Electric load forecasting using parallel RBF neural network

Electric load forecasting plays a critical role for the reliable and efficient operation of power grids. In this paper we propose a load forecasting model using parallel radial basis function neural networks (RBFNN). The proposed implementation of RBFNN allows parallel computation therefore expedites the convergence of training process. The proposed model also employs a new hybrid chaotic genetic algorithm which introduces small scale chaotic variations into the best fit individuals in each iteration to locate an optimal set of parameters in RBFNN. We experiment the proposed load forecasting model with realistic demand data collected from both micro-grid as well as bulk grid levels, i.e., a local institutional micro-grid and one utility in the UK national grid. It is found that both cases can achieve acceptable forecasting accuracy with average error rate smaller than 4%, while forecasting the micro-grid load is more challenging than that of the bulk grid load due to the intermittent fluctuations within the former.