Short term electric load forecasting via fuzzy neural collaboration

An important element of effective power system operation is the well-planned short term scheduling of power generating units. Power system operators use historical load data to schedule available generating units to meet hourly system loads in an economical and reliable manner. This paper describes how a Fuzzy Logic (FL) expert system is integrated with Artificial Neural Networks (ANN) for a more accurate short-term load forecast. The 24 h ahead forecasted load is obtained through two steps. First, a FL module maps the highly nonlinear relationship between the weather parameters and their impact on the daily electric load peak. Second, 12 ANN modules are trained using historical hourly load and weather data combined with the FL output data, to perform the final forecast. Comparisons made between this model, an ANN model, and an Autoregressive Moving Average (ARMA) model show the efficiency and accuracy of this new approach. (C) 2000 Elsevier Science S.A. All rights reserved.