Comparison Study of Neural Network Methods for Electricity Forecasting

Abstract

Load forecasting has an important meaning in economic and secure operation of power systems. So, numerous methods are proposed to enhance the accuracy of load forecasting. In this paper, we introduce these methods and proposes two methods that combines neural networks and time series model. First one is neural networks with global ARMA model and the other one is neural networks with local ARMA model which uses moving window. And then we compares the ordinary artificial neural network model and our proposed models by simulation studies and load demand data from France. Since load demand data has trend and seasonality, we use differenced data to fit the ARMA model. Finally, we compares the forecasted values up to 24 hours to see the accuracy of each models.