Short term load forecasting with radial basis function network

The paper presents experiments with application of radial basis function (RBF) network to short term load forecasting (STLF) problems. The proposed regression model is used to forecast forty-eight hours ahead electric load. The model has been implemented on real data: inputs to the RBF are past loads, weekday and special-day coding and the output is the load forecast for the given hour. Ordinary RBF was applied in the experiments. The centers of the Gaussian basis functions were selected on the base of the quasi-Newton algorithm. Mean absolute percentage error of about 4% is derived from the data from the power system in Crete. The performance of the proposed model has been compared with simulations performed by the MLP network, and former models developed for the distribution company in Poland