A new hybrid day-ahead peak load forecasting method for Iran's National Grid

Abstract

This paper presents a new hybrid forecasting engine for day-ahead peak load prediction in Iran National Grid (ING). In this forecasting engine the seasonal data bases of the historical peak load demand on the similar days with their weather information given for three cities (Tehran, Tabriz and Ahvaz) have been used. Wavelet decomposition is used to capture low and high frequency components of each data base from original noisy signals. A separate ANN with an iterative training mechanism which is optimized by genetic algorithm is employed for each low and high frequency data base. A day-ahead peak demand is determined with the reconstruction of low and high frequency output components of each ANN. Simulation results show the effectiveness and the superiority of the proposed strategy when compared with other methods for daily peak load demand forecasting in ING and EUNITE test cases. © 2012 Elsevier Ltd.