A New Artificial Neural Network Approach with Selected Inputs for Short Term Electric Load Forecasting

A new improved inputs selection approach for artificial neural networks (ANN) is proposed to model and forecast time series. The procedure makes use of the correlation analysis of the time series data both in selecting the inputs to the ANN and in deciding its structure. The correlation linking the time series to another prime input other than the past observations is taken into consideration. The improved inputs selection approach uses the available knowledge of existing physical models or transfer function model to add new selected inputs to the ANN in order to construct the best model for the time series. The approach is used in modeling and forecasting the load demand for the city of Zagazig in Egypt. The performance is compared with two other neural network models of the load time series fed with different set of inputs to show the superiority of the proposed approach. Copyright (C) 2008 Praise Worthy Prize S.r.l. - All rights reserved.