Energy and Power En gi neering, 2011, 3, 190-193
doi:10.4236/epe.2011.32024 Published Online May 2011 (
Copyright © 2011 SciRes. EPE
A Hybrid Short Term Load Forecasting Model of an
Indian Grid
Rabindra Behera1, Bibhu Prasad Panigrahi1, Bibhuti Bhusan Pati2
1Department of Electrical Engineering I. G. I. T. Sarang, Orissa, India
2Department of Electrical Engineering VSSUT Burla, Orissa, India
Received March 21, 20 1 1; revised March 28, 2011; accepted April 8, 2011
This paper describes an application of combined model of extrapolation and correlation techniques for short
term load forecasting of an Indian substation. Here effort has been given to improvise the accuracy of elec-
trical load forecasting considering the factors, past data of the load, respective weather condition and finan-
cial growth of the people. These factors are derived by curve fitting technique. Then simulation has been
conducted using MATLAB tools. Here it has been suggested that consideration of 20 years data for a devel-
oping country should be ignored as the development of a country is highly unpredictable. However, the im-
portance of the past data should not be ignored. Here, just previous five years data are used to determine the
above factors.
Keywords: Short Term Load Forecasting, Parameter Estimation, Trending Technique, Co-Relation
1. Introduction
Electrical energy is a superior form of energy for all
types of consumer needs. The close tracking of system
generation at all time is the basic requirement in the op-
eration of power system. There is a 3% - 7% of increase
of electrical load per year for many years. Short-term
load forecasting (STLF) is essential for an effective en-
ergy management in a deregulated power open market.
However, the electric power load forecasting problem is
not easy to handle due to nonlinear and random-like be-
haviors of system loads, weather conditions, and varia-
tions of social and economic environments.
A wide variety of models have been proposed in the
last two decades for STLF due to its importance etc. A
wide variety of models have been proposed in the last
two decades for STLF due to its importance, such as
Functional clustering and linear regression for peak load
forecasting [1], Mixed price and load forecasting of elec-
tricity markets by a new iterative prediction method [2]
and univariate modeling and forecasting of monthly en-
ergy demand time series using abductive and neural
networks [3] etc. Moreover, the electrical load forecast-
ing depends on many known and unknown variables.
These variables can be considered to have steady values
within a specified reg ion under one electricity regulatory
authority. This type of load forecasting is being termed
as spatial load forecasting.
Aldo Goia, Caterina May and Gianluca Fusai in their
paper “Functional clustering and linear regression for
peak load forecasting” suggested a new approach using
past heating demand data in a district-heating system.
Nima Amjadya, Ali Daraeepour in their paper “Mixed
price and load forecasting of electricity markets by a new
iterative prediction method” suggested real conditions of
an electricity market and short-term load forecasting. R.
E. Abdel-Aalin’s paper “Univariate modeling and Fore-
casting of Monthly Energy Demand Time Series Using
Abductive and Neural Networks” suggested univariate
modeling of the monthly demand time series based only
on data for 6 years to forecast the demand for the seventh
year contrary to multivariate models.
Here the load and weather data of Bhubaneswar (India)
power grid has been collected for six consecutive years.
Also the economic gr owth of the people is studied. It has
been observed that economic growth in short term can be
considered as neglig ible.
A Case study has been conducted on an Indian grid
located at Bhubaneswar, Orissa based on previous load
and weather data.
2. Proposed Model
This model is a combined model of extrapolation and
correlation techniques. Extrapolation techniques involve
fitting trend curves to basic historic data adjusted to re-
flect the growth trend itself. With a trend curve the fore-
cast is obtained by evaluating the trend curve function at
the desired future point. Correlation techniques of fore-
casting relate system loads to various demographic and
economic factors. This approach is advantageous in
forcing the forecaster to understand clearly the interrela-
tionship between load growth patterns and other meas-
urable factors. The advantage is the need to forecast
demographic and economic factors. Typically, the fac-
tors such as population, building permits, business,
weather data and the like are used in correlation tech-
niques [4-8].
No one method of forecasting is effective in all situa-
tions. Here, effort has been given by considering a par-
ticular location and temperature variation in the month of
January. A study has been conducted on the real-time
data collected over past five years. It may be suggested
that the load data can be divided into three parts: 1) con-
stant about 90%; 2) weather dependent about 5% - 6%; 3)
Unpredictable about 4% - 5%. Considering the above
facts we may estimate the constant term as follows:
yk bk
 
bNk k
where, N = Total number of data.
load data at |th interval
yk k
average or mean value
The average weather report of metrological depart-
ment India has been suggested in their website. The
yearly temperature graph in the region is represented in
Figure 1;
The curve fitting weather correlation function may be
defined as:
 
yk AeBk (3)
where the constants A, B and a can be calculated based
upon the amplitude and respective instant.
Hence, the forecasting equation can be given as:
 
ykyky k (4)
A computer programme can be developed to simulate
Figure 1. Statistical data of temperature vs days in a year.
the above equation. Here a matlab simulation is being
3. Results and Discussion
A case study has been conducted on an Indian grid lo-
cated at Bhubaneswar using MATLAB Toolbox. The
details of the case study are mentioned in the following
tables and graphs. Here, the load data for five consecu-
tive years has been taken into consideration. Earlier
various models and simulation results suggested that
more and more real time data can help the forecaster to
obtain a more accurate result. However, here it is be-
lieved that data more than five years are obsolete. This
algorithm may be useful for two years, next we have to
check again the value of the constants, as they are all
time dependent. It is so because the financial growth of
the people, weather condition, advancement of the tech-
nology, policy of a government and global relation
among countries changes. Hence, application of trending
technique becomes more complicated if we will incor-
porate the above factors into account. However, it can be
suggested here that if a model can be prepared including
all of the above factors, accuracy of the forecasting can
be increased.
Here the daily peak load and minimum load are taken
into consideration separately. A week data has been ran-
domly selected then applying the proposed method the
next weeks data are evaluated then the optimum error
and absolute error were recorded and given in Tables 1
and 2. The result has been compared with the proposed
ANN model depicted in [14-15]. The average error for
peak load is mentio ned in the above article [1 6] is 2.35%
which is compared with the present result of this model
Copyright © 2011 SciRes. EPE
Table 1. Average minimum load forecasting for one week.
Time Actual load
(mw) Forecasted
load (mw) Error (mw) Percentage
Abs. error
3.00 32 32.7018 0.7018 2.17
27.00 30 29.8640 –0.14 0.468
51.00 33 33.6784 0.6784 2.02
75.00 23 24.7796 1.7796 7.18
Sunday 34 34.8234 0.8234 2.36
123.0 27 27.2701 0.2701 0.99
147.00 28 27.4538 0.54 1.96
Table 2. Average peak load forecasting for one week.
Time Actual load
(mw) Forecasted
load (mw) Error (mw) Percentage
Abs. error
18.00 64 62.8169 –1.1831 1.883
42.00 63 62.3988 –0.5012 0.803
69.00 55 56.0949 1.0949 1.951
94.00 58 56.2285 –1.7715 3.150
Sunday 52 53.5612 1.5612 2.914
141.0 0 54 55.6114 1.6114 2.897
165.00 54 53.8354 –0.1646 0.31
which is 1.986% and found suitable. Comparing to the
Sunday load forecasting with the holiday forecasting, the
former has 2.94% whereas the ANN model depicted in
[17-18] has 3.56% on average b a sis.
Here, a comparison between forecasted absolute error
for peak load and minimum load is given in Figure 2.
Also a comparison between actual peak load and fore-
casted peak load and that of minimum load are given in
Figure 3 and Figure 4 respectively. It can be observed
that the absolute error of peak load forecasting in case of
Saturday as shown in Figure 2 is much more than that of
the mini mum lo ad on the s ame day. The reason is due to
the government policy on declaration of holiday is dif-
ferent from that of the other parts of the world. Here
second Saturday of each month is given as a holiday.
However some of the organizations follow different
methods to declare Saturday as holiday.
4. Conclusions
The combined extrapolation and correlation technique
was tested and the results were presented as above. The
validation of the proposed model was compared with that
of the results mentioned in [19] of references mentioned
below and found su itable for predictio n.
Figure 2. Percentage abs. error for maximum and minimum
load diagram.
Figure 3. Peak forecasted load diagram.
Figure 4. Minimum forecasted load diagram.
Copyright © 2011 SciRes. EPE
Copyright © 2011 SciRes. EPE
5. References [10] H. Mori and M. Ohmi, “Probabilistic Short-Term Load
Forecasting with Gaussian Processes,” Proceedings of the
13th International Conference on Intelligent Systems
Application to Power Systems, Arlington, 6-10 November
2005, p. 6. doi:10.1109/ISAP.2005.1599306
[1] A. Goia, C. May and G. Fusai, “Functional Clustering
and Linear Regression for Peak Load Forecasting,” In-
ternational Journal of Forecasting, Vol. 26, No. 4, 2010,
pp. 700-711. doi:10.1016/j.ijforecast.2009.05.015 [11] H. Chen, C. A. Canizares and A. Singh, “ANN Based
Short-Term Load Forecasting in Electricity Markets,”
IEEE Proceedings of Power Engineering Society Winter
Meeting, Columbus, 28 January-1 February 2001, pp.
411-415. doi: 10.1109/PESW.2001.916876
[2] N. Amjady and A. Daraeepour, “Mixed Price and Load
Forecasting of Electricity Markets by a New Iterative
Prediction Method,” Electric Power Systems Research,
Vol. 79, No. 9, 2009, pp. 1329-1336.
doi:10.1016/j.epsr.2009.04.006 [12] H. S. Hippert, C. E. Pedreira and R. C. Souza, “Neural
Networks for Short-Term Load Forecasting: A Review
and Evaluation,” IEEE Transactions on Power Systems,
Vol. 16, No. 1, 2011, pp. 44-55. doi: 10.1109/59.910780
[3] A. D. Papalexopoulos and T. C. Hesterberg, “A Regres-
sion-Based Approach to Short-Term System Load Fore-
casting,” IEEE Transactions on Power Systems, Vol. 5,
No. 4, 1990, pp. 1535-1547. doi: 10.1109/59.99410 [13] W. Wang, C. Cheng and L. Qui, “Genetic Programming
with Rough Sets Theory for Modeling Short Term Load
Forecasting,” Fourth International Conference on Natural
Computational, Jinan, 18-20 October 2008, pp. 306-310.
doi: 10.1109/ICNC.2008.141
[4] S. Rahman and I. Moghram, “Application of Knowledge
Based Algorithms in Electric Utility Load Forecasting,”
IEEE Southeastern Conference on Energy and Informa-
tion Technologies, Richmond, March 1989, pp. 380-385.
doi:10.1109/SECON.1989.132399 [14] E. A. Feinberg, J. T. Hajagos and D. Genethliou, “Load
Pocket Modeling,” Proceedings of the 2nd IASTED In-
ternational Conference: Power and Energy Systems,
Crete, 25-28 June 2002, pp. 50-54.
[5] N. Kamel and Z. Baharudin, “Short-Term Load Forecast
Using Burg Autoregressive Technique,” MIT Press,
Cambridge, 1992. [15] D. Genethliou, “Statistical Load Modeling,” Proceedings
of the 7th IASTED International Multi-Conference: Power
and Energy Systems, Palm Springs, 2003, pp. 88-91.
[6] S. Fan, L. Chen and W. J. Lee, “Short-Term Load Fore-
casting Using Comprehensive Combination Based on
Multi-Meteorological Information,” IEEE Industrial and
Commercial Power Systems Technical Conference, Vol.
9, No. 2, 1994, pp. 1-7. [16] K.-B. Song, Y.-S. Baek, D. H. Hong and G. Jang,
“Short-Term Load Forecasting for Holidays Using Fuzzy
Linear Regression Method,” IEEE Transactions on
Power Systems, Vol. 20, No. 1, 2005, pp. 96-101.
doi: 10.1109/TPWRS.2004.835632
[7] C. S. Chen, J. C. Hwang, Y. M Tzeng, C. W. Huang and
M. Y. Cho, “Determination of Customer Load Character-
istics by Load Survey System,” IEEE Transactions on
Power Delivery, Vol. 11, No. 3, pp. 1430-1436.
doi: 10.1109/61.517501 [17] L. Vehviläinen and T. Pyykkönen, “Stochastic Factor
Model for Electricity Spot Price—The Case of the Nordic
Market,” Energy Economics, Vol. 27, No. 2, 2005, pp.
351-367. doi:10.1016/j.eneco.2005.01.002
[8] T. W. S Chow and C. T. Leung, “Nonlinear Autoregres-
sive Integrated Neural Network Model for Short-Term
Load Forecasting,” IEE Proceedings of Generation,
Transmission and Distribution, Vol. 143, No. 5, 1996, pp.
500-506. doi: 10.1049/ip-gtd:19960600
[18] A. G. Baklrtzis, V. Petrldis, S. J. Klartzls, M. C. Alex-
ladls and A. H. Malssls, “A Neural Network Short Term
Load Forecasting Model for The Greek Power System”
IEEE Transaction on Power Systems, Vol. 11, No. 2,
1996, pp. 858-863. doi: 10.1109/59.496166
[9] S. Rahman and O. Hazim, “Load Forecasting for Multiple
Sites: Development of an Expert System-Based Tech-
nique,” Electric Power Systems Research, Vol. 39, No. 3,
1996, pp. 161-169. doi:10.1016/S0378-7796(96)01114-5 [19] R. L. Sullivan, “Power System Planning,” McGraw-Hill
International Book Company, New York, 1997.