抄録
This paper proposes a hybrid method of the fuzzy optimal regression tree (FORT) and the multi-layer perceptron (MLP) of an artificial neural net for short-term load forecasting. The regression tree (RT) is useful in discovering meaningful rules and classifying data so that the relationship between input and output variables is clarified. In this paper, a couple of strategies is developed to improve the performance of RT. One is to make use of tabu search to determine the globally optimal tree structure. The other is to introduce simplified fuzzy inference into RT to enhance the accuracy of the split value. As a prefiltering technique, FORT is used to classify data into one of clusters. MLP is constructed to forecast one-step ahead daily maximum loads for each cluster. The proposed method is successfully applied to real data.
