Phase-space based short-term load forecasting for deregulated electric power industry

This paper describes the application of a phase-space embedding concept to artificial neural network (ANN) based short-term electric load forecasting. Embedding parameters for electric load time-series were determined using the method of integral local deformation. In the reconstructed phase-space modular ANN predictor was trained to predict loads up to five days ahead in one-hour steps. It was found that addition of temperature and cycle variables to the phase-space based input variable set improved forecasting accuracy. The overall number of input variables was much smaller than in the similar cases reported in the literature. In this manner the size of historical data set needed for training was significantly reduced. Forecasting errors were comparable to or better than the ones reported for the similar cases Such characteristics make the approach attractive for short-term load forecasting in the deregulated electric power industry