Different Model Types for Short-Term Forecasting of Characteristic Load Points

This paper presents Neural Network solutions for 24 hours ahead electrical load forecasting of normal working days in Belgium. Our approach introduces several models, each one predicting a different characteristic load point of the day. The differences in behaviour between these load points led us to such differentiation as opposed to homogeneous load curve forecasting. Results compared to the human experts' forecasts indicate the neural networks perform adequately.