direct solar radiation on east/west/south/north Issues in EnergyPlus and DOE-2 Window Heat Transfer
facing surface per minute as well as accumulated value per Calculations, NREL/TP-5500-55787.
hour as the standard value. Then we tried different points Crawley DB, Hand JW, Kurnment M, Griffith BT (2008). Contrasting
(the beginning, the middle and the end) of the hour during the capabilities of building energy performance simulation
the hourly time step and also different sub-hourly time steps programs. Building and Environment, 43: 661  673.
Zhu et al. / Building Simulation 13

Crawley DB, Lawrie LK, Winkelmann FC, Buhl WF, Huang YJ, Henninger RH, Witte MJ (2011b). EnergyPlus Testing with Building
Pedersen CO, Strand RK, Liesen RJ, Fisher DE, Witte MJ, Glazer Thermal Envelope and Fabric Load Tests from ANSI/ASHRAE
J (2001). EnergyPlus: Creating a new-generation building energy Standard 140-2007. Available: [Link]
simulation program. Energy and Buildings, 33: 319  311. energyplus/energyplus_testing.cfm. Accessed Nov. 2011.
DeST (2006). Simulation Method for Building Thermal Environment- Henninger RH, Witte MJ (2011c). EnergyPlus Testing with HVAC
DeST. Beijing: China Architecture and Building Press, 2006. (in Equipment Component Tests. Available: [Link]
Chinese) buildings/energyplus/energyplus_testing.cfm. Accessed Nov. 2011.
DOE-2 (1980). DOE-2 Reference Manual Version 2.1, LBL-7689-M Huang J, Bourassa N, Buhl F, Erdem E, Hitchcock R (2006). Using
Ver.2.1, LBL-8706 Rev.1. EnergyPlus for California title-24 compliance calculations. In:
DOE-2 (1982). DOE-2 Engineers Manual Version 2.1A, LBL-11353. Proceedings of SimBuild, Cambridge, MA, USA.
EnergyGauge (2012). EnergyGauge User Manual 2010. Florida Solar IEA (1995). Building Energy Simulation Test (BESTEST) and Diagnostic
Energy Center, USA. Method. National Renewable Energy Laboratory, USA.
EnergyPlus (2011). Engineering Reference Version 7.0 Documentation. Meldem R, Winkelmann F (1995). Comparison of DOE-2 with
University of Illinois and Ernest Orlando Lawrence Berkeley Measurements in the Pala Test House, LBL-37979. Lawrence
National Laboratory, USA. Berkeley National Laboratory, USA.
EnergyPro (2011). EnergyPro User’s Manual, Version 5, July, 2011. Sullivan R (1998). Validation Studies of the DOE-2 Building Energy
EnergySoft, LLC, USA. Simulation Program, LBNL-42241. Lawrence Berkeley National
eQuest (2009). eQuest Introductory Tutorial, Version 3.63, April, 2009. Laboratory, USA.
James J. Hirsch & Associates, USA. VisualDOE (2004). VisualDOE 4.0 User Manual, March, 2004.
ESRU (1999). ESP-r: A building and plant energy simulation environ- Architectural Energy Corporation, USA.
ment, User Guide Version 9 Series. ESRU Publication, University Waddell C, Kaserekar S (2010). Solar gain and cooling load comparison
of Strathclyde, USA. using energy modeling software. In: Proceedings of SimBuild,
Henninger RH, Witte MJ (2006). Test Report of LBNL DOE-2.1E119 New York, USA.
Based on ANSI/ASHRAE Standard 140-2004. Lawrence Berkeley Yan D, Xia J, Tang W, Song F, Zhang X, Jiang Y (2008). DeST—An
National Laboratory, USA. integrated building simulation toolkit, Part Ⅰ: Fundamentals.
Henninger RH, Witte MJ (2011a). EnergyPlus Testing with ASHRAE Building Simulation, 1: 95  110.
1052-RP Toolkit—Building Fabric Analytical Tests. Available: Zhang X, Xia J, Jiang Z, Huang J, Qin R, Zhang Y, Liu Y, Jiang Y
[Link] (2008). DeST—An integrated building simulation toolkit, Part Ⅱ:
[Link]. Accessed Nov. 2011. Applications. Building Simulation, 1: 193  209.