Comparison of serial and parallel approaches using artificial neural networks for Algerian short term load forecasting

Abstract

Knowing that electrical load is a non storable resource; short term electric load forecasting becomes an important tool to optimise dispatching of electrical load in regular system planning. Several techniques have been used to accomplish this task, from traditional linear regression and Box-Jenkins to artificial intelligence approaches such as Artificial Neural Networks (ANN). This work presents a comparative study of serial and parallel ANN approaches for forecasting 168 hours ahead using a multiple linear regression model as a benchmark for comparison. The results obtained by the latter method, are compared with the ANN serial and parallel developed approaches. These models were trained solemnly on past load consumption data, given by the Algerian national electricity company. This results in Nonlinear Autoregressive Models (NAR), however once the approach validity is proven, the addition of exogenous inputs can only improve model results.