Automation, with neural network based techniques, of short-term load forecasting at the Belgian national control centre

The project described is aimed at automating the short-term load forecasting of the Belgian national power system control centre, usually done with a minimum lead time of 24 hours. It is hoped that the resulting system will improve the quality of forecasting methods, through a better modeling of the nonlinear relationship between load and climatic factors. In view of the various aspects of the problem, the authors intend to develop a hybrid neural network (ANN)-knowledge based system (KBS) application: the ANN will form the basis of the system and will make the forecast in normal situations; the KBS should manage exceptions and special phenomena as well as provide specific knowledge-based facilities. The authors focus on the development of a prototype for the ANN. The ANN is to be a model of the evolution of the load w.r.t. input parameters, therefore the ANN predicts the ratio between the load for one day and the day before, instead of the raw load value.