A load curve based fuzzy modeling technique for short-term load forecasting

A modeling method is suggested in this paper that permits building fuzzy models for short-term load forecasting (STLF). The model building process is divided in three parts: (a) the structure identification based on a fuzzy C-regression method, (b) selection of the proper model inputs which is achieved using a genetic algorithm based selection mechanism, and (c) fine tuning by means of a hybrid genetic/least squares algorithm. To obtain simple and efficient models we employ two descriptions for the load curves (LC's), namely, the feature description for the premise part and the cubic B-spline curve for the consequent part of the rules. The simulation results demonstrate that the suggested model exhibits very good forecast capabilities. The suggested model is favorably compared with the ANN model in terms of prediction accuracy, robustness and model complexity. (C) 2002 Elsevier Science B.V. All rights reserved.