One-Hour Ahead Load Forecasting Based on Wavelet Neural Networks

Abstract

This paper presents the improvement on accuracy and reliability of the load forecast model. It is well-known that characteristics of a load series is a non-stationary data, which is a constraint for the load forecast methods to achieve accurate and robustness responses. To overcome this limitation, a synergized method between wavelet transform and artificial neural network is proposed for short-term load forecasting. The modeling processes such as minimizing distorted data due to convolution operator of the wavelet transforms, model inputs and neural network design are presented. The proposed method is tested using historical load data of independent system operation New England. The results of the proposed model significantly outperform either accuracy or robustness results over neural network model.