MSc in Electrical Installation

Research Proposal

Demand Side management and AI assistant load forecasting for

commercial installations.

W.L.M Fernando 239908T

Faculty of Engineering
University of Moratuwa
May 2025
 TABLE OF CONTENTS

TABLE OF CONTENTS ..........................................................................................................3

LIST OF FIGURES ..................................................................................................................4
LIST OF TABLES ....................................................................................................................4
LIST OF ABBREVIATIONS ...................................................................................................5
Chapter 1: Introduction ......................................................................................... 7
1.1 Tentative Title of the Thesis ...........................................................................................7
1.2 Background of Industrial Power Consumption ..............................................................7
1.3 Areas to be studied..........................................................................................................7
Chapter 2: Literature Review ................................................................................ 8
Chapter 3: Aim and Objectives ............................................................................ 13
Chapter 4: Methodology ....................................................................................... 14
Chapter 5: Expected Outcomes ............................................................................ 18
AI model to forecast the load profile ......................................................................................18
Chapter 6: Proposed Time Line ........................................................................... 19
Chapter 7: References .......................................................................................... 20
 LIST OF FIGURES

Figure 3.1 Proposed Methodology ........................................................................ 14

LIST OF TABLES

Table 2.1: Summery of the literature review ......................................................... 12

 LIST OF ABBREVIATIONS

AI Artificial Intelligence

ANN Artificial Neural Network

CEB Ceylon Electricity Board

DER Distributed Energy Resource

DSM Demand Side Management

DR Demand Response

GDP Gross Domestic Product

HVAC Heating, Ventilation, and Air Conditioning

ICT Information and Communication Technology

LSTM Long Short-Term Memory (Neural Network)

ML Machine Learning

MAPE Mean Absolute Percentage Error

PV Photovoltaic

RNN Recurrent Neural Network

TAR Time-of-Use Tariff Adjustment Ratio

TOU Time-of-Use

TS Time Series

T&D Transmission and Distribution

VRE Variable Renewable Energy

Chapter 1: Introduction

1.1 Tentative Title of the Thesis

‘Demand Side management and AI assistant load forecasting for commercial

installations’.

1.2 Background of Industrial Power Consumption

The industrial sector is a major consumer of energy, with key industries like cement,
steel, and chemicals using large amounts of electricity and thermal energy. This high
demand places pressure on the national grid, especially during peak hours. Demand
Side Management (DSM) plays a crucial role in addressing these challenges by
encouraging industries to optimize their energy use patterns. Through DSM
strategies—such as load shifting, process optimization, and adoption of energy-
efficient technologies—industries can reduce peak demand, lower operational costs,
and improve overall system reliability. The report [1] emphasizes that by
benchmarking energy performance and implementing DSM measures, industries not
only enhance their energy efficiency but also contribute to a more stable and
sustainable energy future.

1.3 Areas to be studied

In recent years, the load profile of commercial and industrial electricity consumers has
changed significantly. One of the primary factors behind this shift is the increased
overall demand for electricity. However, to fully understand this transformation, it is
essential to investigate several interconnected factors that influence load behaviour in
commercial installation. This study will explore the following areas:

o Increment of demand of electricity in commercial sectors

o Shifts in consumption patterns across different types of commercial
users
o Economic growth of the country and structure of the economy
These changes are directly relevant to DSM, as understanding when and how loads
vary can help optimize energy usage and reduce peak demand. The study aims to
analyse the causes of variations between day and night peak loads and assess whether
a simplified load profile with a more predictable daytime peak could support more
effective DSM strategies. This simplification could lead to both operational
efficiencies and economic benefits for utilities and end users.

The vital purpose of this study is to come across with the reasons for the changes of
day peak and night peak variations and study the feasibility of make load profile
simpler than available load curve by observing benefit of significant change of the day
time peak increment. And also the study will focusing how this change can be make
economically beneficial for the country as well. And also the studies will focus on the
AI modelling of load forecasting in order to develop an accurate AI model for load
forecasting.

Chapter 2: Literature Review

The paper [2] is reviewing that industrial DSM measures are primarily driven by peak
load reduction , energy cost minimization , integration of renewable energy sources
and improved operational flexibility . The various DSM techniques cited in the
literature include, Load shifting and shedding, Time of use pricing schemes, Demand
response programs and Process optimization using advanced control systems. DSM
implementation in industries is heavily influenced by market structure, regulatory
frameworks, and the availability of automated systems.

Load forecasting is essential for DSM to ensure that energy demands are predicted
accurately. The reviewed work emphasizes the integration of AI techniques to enhance
forecasting accuracy in industrial environments. Key AI models and methods
discussed include that Artificial Neural Networks, Support Vector Machines, Fuzzy
Logic Systems and Decision Trees and Hybrid models. AI based forecasting provides
real time adaptability, predictive maintenance support, and smarter DSM execution.
These models also facilitate intelligent decision making for energy scheduling and
automation in industrial process [2].
This paper identified several challenges in implementing DSM with AI models. They
are Data quality and availability in industrial settings, Integration of legacy systems
with modern AI tools, Need for domain specific customization of forecasting models
and Limited real world testing and deployment case studies [2].

The report [3] shows this perspective by framing industrial demand side flexibility as
a cornerstone of a just and inclusive energy transition. It asserts that industrial DSM
should not only focus on efficiency and peak management but also serve as a platform
for enabling renewable integration and economic resilience. The report identifies that
policy support, equitable market access, digitalization and financial mechanisms are
essential to scale DSM across sectors. Notably , it stress the importance of including
small and medium sized enterprises in DSM frameworks, which are often excluded
due to high investment barriers and limited digital readiness [3] .

Moreover , [3] underscores the importance of automated , AI driven systems in

achieving reliable DSM , particularly in contexts with variable production cycle .

This paper [4] introduced the typical demand response and characteristic of industrial
demand-side management. A review of the most important features of the industrial
demand-side management has been presented. Planning and scheduling for industrial
demand-side management under the environment of Energy Internet and Industrial
Inter net has been described in detail. Finally, the policy guarantee and market
mechanism of industrial demand-side management has been designed in this paper.
The construction of DRM is the basic of market design. The DR market will be
operated greatly with the coordination of the ASM, SM and CM. All in all, in the
background of deep integration of Energy Internet and Industrial Internet, facing the
existing challenges in demand response of power system, the authors of this paper
believe in the following solutions. Firstly, we must research the production data-driven
industrial load controllable characteristic modelling method, explore the coordination
mechanism of multiple types of industrial loads that adapt to multiple regulation
scenarios of the power grid, propose real-time industrial load control methods for real-
time regulation needs of the power grid, and develop new low-cost technologies to
realize the cooperative and interactive response of source-grid- load and storage.
Secondly, we must develop multi-type demand response resources combined with
cross-disciplinary deep integration means such as artificial intelligence, build a data-
driven industry-wide demand-side adjustable resource pool, and realize accurate two-
way sensing and interaction between power grid and load. This paper [4] can guide a
promising path for the vigorous development of industrial demand-side management
in the future.

This paper [5] provides a comprehensive analysis of the different technologies,

approaches used in DSM as well as the impact of distributed renewable energy
generation and storage technologies in SG. The main goal of these methods is to
decrease peak load demands and achieve advanced synchronization between network
operators and customers via the development and application of power-saving
technologies, financial incentives, the price of energy, and government rules. This
research thoroughly investigated DSM implementation issues that must be overcome
for DSM to be effectively integrated into the SG with some proposed solutions, DSM
optimization methodologies, and their related solutions, which were not included in
the earlier review article. As a consequence, a comprehensive comparison of many
algorithms used in DSM optimization problems is provided in terms of a variety of
factors such as energy cost reduction, PAR, waiting time, power scheduling, Voltage
limitations, DR, risk management, client privacy, and carbon emission. We
determined, after examining multiple DSM-based research, that a single strategy is not
the best solution to handle the high complexity of the DSM optimization problem due
to its poor performance and low convergence rate.
As a consequence, hybrid algorithms may outperform single algorithms in terms of
convergence rate, complexity, noisy environment, imprecision, uncertainty, and
ambiguity.
Furthermore, these tactics may be improved in the future to improve SG’s efficiency
by balancing supply and demand. Even though these current breakthroughs in the use
of optimization techniques in DSM are widely known, extra research is undoubtedly
necessary to discover the optimal solutions in many real-world scenarios.
The power system’s functioning will become more difficult if corrective control is
used. This is just another obstacle to the adoption of DSM. Yet, given that adaptability
is increasingly seen as a key tool for coping with the unpredictability of future
developments, together with the ongoing cost reductions of DSM technologies, it is
anticipated that DSM will become noticeably more competitive in the near future.
Increasing trust in the employment of DSM schemes for the provision of system
security will benefit from the establishment of targeted trial schemes. This
comprehensive review of DSM will assist all researchers in this field in improving
energy management strategies and reducing the effects of system uncertainties,
variances, and restrictions [5].
Paper Content Gap
No:
[2]  Developed a deterministic DSM  Lack of real-time adaptability:
model that incorporates time-of- deterministic models struggle with real-
use tariffs and on-site generation. time adjustments in load and renewable
 Focused on linear programming generation uncertainty.
to optimize load schedules for  No integration with AI-based forecasting,
cost reduction and grid reliability. limiting accuracy in dynamic
 Applied in a case study of a environments.
factory with peak load shifting
and energy cost minimization.
[3]  Reviewed barriers to adopting  Scarcity of empirical work combining AI
DSM in factories, including with deterministic scheduling in
regulatory challenges, industrial DSM.
infrastructure, and lack of  Need for frameworks that unify
automation. forecasting, control, and optimization
 Emphasized the need for digital technologies for real-time deployment.
technologies like AI and IoT to
improve load flexibility.
[4]  Proposed a hybrid method  Lacks advanced machine learning
combining deterministic planning forecasting for highly dynamic
with stochastic adjustments. production environments.
 Focused on integrating  Deterministic models remain limited
production scheduling with DSM by rigid assumptions about demand
to improve load profiles. and resource availability.

[5]  Surveyed AI applications in DSM  Most AI models are isolated from

for smart grids, emphasizing deterministic DSM models.
factory use-cases.  Need to bridge deterministic
 Discussed neural networks, scheduling with AI-based forecasting
reinforcement learning, and to ensure both accuracy and reliability
hybrid models for forecasting and in industrial systems.
control.
Table 2.1: Summery of the literature review
Chapter 3: Aim and Objectives

This research aims to explore multiple dimensions of load behaviour in commercial

installation, with a focus on enhancing Demand Side Management and developing an
AI assisted load forecasting system. The main objectives are as follows:
1. Quantitative Analysis of Load Profile Changes
The first objective is to identify and quantify the significant changes in the load
profiles of industrial and commercial consumers. This includes analyzing historical
data to observe how consumption patterns have evolved over time and determining
the underlying causes of these changes such as increased demand, operational shifts,
or economic growth.
2. Categorization of consumer Load profile
To improve the effectiveness of DSM initiatives, the study will segment the overall
load profile into distinct customer categories (e.g., office buildings, retail, and
manufacturing). A granular understanding of the consumption behaviour of each
category is essential for targeted DSM strategies. Since existing studies often lack a
properly disaggregated load profile, this research aims to propose a method for
achieving meaningful segmentation using available data.
3. Development of forecasting models
Forecasting the load profile for the next decade is a key objective. The study will
review existing load forecasting methodologies and recommend the most suitable
approach for commercial installations. The goal is to develop a reliable, AI-based load
forecasting model that accounts for both historical patterns and emerging
consumption trends
4. Assessment and integration of DSM techniques
A critical component of the research involves identifying and evaluating various DSM
techniques—such as peak load shifting, demand response, and time-of-use pricing.
These techniques will be assessed quantitatively in the context of the forecasted load
profiles to determine their potential impact and feasibility.
Chapter 4: Methodology

The methodology for this research is designed to systematically analyze the load
profile changes in the industrial sector and develop an integrated framework that
combines Demand Side Management (DSM) techniques with AI-assisted load
forecasting.

Identify the Primary Changes of

Industrial Load Profile
(Existing Data & Literature)

Load Forecasting DSM

Identification of DSM
Data Collection and Preprocessing
opportunities

Model Selection Technique Evaluation

Model Selection Impact Quantification

Model Training and validation

Economic Analysis
Future Load Projection

Figure 4.1 Proposed Methodology

  Identify the Primary Changes of Industrial Load Profile

Initially the research project will starts with the identification of the primary
changes of the load profile of industrial factories through available data and the
literature review. Our literature review has been carried out through five areas.
o Need and potential for Demand Side Management (DSM)
o Available DSM techniques
o Existing forecasting methods
o Long term generation expansion plans.
Then the research project will be carried out through two basic sections which can be
identified as Load forecasting. Eventually these sections contain five steps each.

Load Forecasting

The load forecasting in section in the methodology contains five steps.

Splitting the Existing Load Profile

Splitting the existing load profile according to the tariff categories will be done as the
first step in order to identify the contribution of each tariff category to entire load
profile.

Observe the current load forecasting methods

The available load forecasting methods will observe as the next step of the
methodology. This aspect will be giving an idea of the load forecasting techniques and
also the factors that need be considered for load forecasting.

Forecast of the load profile for future decade

After observing and studying the current load forecasting methods and suggested
forecasting method the load profile will be forecasted next two decades.
Observe the accuracy of forecasted data
The accuracy of available forecast models will be identified during this step. In order
to evaluate accuracy of the load profile, past load profiles will run with suggested
forecasting model and accuracy will evaluate for an available load profile. After
considering all these aspects a new data forecasting model will be developed as the
next step.

Selection of most accurate method

The most fitted load forecasting model with past years will be selected as the new
mathematical model for load forecasting.

In order to have a clear picture of future increment of demand the user end demand
forecasting is necessary for the utility. This may help utility to identify whether they
are capable enough to supply the demand in future with new establishment of power
plants can be delayed using DSM initiatives.

Demand Side Management

This section also contains five steps.

Identify the DSM initiatives

 Definition for Demand Side Management

The term “Demand Side Management” was originated in 1980s following the
1973 and 1979 energy crisis. Demand Side Management (DSM) was widely
discussed in the 1980s as the alternative to supply side “overspending” in
energy systems. The basic concept behind DSM was that the least-cost option
for the energy system performance should be chosen when more supply or less
demand were compared on equal terms. The solutions were focusing on two
issues, one was to reduce the demand for energy (i.e. energy conservation) and
the other was to shift demand from peak periods to off-peak periods (i.e. load
shifting).
 In this step the existing DSM initiatives were identified and comprehensively
studied their capability to effect on system load profile. By studying the
available DSM techniques the way forward for introducing the new DSM
techniques was initiated.

Viability of DSM in industrial sector

Even though it was identified that the DSM techniques were cost effective, have
lower gestation periods it is necessary to analyze the viability of DSM in industrial
sector. In this step it will considered significantly the need and potential of DSM
comparing to the electricity demand of the country.

Impact of DSM for forecast load profile

The impact for the forecast load profile with suggested DSM techniques will be
evaluated in order to identify whether the suggestions are reaching the project
objectives or not.

Analyzing of energy saving and capacity saving to the system

The saving of energy and capacity to the system will be analyzing as last step after
coming across the all the steps of methodology. This step will give a clear picture
whether the research question is anticipated or at least the project is able to minimize
the research gap.
Chapter 5: Expected Outcomes

A methodology to predict the load profile, to assess impacts of DSM strategies and
programs

As suggested in the objectives of the research minimizing the variations of the existing
load profile and observe a simple load profile will be a one of the key outcome the
project. With the implementation of new DSM initiatives the variations of the load
profile are expected to minimize. Reducing the peak demand, shift the peak load to the
off-peak time and filling the valleys of the will be consider as key DSM techniques
which need to consider to minimize the variations of the system load profile.
Introducing DSM to industrial sector the utility or suppliers can minimize their cost to
establish new power plants and also they do not need be expect unnecessary black-
outs in the system.

AI model to forecast the load profile

Forecasted load data is necessary when the future generation expansion plans are
designing. An accurate forecast load will give a clear picture that how the system
demand is increasing since the utility can decide when they have to come up with new
generation units in order to satisfy the demand in the future. New AI models which
will higher accuracy than the existing mathematical model will obtain as final outcome
of the project.
Chapter 6: Proposed Time Line
Chapter 7: References

[1] S. Lanka and S. Energy, “Energy Consumption Benchmark Analysis.”

[2] B. Williams, D. Bishop, P. Gallardo, and J. G. Chase, “Demand Side
Management in Industrial, Commercial, and Residential Sectors: A Review of
Constraints and Considerations,” Energies, vol. 16, no. 13, 2023, doi:
10.3390/en16135155.
[3] R. Heffron, M. Körner, J. Wagner, M. Weibelzahl, and G. Fridgen, “Industrial
demand-side flexibility : A key element of a just energy transition and
industrial development,” vol. 269, no. April, 2020.
[4] S. Chen et al., “Planning and scheduling for industrial demand-side
management: State of the art, opportunities and challenges under integration
of energy internet and industrial internet,” Sustain., vol. 13, no. 14, 2021, doi:
10.3390/su13147753.
[5] M. S. Bakare, A. Abdulkarim, M. Zeeshan, and A. N. Shuaibu, “A
comprehensive overview on demand side energy management towards smart
grids: challenges, solutions, and future direction,” Energy Informatics, vol. 6,
no. 1, 2023, doi: 10.1186/s42162-023-00262-7.