A neural network short-term load forecaster

Abstract This paper presents a neural network based approach to short-term load forecasting, which plays an important role in the day to day operation and scheduling of power systems. A four-layer feedforward neural network, trained by a back-propagation learning algorithm, has been applied for forecasting the hourly load of a power system. In this paper, the performance of the network is compared with some carefully chosen experimental methods. This new approach promises to provide results unobtainable with more traditional time series methods. It is shown that, with careful network design, the back-propagation learning procedure is an effective way of training neural networks for electrical load prediction. The choice of transfer function is an important design issue in achieving fast convergence and good generalization performance. The network is trained on real data from a power system and evaluated for short-term forecasting with hourly feedback. The network learns the training set nearly perfectly and shows accurate prediction with 1.07% error on weekdays and 1.80% error on weekends.