Modeling and forecasting short-term electricity load:A comparison of methods with an application to West Azarbaijan data

To improve the short term load forecasting methods, at first the successfull ones must be compared and evaluated. The goal of this paper is to compare some methods, which are applied successfully to daily load forecasting (DLF), for the hourly electricity load in the area covered by an electric utility of West Azarbaijan State (WAS) located in the northwest of Iran. The proposed input variables selection algorithms are based on mutual information (MI), Gamma test and correlation analysis for neuro-fuzzy modeling, with locally linear model tree (LOLIMOT) learning algorithm. Then the results are compared with our previous paper on WAS power system (WASPS) to get the final conclusion. The performance of each method is evaluated over the low, medium and peak loads of WASPS during the summer week from August 25 to September 1, 2008. The simulation results show that model dependent input selection methods can be better than the model independent ones for several hours ahead forecasting. Also, for each load group the corresponding suitable forecasting method is suggested.

Keywords:

Daily Load Forecasting , Input Selection , Mutual Information , Gamma Test , Correlation , Neuro-Fuzzy Modeling , LoLiMoT , Neural Networks