An Expert System for Short Term Load Forecasting by Fuzzy Decision

Abstract

There have been many researches about load forecasting, which can be classified into two classes, time series method and factor analysis method. But the former is not adaptive for sudden change of a correlated factor and the latter is inefficient as the factor estimation itself is not easy. To make matters worse, both of them are not good for the estimation of special days. It is because the load forecasting is not a problem modeled precisely in mathematics, but a problem requires experience and knowledge of an expert. In this viewpoint, an expert system is proposed which can use complicated experience of an expert. Days are classified into two groups, ordinary and special. Moving average method is used for the estimation of ordinary days and fuzzy decision method is applied to special days to decide the application order of correlated factors. Load estimation for special days is taken by compensation for moving average. The compensation value is calculated from factors considered in order of importance. In fuzzy decision, both membership and utility of a factor should be considered. Fuzzy factors such as rain, hot, holiday are expressed by fuzzy membership. Utilities are composed of relative differences between moving average and real data. The results are appropriate. Percentage error is about 1.4 for ordinary days and about 2 for special days.