Regularisation of neural networks for improved load forecasting in the power system

A regularisation procedure for neural-network reduction in order to obtain the best results for load forecasting in a power system is presented. The OBD pruning method was applied in the solution. The numerical experiments were concentrated on the prognosis of the load in the power system. Two kinds of experiments are described: a 24-hour forecast and the forecast of the daily mean of the load. It was shown that the application of the regularisation of the neural network employed for prediction resulted in a significant improvement of the forecasting accuracy