Artificial neural network based short term load forecasting

Abstract

The artificial neural network approach has attracted a number of applications, especially in the field of power systems since it is a model-free estimator. Progress in application of artificial neural network technology to power systems in the areas of load forecasting, security assessment and fault diagnosis among others, has led to overcoming some of the limitations in the short term load forecasting problem. Improvements in forecasting are of paramount importance as an SLF program must guide the decision making of power system operators and also respond accurately and consistently to system changes. In this paper, a fast emerging field of ANN is described for a short term load forecasting (ANNSTLF) model, which implements the multiLayer feedforward backpropagation algorithm. The backpropagation algorithm with MLP model of artificial neural network is developed for the problem of short term load forecasting with a lead time of at least 24 hours. The best performance was obtained for the load forecasting for the Tuesday for which the maximum and average percentage error of 2.00% and 0.20% respectively was found. This came very close to the precision obtained by the human forecaster. The ability of a neural network to generalize the information presented is used to train the network.