A hybrid short-term load forecasting with a new data preprocessing framework

Abstract

This paper proposes a hybrid load forecasting framework with a new data preprocessing algorithm to enhance the accuracy of prediction. Bayesian neural network (BNN) is used to predict the load. A discrete wavelet transform (DWT) decomposes the load components into proper levels of resolution determined by an entropy-based criterion. Time series and regression analysis are used to select the best set of inputs among the input candidates. A correlation analysis together with a neural network provides an estimation of the predictions for the forecasting outputs. A standardization procedure is proposed to take into account the correlation estimations of the outputs with their associated input series. The preprocessing algorithm uses the input selection, wavelet decomposition and the proposed standardization to provide the most appropriate inputs for BNNs. Genetic Algorithm (GA) is then used to optimize the weighting coefficients of different forecast components and minimize the forecast error. The performance and accuracy of the proposed short-term load forecasting (STLF) method is evaluated using New England load data. Our results show a significant improvement in the forecast accuracy when compared to the existing state-of-the-art forecasting techniques.