Enhancing the performances of a basic neural network tool for short-term forecasting of Italian electric load

In a previous study a basic multilayer perceptron artificial neural network for short-term forecasting of Italian hourly electric load has been implemented. The procedure has permitted adequate forecasting accuracies to be obtained for all the days with normal load conditions. However, the techniques adopted show some accuracy deficiencies while forecasting days with anomalous load conditions due to social factors (vacation periods, holidays, long weekends) and peculiar meteorological conditions as well. The paper illustrates the research activity conducted for enhancing the forecasting performances of the basic artificial neural network during these periods.