See discussions, stats, and author profiles for this publication at: [Link]

net/publication/3582208

Statistical maximum demand measurement

Conference Paper · May 1990

Source: IEEE Xplore

CITATIONS READS

5 801

1 author:

Andrew Berrisford
Berrisford Power Measurement and Analytics
11 PUBLICATIONS   101 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

piNILM View project

Harmonics impact project View project

All content following this page was uploaded by Andrew Berrisford on 24 September 2019.

The user has requested enhancement of the downloaded file.

 52

STATISTICAL MAXIMUM DEMAND MEASUREMENT

A. J. Berrisford
ESKOM Engineering Investigations, Republic of South Africa

SYNOPSIS into account. The relationshipa between

measured M.D., length of integrating period,
and diversity of demand are complex and
The development of a Statistical Maximum non-linear, but in general, as the M.D.
Demand Meter and field tests carried out integrating period decreases, the measured
with the device are discussed, indicating M.D. increases. If the measured M.D.
that this method of maximum demand metering increases, the measured diversity of
produces more realistic results than the demand for a number of loads will increase.
thermal or block interval methods currently It follows that measured diversity is a
used, particularly when applied to rapidly function of the type of M.D. measurement
varying loads. Commercial tariffing by this used. The above relationships for different
method is also discussed. types of ESKOM customer are currently being
studied.
J NTRODUCTI OK
PROBLEMS W m MAXIMUM DEMAND W R E M E N T
The reason for the maximum demand component
in a large power user's electricity bill is Most electricity supply tariffs have both
based on the utility's [Link] recover the energy and demand components, and the
cost of providing generation and distribution energy component is clearly defined and
plant for the customer's 'peak' load. This accurately measured, but the maximum
may be a lot higher than the average load demand component presents a number of
as determined by monthly kWh consumption. problems:
The 'peak' that a utility can supply is
limited by the thermal overload rating of
its plant. Thus maximum demand (M.D.) 1 ) Different definitions and methods
definitions are supposedly related to the
thermal rating of the utility's plant.
As cables, transformers, etc. do not ESKOM presently measures maximum demand in
overheat instantaneously, these definitions three different ways ( 16 minute thermal
are based on an average load over a short kVA, 60 minute block kW, and 30 minute block
period, or the response of a thermal demand kVA ) each of which would give different
meter, to simulate the plant's thermal demand figures for any load. The undefined
response. The structure of the tariff, and response of the thermal demand meter for
issues such as co-incidence of a anything other than a full scale step change
customer's demand with the supply utility in load, as well as its poor accuracy and
demand, and what proportion of revenue resolution ( 6 ) have long been a source of
should be recovered from a M.D. charge, concern for ESKOM metering staff. Having
are not covered here. Rather, this paper standardised on the 30 minute kVA block
deals with the actual measurement of a interval M.D., ESKOM now has difficulty in
customer's maximum demand, and the results justifying the change to customers still
are still under consideration by ESKOM. on the 60 minute kW M.D. tariff. There
simply is no evidence that ESKOM's thermal
response is 30 and not 60 minutes. The
MAXIMUM DEMAND DEFINITIONS IN ESKOM definition of simultaneous kVA M.D. is also
a problem. What of leading VArs, and should
the cross-summation be done on an arithmetic
Although no internationally accepted or vectorial basis ?
definition for Maximum Demand exists,
most M.D. metering is based either on the
Block Interval method or the Thermal Demand 2 ) Peak-splitting errors
meter. Integrating periods used in various
countries vary from ten minutes to three
hours, with 16 or 30 minutes being the moat Although the M.D. definition appears simple,
common. The maximum demand definition used its implementation as block interval M.D.
by ESKOM is " the highest [implied average1 metering leaves a lot to be desired, when
load supplied by the Licensee to the consumer studied closely. Consider the hypothetical
during any period of 30 (thirty) consecutive case: Two identical 30 minute Block Interval
minutes in the month measured in kVA." This metering systems on the same load, but with
definition has now been standardised demand reset times 15 minutes apart, could
throughout ESKOM, although many old customers produce demand readings that differ by
in the ex RLOFS region are still billed on anything up to 60 X (ie one could indicate
a 60 minute kilowatt M.D. tariff, for 120 MW and the other 240 MW), and both
historical reasons. Strictly speaking, this acceptable in terms of the definition
makes Thermal Demand meters illegal in ESKOM, (Figure la). This is the worst case, but as
although they are still used on small loads. Bolton ( 2 ) states, a method which is capable
( even for worst case conditions ) of
producing a 50 X variation in the measurennt
M M U M DEMAND EIEASUREI(ENT AND DIVERSITY of a given quantity, "is hardly one on which
the measurement engineer can pride himself".
As diversity of demand is a factor which Modern load controllers make use of this
affects the impact of a customer's M.D. shortcoming in M.D. metering, often allowing
on the utility, the averaging process of a large peaks for short periods and then
block interval M.D. is supposed to take this limiting load as the demnnd period approaches
 53

its end, so as not to exceed a preset Block

Interval demand. Note that such load control
does not actually limit the true peak load
to the preset maximum, but merely hides it
from the Block Interval M.D. measurement.

3 ) Block Interval Timing Problems

e
ESKOM presently grants the benefit of
B
Y

diversity to many large customers with

multiple points of supply. This creates
problems such as the synchronising of
integrating periods at points tens of
kilometres apart. While synchronous clocks
provide the best long term synchronisation
of block interval periods, at the time of
national peak demand it has occurred that
system frequency has dropped, effectively
lengthening the integrating period. On the I r@
other hand, crystal based clocks tend to
drift apart after a few ronths.

l! .-
6

e
4) Rapid Load Variatipns

A plant producing large widgets slowly could

d
Y
have a much higher U.D. than an identical
plant producing small widgets at a higher
rate (Figure lb). The only difference
between the two loads is the rate at which
E
they change. Both require the same peak
demand from ESKOM for the same amount of
time per month, and the s u e M o u n t of
energy - yet their bills could differ
considerably. As industry makes more use of
electronically controlled loads, this
shortcoming in U.D. metering is becoming
more apparent. Examples of such loads are
thyristor controlled equipment, walking
draglines, arc furnaces and mine hoists.

5) Treatment of diversity of demand

Load Duration Curve
One of the reasons for not shortening the
M.D. integrating period is to grant the
customer the benefit of system-wide diversity
for short peak loads. Block interval U.D. for a total of 10 hours
implies that diversity exists for peaks
shorter than the integrating period, but
not for longer peaks; and that customers 600
with repetitive short peaks be granted the
same benefit of diversity as those with one
short peak during the billing period.
L
!40 1
500

The first two problems have been addressed

in the past. In 1939 Lincoln and Sprole ( 1 )
described the effect of different time
constants in thermal demand meters on U.D.
readings, as well as the different figures
produced by block interval and thermal demand
meters on the same loads. Bolton ( 2 ) in 1941
{ 300
i
!;\
This orea represents the total
[ Load was zero tor
the month
during
made an excellent study of the effects of
block interval starting times and integrating monthly consumption in kVA hours
period lengths on various loads. Bolton also
demonstrated that the demand period bore no
relation to the complex thermal response of
100 4 25OOkVA hours Ii
power transformers.
. .
In the discussions on his paper the concept of 0 100 200 300 400 500 600 700 800
a moving window, or 'floating' Block Interval Time (hours)
M.D. is proposed. The idea at the time used a
number of Mere indicators with staggered
demand intervals. ESKOM also investigated
the Moving Window U.D. ( which is a more Figure I C
accurate implementation of the block interval
 54

definition ) and found that only in

cases of intentional ‘peak-splitting’ would Load Profile Recorder
the idea be really useful. On typical
industrial loads (without load controllers)
the Uoving Window M.D. was 2% to 3% C.T. H.V. Line
higher than the fixed Block Interval U.D.

STATISTICAL -
m
The fourth problem, that of rapidly varying
loads, was highlighted in a technical note
by Korvink in 1977 (3). In his note Korvink
proposed a statistical method of determining
M.D. using a Load Duration Curve ( L D C ) , which
is a plot of X load againat % time, over the Watt / VAr / VA
period monitored, providing a ‘cumulative
distribution curve’ of the load (Figure IC).
In the past, producing this plot was a
laborious process, used for statistical load
forecasting and network planning. It is

I .................
D I
now possible to use the same technique to
determine a customer’s maximum demand.
Figure 2
1 1 -
A /

It was shown by Korvink that, if the micro

j k$+\
cyclic period of a varying load is less
than the Demand Integrating Period ( or the Processor
response time of a Thermal Demand Ueter ),
the demand meter will tend to indicate the .................
average load rather than the peak. A means of
characterising a load in a time-independent Printer
manner is necessary to find the U.D.
independently of the load variation rate.
The problem is the word ‘consecutive’ in
the presently used definition of M.D. Does lo4 -
it matter ? After all, if the customer draws w-
100 kVA for 30 consecutive minutes, or for Gold Mine Feeder
two 15 minute periods during the month, or 0-
for one minute during each day of the month,
ESKOM must be capable of supplying 100 kVA. 70.
A load duration curve for the month would
indicate the peak load drawn for at least z eQ-
I
30 minutes of the month, consecutively or L 50-....
not. This would treat all customers with ..,......
varying loads in the same way, irrespective
of the rate of load variation or the times
:a - ......._
that peak loads started or ended. The M.D. W-
definition would no longer be based on a
20.
guess of the plant’s thermal response, but
on a clearly defined amount of time per 10 -
billing period that the instantaneous demand
exceeded a certain value, a consistent and
repeatable measure of the customer’s peak
0 , . . . . . . . . .
load requirements. The ’30 minutes’ is
arbitrary, and ESKOU’s Marketing and Tariff
departments would decide on the correct ti-
to use. Whatever time is chosen, all loads
would be treated in the same manner. This
method also addresses the diversity aspect.
as it distinguishes between the customer
with a single, non-repetitive short peak
and the cyclic load. Another method of
determining M.D. from a LDC is to define it
as the load above which a certain percentage
of energy was consumed. For the purposes of
this paper ‘instantaneoue demand’ ia the
demand that would be registered by lost AC
power indicating instruments or tranaducera,
i.e. an average with a response time of 300
to 500 ms.

ND -
F
O
R
P

The advent of microprocessors ham X Time

made it possible to design a s ~ l and
l cheap Figure 3b
Demand Profile Recorder, to evaluate this
method of determining U.D.
 55

The DPR was designed to produce a table of B l Walking Dragline

figures or a graph of the daily load
duration curve on a printer (Figure 2). The
DPR has an input for a 5-0-6mA signal from Having proved that the DPR produced the
any transducer, although Watt, VAr or VA correct results, as expected, when metering
transducers will normally be used. It has a a steady load, it was then installed in
built-in printer and a serial port for an parallel with the Thermal Demand meter at a
external printer. The input signal range is walking dragline, again with a MDR for
divided into twenty 5% sub-ranges, and comparative studies. Figure 3(b) shows the
counters are set up for each sub-range. typical 'triangular' duration curve from
such a load. The dragline regenerates power
during part of its load cycle, which takes
The value of the input signal is monitored less than 2 minutes to complete. The
at 2 second intervals and the appropriate dragline's load cycle has been analysed
counter and all those below it are in technical literature ( 4 ) where power
incremented. At the end of the day, the X factor, regeneration, and other factors have
of time that the load was above each X load been studied. Uere the author is concerned
is calculated and printed, the counters are only that it has a fast load cycle.
reset, and the process is repeated. The
resulting daily printout is a profile of
the load duration curve. In the prototype The load curve for a month ha8 been
DPR the number of load 'bins' and the load expanded in Figure 5 , and the U.D. as
monitoring interval were dictated by the indicated by the installed Thermal meter,
limited m e w r y available on the 6802 as well as the 30 and 60 minute Block
microcomputer hoard used. The next generation Interval have been marked on the curves.
of DPR's will have a load binning resolution Note the following:
of 0,lX and a monitoring interval of 0.3
seconds to make the most of the accuracy and
response time of the transducers used. a ) The duration plot shows that the M.D. was
almost 70% higher than that indicated on the
installed Thermal Demand Meter for 30 minutes
of the month.

In order to test the statistical demand b) Although the Block Interval Demand method
measurement technique, it was decided to gave higher indications than the thermal meter
install the DPR at a relatively steady load they too could not extract the true "30
with an almost 'rectangular' load duration minutes per month" demand.
curve, and then at a rapidly varying load
with a 'triangular' load duration curve. The
c ) The mine consumed 0 , 0 6 X of its energy

-
demand determined by the DPR on the steady
load should correspond closely with that when drawing more than its metered maximum
measured by the installed block interval demand, while the dragline used about 6 X ,
equipment, while it was expected that the DPR a difference of two orders of magnitude.
would indicate a higher demand than the
thermal meter at the varying load site.

A) Gold Mine Supply Feeder

The walking dragline tests have shown clearly
that the DPR overcomes the problems inherent
The DPR was installed on a feeder to a in Block Interval and Thermal Demand U.D.
large gold mining complex for a period of measurements when metering rapidly varying
three months, together with a Uetering Data loads. The difference in terms of revenue in
Recorder (MDR). The UDR was fed from an considerable. Taking the case of the walking
impulsing Ferraris type meter and the DPR dragline, the installed Thermal Demand meter
from a power transducer. This arrangement indicated a M.D. of 3,6 MVA, whereas more
allowed the author to compare the DPR demand than 6 MVA was drawn for 1 to 2 hours per
with a variety of block interval and moving month. This is 67% more than the metered
window periods, as the MDR can record at 5 value, and represents about 26 X of the
minute intervals. The data obtained can then dragline's electricity bill. Either this
he processed as 6 , 30, or 60 minute Block customer is paying too little for his M.D.
Interval or Moving Window information. or his fellow customers with steady loads
are paying too much, because of an inadequate
maximum demand definition.
A typical monthly load duration plot for
this feeder is shown in Figure 3(a), and the
most important part of this plot has been Although the dragline is a severe case, with
expanded in Figure 4. As can be seen, the the statistical U.D. 67% higher than the
Uaximum Demand determined by 5, 30, and 60 Thermal M.D., it is not the worst. Many of
minute integrating periods are very close, the larger industrial loads show rapid load
indicating a steady load. The DPR plot fluctuations not seen by normal M.D. metering
corresponds well, showing that the load equipment. A mine hoist recently studied, for
exceeded the metered M.D. by about 5% for example, had a 4 minute load cycle with an
2 hours total and 10% for only 20 minutes average load of 3,s MW and a ten second pcak
during the month. It should be noted that of 15 - 16 UW. The statistical demand there
this mine is billed on a 60 minute Block is probably 3 to 4 times the demand that
Interval kW tariff. conventional metering would indicate.

While doing statistical analysis on the test

data, it was seen that the standard deviation
 56

of the mean load provided a good iadisation Further sources of error in the Block
of the extent of the load v d a t i o n . This Interval method, such as the use of
'figure of merit' can be calculated by the synchronous clocks and the duration of the
statistical demand meter, and be used by the resetting pulse, and the reaolution on Uer8
utility as a measure of the customer's load and Thermal scales ( 6 ) , are also overcome by
factor. this method.

mTISTICAL FOR TARIPFINO The author would like to thank the ESKOU
management and the late Mr. Q.J. Korvink for
their support, and Uessra U. Gillour and
Uaximur Demand can be derived from a LDC in L.C. de Wet for their assistance.
a number of ways, the most obvious being
either the demand that was exceeded for a
certain amount of time, or the demand above RBFBRENCBS
which a certain percentage of total energy
was consumed.
1 Lincoln. P.M. and Sprole. R.R. 1939.
"Demand Meter Time Periods" AIE8
LDC's can be based on the entire billing [Link], 68, 729 734. -
period, or the utility's peak periods, or
weighted by time-of-use factors if so 2 Bolton, D.J. 1941. "Maximum Demand
desired. The tariff designer can use the Metering : A Study of the Tiling and
Length of the Integrating Poriod",
statistical M.D. in any manner poasible, as
it is no longer necessary to limit Vo1 89, Part 11, 96 120. -
U.D. definitions or measurements to the
capabilities of 80 year old technology. The 3 Korvink, Q.J. 1977. 'A Note on Uaximum
statistical U.D. can be implemented in any Demand metering" ESKOU.
of the latest generation of microprocessor
based meters. 4 Nichols, S . , Barton, T. C Ounthorpe, 0.
1981, "Load Model of a Dragline" Iggg
m. Ind. BEE, Vol 1A-17, 4, 366-361.
The simplest approach would be to remove the
word 'consecutive' from the present U.D. 6 Korvink, Q.J. 1978, 'A Note on Demand
definition, and take the demand exceeded for Meter Reading Resolution", ESKOU.
30 minutes during the billing period as the
M.D. The effect of such a change on the
metered demands of customers with steady
loads would be minimal, but customers with Gold M i n e M.D.
regular short peaks, or those hiding peaks Figure 4
by peak-splitting, would register higher
demands. This would result in an adjustment 11 2 , 3 MWh (0,05%)
of M.D. rates. to the financial benefit of 10

Defining W.D. as the demand above which a

customer connumed a small percentage of M 4-
his energy has the additional advantage of w 3- 4700 MWh (99.95%)
not unduly penalising a customer for a short
lapse of demand control. 2-
1-

0,

D r a g l i n e M.D
Figure 5
The design, production, and testing of the
DPR have shown that the statistical method
of maximum demand determination has the L 61---1
following advantages:
I
a) The load cycle period doe. not influence 30 Min 4.3 MVA 30 MVAh (6 X)
the U.D. measured. ............................................... ............
___.....
b) The Block Interval 'peak splitting' and

'
M 3 60 Min 4.2 MVA
timing problems are overcome. V
c) The LDC contains additional information
about the demand profile, not normally
available from a U.D. measurement.
The statistical maximum demand algorithm
can be included in any of the latest micro-
processor based meters, at little or no
A :I
0
0 30 60 90
3.7 MVA

120 150
480 MVAh (94 %)

180 210 240

additional hardware cost. Time (Minutes per Month)

View publication stats