Short-Term Load Forecasting Using Neural Network Technology

Managing electrical energy supply and generating exact amount of electricity are complex tasks. The most important part of managing energy supply is that of forecasting the future load demand. This is usually carried out by constructing a number of models on relative information such as previous load demand data. In this connection, this paper examines and analyzes the use of Artificial Neural Networks (ANNs) as a forecasting tool, along with traditional forecasting tools such as regression methods for forecasting the power demand for three days ahead, and this comes under Short-Term Load Forecasting (STLF). Specifically, the ability of ANN models trained by Back Propagation (BP) algorithm to predict future electricity load is tested. Results obtained from neural network model are compared with the traditional forecasting method, multiple linear regression analysis. In this study, a multi-layer feed forward neural network model has been proposed for implementing BP learning. The model works on the principle of Generalized Delta Rule, uses Sigmoid Logistic function as the activation function and calculates error using the method of Root Mean Square Error (RMSE). General software for forecasting was developed in MATLAB software and was successfully trained and then tested for calculating the values of daily power demands for the next three days. The model was trained and tested with different architectures with the same input pattern and different values of learning rate and constant momentum factor. The result obtained from BP method is compared with that of the regression method