Evolving wavelet-based networks for short-term load forecasting

Abstract

A new short-term load forecasting (STLF) approach using evolving wavelet-based networks (EWNs) is proposed. The EWNs have a three-layer structure, which contains the wavelet (input-layer), weighting (intermediate-layer), and summing (output-layer) nodes, respectively. The networks are evolved by tuning the parameters of translation and dilation in the wavelet nodes and the weighting factors in the weighting nodes. Taking the advantages of global search abilities of evolutionary computing as well as the multi-resolution and localisation natures of the wavelets, the EWNs thus constructed can identify the inherent nonlinear characteristics of the power system loads. The proposed approach is verified through different types of data for the Taiwan power (Taipower) system and substation loads, as well as corresponding weather variables. Comparisons of forecasting error and constructing time reveal that the performance of the EWNs could be superior to that of the existing artificial neural networks (ANNs).