SYNAPSIS
Power Systems now a days are expanding drastically and they are
interconnected to meet the growing demand and improve the reliability. For
example, the southern region of Indian Power System Network comprises of
APTRANSCO, TNEB, KSEB, KEB and Pondy as independent areas and draws
their central sector shares from the Central Generating Stations (CGS). These five
areas are interlinked as a Single Integrated System through the 400 KV Central Tie
Transmission Network. The operation of such network cannot be operated by
Power System operators without continuous information of load patterns and
power generation. Short-term Forecasting of loads for different conditions and
Load Dispatch planning needed to be established for handling satisfactorily the
operations of the complex practical Power Systems mentioned above.
Several electric power companies are now forecasting load power based
on conventional methods. Since the relationship between load power and factors
influencing load power is nonlinear, it is difficult to identify its nonlinearity by
using conventional methods namely extrapolation, exponential smoothing,
correlation methods involving regression analysis, weighted least square method
(WLSM) etc. The Artificial Neural Network (ANN) represents an effective
alternative approach rather than the conventional solution methods. Artificial
Neural Network neither depends on human expert nor the explicit pre-assumed
functional relationship between the past load data and the weather variables. It
allows representation of complex systems which are difficult or model with the
traditional techniques or knowledge based expert systems. Artificial Neural
Network based methods offer better forecasts than traditional Load Forecasting
techniques as the average error is low in the ANN based methods.
1
This project envisages mainly the comparison of “Weighted Least
Square Method” (Conventional) and “Artificial Neural Network” based Short-
Term Load Forecasting and its better usage in practical Power System operation
and control. The proposed Artificial Neural Network and Weighted Least Square
Methods are implemented by simulating the similar day data of practical loads for
the span of 8 weeks for one-hour-ahead load forecasting by adding a correction to
the selected similar day data. ANN implements a multilayer feed forward network
using sigmoid function as the activation function, Back Propagation algorithm as
the learning method and mean square error as the measure of error.
The software tool has been developed in ‘C’ language along with the
usage of ‘MATLAB’ for Load Forecasting with the help of Neural Networks
approach and test results are compared with the conventional method. Tables 5.2.1
– 5.2.4 gives the observed and predicted power demands and the difference
between them for 24 hours. From the tables and the figures 5.2.1 – 5.2.4 it has
been observed that the standard deviation has been reduced to approximately half
that of the conventional method (WLSM) with Neural Networks approach.
Chapter-1 deals with the role of forecasting in electrical demand
management and in electrical power systems. It also deals with the role of neural
networks in load forecasting. Chapter-2 deals with various approaches to load
forecasting. Chapter-3 gives the introduction, architecture and applications of
Artificial Neural Networks. Chapter-4 explains the method of short-term load
forecasting using ‘Weighted Least Square Technique’. Chapter-5 explains the
method of short-term load forecasting using ANN and simulated results in
comparison with weighted least square method. It also gives an approach to long-
term load forecasting. The conclusions are given at the end.
