Demand

Forecast
 Demand Forecast
• Main reference for this material is [Seifi 2011]
• Demand forecast refers to predicting the future electric demand.
• Other terms: load forecast, power forecast.
• The forecast of future power demand is an essential step in any
power system planning study.
• Three time frames for load forecast:
– Short-term load forecast (STLF)
– Medium-term load forecast (MTLF)
– Long-term load forecast (LTLF)

54
 Demand Forecast
• LTLF is usually a peak demand forecast
• In general, the longer the time frame, the less accurate the
forecast is.
• Hence, LTLF is often performed over multiple scenarios, such
as scenarios of very high, high, and moderate GDP growth.
• Sometimes, spatial load forecasting is performed. Then,
aggregate forecasting is obtained using coincidence factors, as
discussed previously.

55
Demand Forecast Driving Parameters

56
 LTLF Methods
• The following LTLF methods will be introduced:
– Trend Analysis
– Econometric Modeling
– End-Use Analysis
– Combined Analysis

57
 Trend Analysis
• This is a trend extrapolation
method
• It uses the information of the past
to forecast the load of the future.
• A simple example is shown in Fig.
4.5, in which load is shown for the
last 10 years and predicted to be
2906 MW in 2015.
• A curve fitting approach may be
employed to find the load of the
target year.

58
 Trend Analysis
Pros?
• Simple to understand and inexpensive to implement.

Drawbacks?
• It assumes that the trends in various load driving parameters
remain unchanged during the study period.
• For instance, if there is a substantial change in economic
growth, the approach fails to forecast the future load, accurately.

• A modified version of this method, more weights may be

attached to the loads towards the end of the past period.
• In this way, the prediction may be improved.
59
 Econometric Modeling
• Initially the relationship between the load and the driving
parameters is estimated.
– The relationship may be nonlinear, linear; additive or in the
form of multiplication.
– This relationship is established based on available historical
data.
– Various driving parameters may be checked to find the ones
that have the dominant effects.

60
 Econometric Modeling
• A common nonlinear estimation is:

Di =a(per capita income)bi (population)ci (electricity price)di

where i denotes the year
a, b, c, d are the parameters to be determined from the
historical data.

61
 Econometric Modeling
• Once this relationship is established, the future values of the
driving variables (i.e. per capita income, population, electricity
price, etc.) should be projected.
• Di for a future year can then be determined.

62
 Econometric Modeling
• This approach is widely used and may be applied to various
customer classes (residential, commercial, etc.) and to the system
as a whole.

Pros?
• Relatively simple to implement.

Drawbacks?
• It assumes that the relationship established for the past will be
still applicable for the future.
• In this way, the influence of any new driving parameter cannot be
taken into account.
63
 End-Use Analysis
• Mostly for residential loads.
• May be applied with some modifications to other load
classes, too.
• Example: Forecast refrigeration load…
• The number of refrigerators for a future year may be
estimated based on
– the number of households and
– the percent of households having a refrigerator (estimated),
• Following that and based on average energy use of one
refrigerator, the total energy consumption of refrigerators
may be estimated. 64
 End-Use Analysis
• The same procedure may be applied to other appliances in
order to forecast the total energy requirement.
• This approach explicitly predicts the energy consumption.
• If the load is to be estimated, some indirect approaches
have to be used to convert the predicted energy to load
(power demand).

65
 End-Use Analysis
Pros?
• Very accurate
• Various driving parameters effects may be taken into
account.

Drawbacks?
• Extensive amount of data is needed.

66
 Combined Analysis
• End-use and econometric methods may be simultaneously
used.
• A compromise.

67
 EE524

Forecast Accuracy

• Forecast error: Difference between forecast and actual demand

MAE / MAD
• Mean absolute error / deviation

RMSE
• Root Mean Square Error

MAPD
• Mean absolute percent deviation

Cumulative error

Average error or bias

68
 EE524

Mean Absolute Error (MAE)

Mean Absolute Deviation (MAD)
𝑛
1
𝑀𝐴𝐸 = 𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝑥𝑎𝑐𝑡𝑢𝑎𝑙
𝑛
1

where
t = period number
𝑥𝑎𝑐𝑡𝑢𝑎𝑙 = demand in period t
𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = forecast for period t
n = total number of periods
= absolute value

69
 EE524

Root Mean squared error (RMSE)

𝑛
1
𝑅𝑀𝑆𝐸 = (𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝑥𝑎𝑐𝑡𝑢𝑎𝑙 )2
𝑛
1

t = period number
𝑥𝑎𝑐𝑡𝑢𝑎𝑙 = demand in period t
𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = forecast for period t
n = total number of periods

70
 EE524

Other Accuracy Measures

Mean absolute percent deviation (MAPD)
|𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝑥𝑎𝑐𝑡𝑢𝑎𝑙 |
MAPD =
𝑥𝑎𝑐𝑡𝑢𝑎𝑙

Cumulative error
E= 𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 −𝑥𝑎𝑐𝑡𝑢𝑎𝑙

Average error
(𝑥𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 −𝑥𝑎𝑐𝑡𝑢𝑎𝑙 )
E= n
71
 Numerical Example I: End-Use Analysis
• Figure 4.6 [Seifi] (next slide) shows the region for
which the load is to be forecasted.
• It consists of eight areas.
• Each area consists of some subareas, supplied through
some substations, either existing or new.

• A summary of the data is shown in Table 4.1 (next

slide),

Goal: To predict the peak load and the energy demand of

the regional utility for 10 years from the current year; with
a time step of 1 year.
72
 Numerical Example I

Fig 4.6 [Seifi]

74
 Numerical Example I
• Spatial load forecasting will be used.
• That is, the process starts from forecasting the load of
the subareas; moving upwards to the areas then to the
load for the region.
• It is assumed that the geographical characteristics of the
subareas as well as their load data for the last 10 years
are known.

75
 Numerical Example I
• It is assumed that each subarea consists of the
following three types of loads:
– Urban
– Rural
– Large customers

76
 Numerical Example I
• The urban loads, typically, consists of
– Residential
– Commercial
– Public
– Small industrial
– Distribution losses
• Historical data as well as extensive data from the
regional departments, in charge of the above mentioned
sections, are required to reach at reasonable predictions.
• An urban load is not, actually, concentrated at a specific
geographical point and is distributed throughout the
urban territory. These points have to be observed. 77
 Numerical Example I
• The rural loads, typically, consists of
– Residential
– Agricultural
– Others (public, small industrial, etc.)
• The residential part may be estimated based on
– the estimated number of homes and
– the estimated power consumption of each home, which is determined
based on its existing figure and the possible increase in usage due to
various reasons (say, new appliances and technologies appearing in rural
areas).

78
 Numerical Example I
• The rural loads, typically, consists of
– Residential
– Agricultural
– Others (public, small industrial, etc.)
• The agricultural part is determined based on
– the estimated number of wells,
– their average depths and
– their average water flows.

79
 Numerical Example I
• The rural loads, typically, consists of
– Residential
– Agricultural
– Others (public, small industrial, etc.)
• For instance: in a subarea, there may be a total number
of 491 deep well, with 75 meter average depth and 25
liter/sec average flow. These figures may be 2735, 36
and 15, respectively, for semi-deep wells in the same
subarea. Based on these figures, the agricultural load of
the subareas and, as a result, the area may be
determined.
80
 Numerical Example I
• The rural loads, typically, consists of
– Residential
– Agricultural
– Others (public, small industrial, etc.)
• The “Others” part should also be estimated. If difficult,
sometimes, a fixed percentage (say 25%) may be
considered.

81
 Numerical Example I
• The large customers’ loads, are considered separately,
as they do not obey any specific rule, in terms of, the
forecasted loads.
• They are typically the customers with more than 1 MW.
• They may be either existing or new.
– The future loads of existing types may be estimated based on
their previous and foreseen performances.
– The loads of new types are determined based on the demands
of their respective contracts with the utility.
– Both require extensive data gathering and communications
with the large customers and the departments in charge of
large customers.
82
 Numerical Example I
• Based on the above, for each subarea, the peak forecasted load is
determined for each class of the loads.
• coincidence factors should then be used to find out the forecasted
load of the area and then the region.
• These factors may be determined based on both historical data
and some engineering judgments.

83
 Numerical Example I

Fig 4.6 [Seifi]

84
 Numerical Example I
• For the region under
consideration, the results for
area E will be shown next.
• This area consists of four
subareas, see Table 4.1.
• There are two main urban loads
in subareas 1 and 2 (UE1 and
UE2).

The details are as

follows:
85
 Numerical Example I
• Area E’s results for the urban parts of subareas 1 and 2 are
shown in Table 4.2 for some selected years.
• Note: This assumes subareas 3 and 4 consist of no urban
parts.

86
 Numerical Example I
• Area E’s results for the rural-agricultural part of the area are
shown in Table 4.3

87
 Numerical Example I
Area E’s results for
the rural-residential
and rural-others
loads are shown in
Table 4.4

rural-others is
assumed to be about
27.5% of rural-
residential

88
 Numerical Example I
Area E’s results for urban, rural and large customers’ loads are
shown in Table 4.5

89
 Numerical Example I
Area E’s results for urban, rural and large customers’ loads are
shown in Table 4.5

90
Numerical Example I

91
 Numerical Example I
Area E’s results for urban, rural and large customers’ loads,
subarea-based, are shown in Table 4.6

92
 Numerical Example I
• The previous results were all for Area E.
• The same procedure is repeated for all the other areas in the
region.

93
 Numerical Example I
Region’s results for
the rural-residential
and rural-others loads
are shown in Table 4.7

94
 Numerical Example I
Region’s results for the rural-agricultural loads are
shown in Table 4.8

95
 Numerical Example I
Region’s results for the rural-agricultural loads are
shown in Table 4.8

96
 Numerical Example I
Region’s results for urban, rural and large customers’ loads are
shown in Table 4.9

97
 Numerical Example I
Region’s power and
energy results.
Note: The load factor
is defined as

Historical data was

used to identify the
LF. Then, that load
factor and the load
power are used to
obtain the total
energy
99