Unit 3

Electrical Load Forecasting

BG
TU-IOE

7/24/2025 BG/TU, IOE 1

 Variable Load on a Power System:
 The function of a power station is to deliver power to a large
number of consumers.
 However, the power demands of different consumers vary in
accordance with their activities. The load is never constant.
 Most of the complexities of modern power plant operation arise
from this inherent variability of loads.
The power engineer would like to run all the generators at their
rating for max. efficiency but the demands have wide variation.
 Unfortunately, electrical power can’t be stored. So power must
be produced as and when demanded.
Effect of variable load:
Need of additional equipment and staff:
• Back up units and additional floor area required for them.
• Need of constantly changing the flow of water/coal/air acc. to
the change in load. This demands technical sophistication.
Back up units and technical sophistication increase cost.
7/24/2025 BG/TU, IOE 2
 3.1
Load Curve: The curve showing the variation of load on the power
station w.r.t. time is known as a load curve.
Daily load curve: load variations plotted for a day
Monthly load curve: load variations plotted for a month
Yearly load curve: load variations plotted for a year

7/24/2025 BG/TU, IOE 3

 Load duration curve:
When the load elements of a load curve are arranged in the order
of descending magnitudes, the curve thus obtained is called a load
duration curve. The load duration curve is obtained from the
same data as the load curve but the ordinates are arranged in the
order of descending magnitudes. In other words, the maximum
load is represented to the left and decreasing loads are represented
to the right in the descending order.
Hence the area under the load duration curve and the area under
the load curve are equal. The figure below shows the daily load
curve. The daily load duration curve can be readily obtained from
it. It is clear from daily load curve, those load elements in order of
descending magnitude are 20 MW for 8 hours; 15 MW for 4
hours and 5 MW for 12 hours. Plotting these loads in order of
descending magnitude, we get the daily load duration curve as
shown in Figure below.

7/24/2025 BG/TU, IOE 4

7/24/2025 BG/TU, IOE 5
 In a load curve:
(i) Display of variation of load
(ii) The peak of the curve represents maximum load
(iii) The area under the curve represents the total units generated
during the specified period.
Load curve is a major reference for selecting the size and number of
generating units and preparing the operating schedule of the plant.
Average Load: The ratio of area in kwh of the load curve and time
E.g. Average daily load=Area in kwh under daily load curve/24 hrs

Connected Load: Sum of the continuous ratings of all the

equipment connected to the supply system.
7/24/2025 BG/TU, IOE 6
 Maximum Demand: Maximum demand during the given period

The idea of demand factor was introduced due to the fact that not all
the equipment connected to the system will be working at the same
time in practice. So demand factor always less than 1.
The minimum capacity of the plant must be able to meet the
maximum demand.

The maximum demands of the individual consumers generally don’t

occur at the same time. So diversity factor is always more than 1.
Coincidence Factor: It is the reciprocal of the diversity factor. Its
always less than 1.

7/24/2025 BG/TU, IOE 7

 Significance of Load Factor and Diversity Factor
Higher the values of load factor and diversity factor, lower will be
the overall cost per unit generated.
Higher load factor means greater average load resulting in greater
number of units generated for a maximum demand. Thus the
standing charges which are proportional to the maximum demand
can be distributed over a large number of units supplied. Therefore
the cost per unit of electrical energy generated is reduced.
The capital cost of the power station depends upon the capacity of
the station and that depends upon the maximum demand. With a
given number of consumers, the higher the diversity factor, the
smaller will be the plant capacity required and consequently the fixed
charges due to capital investment will be much reduced.
The suppliers should always try to improve the load factor and
diversity factor by inducing the consumers to use electricity during
off-peak hours. They are charged at lower rates for such schemes.
7/24/2025 BG/TU, IOE 8
 3.1.1 Types of Load
(i) Domestic (residential) load:
Lighting, small motors, home appliances, electronic equipment.
Domestic load generally peaks at evening.
(ii) Commercial load:
Lighting, decorative lighting, medium sized motors, large
refrigerators, music, theatres, movies, entertainment industry,
stadiums.
(iii) Non-commercial (municipal) load:
Lighting, display signs, advertisements
(iv) Industrial load:
Large motors, welding machines, lighting
(v) Irrigation load:
Large motors (water pumps)
(vi) Transport (traction) load:
Tram cars, trolley buses, railways, cable car
7/24/2025 BG/TU, IOE 9
 Consumer Demand Factor Diversity Factor
Domestic Lighting (up to 1 kw) 0.5 ~ 0.7 3-5
Domestic Power 0.5 1.5 ~ 2
Commercial Loads 0.5 ~ 0.7 1.5 ~ 2
Industrial Loads 0.5 1.5 ~ 2

7/24/2025 BG/TU, IOE 10

 Base Load and Peak Load
Base Load: The unvarying load which occurs almost the whole
day on the station is known as base load.
Peak Load: The various peak demands of load over the base load
of the station is peak load.
Intermediate load is the load in between base and peak loads.

7/24/2025 BG/TU, IOE 11

 Methods of managing base and peak loads:
 The more efficient and low cost stations are operated to supply
for the base load and less efficient ones can be chosen for
supplying in times of peak loads.
 Interconnection of grids is also a very reliable way of managing
the peak loads.

7/24/2025 BG/TU, IOE 12

 3.2 Load forecasting
Prediction of load and energy level for future by a utility for a given
region.
Objectives:
• To make advanced decision for profitable and judicious investment .
• To estimate quantity and timing of additional power to satisfy
future needs.
• To keep the system reliability at desired level.
• To provide basis for examining energy supply problem of a country
in a broader perspective.
• To assist in determining capital expenditure requirements for system
growth and system adequacy
• To assist in determining fuel supply requirements .

7/24/2025 BG/TU, IOE 13

 Planning/Building Period and Life Expectance of Power Plants:

Type Planning and Building, Life Expectancy,

Years Years
Thermal 2~3 20 ~ 25
Nuclear 3~6 15 ~ 20
Hydro 4~8 More than 50

7/24/2025 BG/TU, IOE 14

 Classification(1)
Based on duration of future demand
• Long term forecasting
 20-30 years duration
 Uses past data for prediction
 Used for planning the generation expansion strategy
 Examines energy supply problems in a broader perspective
• Short term forecasting
 2-5 years duration
 Used to decide operating procedure and budget estimates
• Immediate forecasting
 Next hour demand
 Vital for efficient operation of thermal system
 Depends on weather variations, special events

7/24/2025 BG/TU, IOE 15

 Classification(2)
Based on geographical area
• National load forecast
 Annual peak load and annual energy for the whole country.
 Done usually by CGOU on a long term basis.
• Regional forecast
 Carried out for some large geographical region of a country.
 Usually done both by CGOU and privately or investor owned
utility.
• Specific area wise forecast
• Usually carried out in a particular area by small local utility.

7/24/2025 BG/TU, IOE 16

 Classification(3)
Based on power generation, transmission and distribution
• Load forecasting for generation
 Finds the size of new power plants on the basis of forecasted peak
demand plus spinning reserve.
• Load forecasting for transmission and distributions
 To estimate optimum size of transformer and conductors.
 To maintain good voltage profile
 To operate transmission line without stability limit violated.
Classification(4)
Based on Global Consideration
• Global forecasting
 Evaluates the energy resources scenario and the environmental
impacts associated with its utilization in the global context.
 Carried out usually by institutions like World Bank, World Energy
Commission etc.
7/24/2025 BG/TU, IOE 17
 3.3 Tools and Approaches
Statistical Method of Load Forecasting
The fundamental law of statistics is that every phenomenon
in nature obeys some mathematical law which can be
represented by a smooth curve. So electrical load forecasting
too should obey this rule, except in cases of unforeseen
developments such as sudden establishment of large
industries or long impacting natural disaster.

In this method, future demand is forecasted on the basis of

mathematical analysis of data of past several years and other
parameters existing.
Regression

7/24/2025 BG/TU, IOE 18

 Factors to be kept in mind while forecasting
 The geographical region where power is to be sold.
 Topography and climate of the region and its effect on the load
requirement.
 Population distribution, past growth and estimated future
growth.
 Availability of labor, past distribution and estimated future
available labor.
 Income per person
 Main economic activities i.e. agriculture, textiles etc.
 The existing use and the existing rate of consumption of
electrical energy.
 Relative cost of electrical energy and the other forms of energy
available in the regions.
 Natural resources of the region viz. agriculture, timber, minerals
and raw materials for manufacturing industries
7/24/2025 BG/TU, IOE 19
3.4 Errors and Uncertainties in Load forecasting
Whatever be the method of forecast, the actual value in the nth year
never equals the predicted value for that year.
Uncertainty results from error in any one of the past data, presently
collected data and the error in the degree of certainty with which
historical trend supposed to continue in the future, the error in the
survey data (initialization data) and test data etc. The chief sources
of errors are as follows:
• Error in the historically recorded data.
• Error in measuring instruments.
• Error in calculation.
• Error introduced by particular forecasting method
• Error in coincidence factor and load factors.

7/24/2025 BG/TU, IOE 20

 Uncertainties
Negative Uncertainty: When the system peak load exceeds the
installed capacity (based on the forecast) of the system, its
‘negative uncertainty’. The reliability of the system deteriorates
and system security falls. So the cost of unreliability is
increased.
Positive Uncertainty: When the system installed capacity
exceeds the system peak demand (by large margin), its ‘positive
uncertainty’. The reliability of the system is increased, security is
increased but the cost of reliability is increased.

7/24/2025 BG/TU, IOE 21

Approaches to Load Forecasting
Time series approach
In this approach of forecasting, the system is treated as a black box and no
attempt is made to discover the factors affecting its behavior. The reasons
for treating the system as black box are:
- the system may not be understood and even if it is understood, it may be
extremely difficult to measure the relationship governing its behavior.
- The main concern may be only to predict what will happen and not to
know why it happens.
Explanatory approach
In this approach, any change in the inputs will affect the output of the
system in a predictable way assuming that the cause-and-effect relationship
between input and output is constant.
So, the first task while forecasting is to find the cause-and-effect
relationship by observing the output of the system and relating that to the
corresponding inputs.
7/24/2025 BG/TU, IOE 22
 Load Forecasting by Survey Method
• Survey method are used when historical data are not available
• The loads are grouped under different categories such as
domestic, commercial, industrial, etc.
• Survey sheets are used to collect detail information area wise
and category wise
• Information like demand factor, group diversity factor (group
coincidence factor) and Peak diversity factor(peak coincidence
factor) etc. can be found from the survey sheet
• Peak load for each category and hence the total maximum
demand can be found using the above information

7/24/2025 BG/TU, IOE 24

 Sample of survey sheet:
Zone: District: Municipality:
Ward no.: Village: Type of consumer:
Consumer No.:

Type of load Time period in a day when the load is on

Light (bulb) 6 pm to 9 pm
Light (tube) 6 pm to 9 pm
Fans 11 am to 4 pm
….. ---

7/24/2025 BG/TU, IOE 25

 Mathematical Modeling of Data for Load Forecasting
The available past data can be modeled into simple regression curves.
Based on these curves, future forecasting can be made. In regression
for load forecasting, demand (in MW) or consumption (in KWh) is
plotted in Y-axis against time in X-axis.
P = f (T) or E = g(T)
Linear Regression:
Modeling of the available data in linear form, y = a+bx, where a and
b are constants.
Exponential Regression:
Modeling of the available data in exponential form, y = aebx, where a
and b are constants.
Parabolic Regression:
Modeling of the available data in parabolic form, y = a+bx+cx2,
where a, b and c are constants.
Curves are fitted by the least square method.
7/24/2025 BG/TU, IOE 26
 Year Demand, MW Linear
1 150
2 175
3 225
4 300
5 400
6 550
7 800
8 1000
9 1300

Exponential Parabolic

7/24/2025 BG/TU, IOE 27

 Measures of Accuracy of Forecasted Data
‘Accuracy’ means the ‘goodness of fit’ which is how well the
forecasting model is able to reproduce the data that are already
known. Generally, a subset of the known data is used to develop the
model and to forecast the rest of the known data, which enables the
study of accuracy of the forecasts. Or the study of accuracy can be
made in the whole of the known data.
Measures of Accuracy:
If Xi is the actual datum for time period i and Fi is the forecast (of
the fitted value) for the same period, then the error for that point
(time period) is defined as;
ei = Xi - Fi
So if there are n observations, there will be n fitted values and n
errors.

7/24/2025 BG/TU, IOE 28

Standard Statistical Measures Relative Measures

7/24/2025 BG/TU, IOE 29

 Examples:
(1) Following are the annual peak demand data for a utility for the last
14 years. Find out the best fitting model out of linear, exponential and
parabolic mathematical models based on MAPE. Using the best fitting
model, forecast the annual peak demand for the next 20 years. Take
the first 10 data as the initialization set
Year 1999 2000 01 02 03 04 05 06 07 08 09 10 11 12
Maximum
141 150 176 204 216 220 231 244 275 278 282 284 298 312
Demand, MW
Solution:
Equations for the three mathematical models obtained from the
provided data;
Linear: y = 136.18 + 13.472x
Exponential: y = 144.2e0.0638x
Parabolic: y = 116.98 + 21.703x + - 0.6331x2
Where x is ‘year (in no.)’ and y is ‘maximum demand’.

7/24/2025 BG/TU, IOE 30

(i) for linear model;
Absolute Relative
Actual Forecasted
Error,
Year Year in no. Demand, MW Demand, MW
|(Xi - Fi)/Xi|
(Xi) (Fi)
1999 1 141
2000 2 150
01 3 176
02 4 204
03 5 216
04 6 214
05 7 231
006 8 244
07 9 275
08 10 272
09 11 278
10 12 284
11 13 298
12 14 312
7/24/2025 BG/TU, IOE 31
(ii) for exponential model;
Absolute Relative
Actual Forecasted
Error,
Year Year in no. Demand, MW Demand, MW
|(Xi - Fi)/Xi|
(Xi) (Fi)
1999 1 141
2000 2 150
01 3 176
02 4 204
03 5 216
04 6 214
05 7 231
006 8 244
07 9 275
08 10 272
09 11 278
10 12 284
11 13 298
12 14 312
7/24/2025 BG/TU, IOE MAPE 32
(iii) for parabolic model;
Absolute Relative
Actual Forecasted
Error,
Year Year in no. Demand, MW Demand, MW
|(Xi - Fi)/Xi|
(Xi) (Fi)
1999 1 141
2000 2 150
01 3 176
02 4 204
03 5 216
04 6 214
05 7 231
006 8 244
07 9 275
08 10 272
09 11 278
10 12 284
11 13 298
12 14 312
7/24/2025 BG/TU, IOE MAPE 33
 As can be seen from the three tables above, MAPE is lowest for the
linear model. So this model is used to forecast for next 20 years.
Now the year no. starts from 15.
Linear: y = 136.18 + 13.472x

7/24/2025 BG/TU, IOE 34

(2) Following are the annual peak demand data for a utility for the last
9 years. Find out the best fitting model out of linear, exponential and
parabolic mathematical models based on MAPE. Using the best fitting
model, forecast the annual peak demand for the next 20 years. Take
the first 6 data as the initialization set.
Year 2012 13 14 15 16 17 18 19 20
Maximum
7.3 7.6 7.9 8.3 8.6 8.8 8.9 9.5 10.6
Demand, MW
Solution:
Equations for the three mathematical models obtained from the
provided data;
Linear: y =
Exponential: y =
Parabolic: y =
Where x is ‘year (in no.)’ and y is ‘maximum demand’.

7/24/2025 BG/TU, IOE 35

7/24/2025 BG/TU, IOE 36
 (3) When the annual load factors for sufficiently long historical data
of a utility are plotted against time in year, a linear relationship as
shown below was obtained;
L.F. = 0.42 + 0.238 x
(x = 0 for years upto 2009, x = 1 from 2010 onwards)
The annual energy consumption in MWhr for the last 16 years is
tabulated below. Taking 12 data as the initialization set, forecast the
annual peak energy demand for the next 25 years assuming the
expression for the load factor holds true for coming 25 years.
Choose the best fitting model among linear, exponential and
parabolic models, based on MAPE.

7/24/2025 BG/TU, IOE 37

 Year MWhr
2003 252
04 256
05 267
06 281
07 278
08 279
09 282
10 286
11 289
12 293
13 314
14 327
15 335
16 333
17 340
18 349
7/24/2025 BG/TU, IOE 38
Solution: Year Consumed Energy, MWhr L.F. Max. Energy Demand, MWhr
2003 252
04 256
05 267
06 281
07 278
08 279
09 282
10 286
11 289
12 293
13 314
14 327
15 335
16 333
17 340
18 349
7/24/2025 BG/TU, IOE 39
 Equations for the three mathematical models obtained from the
data for the maximum energy demand;
Linear: y =
Exponential: y =
Parabolic: y =
Where x is ‘year (in no.)’ and y is ‘maximum demand’.

7/24/2025 BG/TU, IOE 40