Short term load forecasting using a novel recurrent neural network

Abstract

Short term load forecasting is essential to the operation of electricity companies. It enhances the energy-efficient and reliable operation of power system. Neural networks (NNs) have powerful nonlinear mapping capabilities. Therefore, they have been used to deal with predicting, in which the conventional methods fail to give satisfactory results. A novel Recurrent neural network (RNN) is proposed in this paper. Many types of computational intelligent methods are available for time series prediction. The novelty of this RNN lies in the usage of neurons instead of simple feedback loops for temporal relations. There is flexibility to use any type of activation functions in both feed forward and feedback loops. Number of hidden neurons can be changed on case to case basis for maximum accuracy. The performance of the RNN is demonstrated to be better than several other computational intelligent methods available.