Application of Support Vector Regression to Temperature Forecasting for Short-term Load Forecasting

Abstract

This paper proposes a new method for temperature forecasting in power system short-term load forecasting. In recent years, power markets become more deregulated and competitive in power systems. As a result, it is of importance to deal with one-day ahead daily maximum load forecasting appropriately. To improve the forecasting model accuracy, it is a key to predict the weather conditions of input variables. In particular, daily predicted maximum temperature is one of the most important input variables. In this paper, an SVR-based method is proposed for maximum temperature forecasting in short-term load forecasting. It is an extension of SVM that makes use of the kernel trick to maximize a margin between different data sets. SVR corresponds to the regression version of SVM. The proposed method is successfully applied to real data of maximum temperature in Tokyo. A comparison is made between the proposed and the conventional ANN methods.