Incorporating Stock Index in a Support Vector Regression Model to Improve Short Term Load Forecasting Accuracy

Abstract

Recently emerged Support Vector Regression (SVR) model which incorporated temperature and calendar factors as input variables performed well in short term load forecasting. However, lacking of enough economic information in the regression model leads to poor prediction accuracy in months affected by the worldwide economic slowdown effect during years 2008 and 2009. To overcome this problem and improve load forecasting accuracy to better account for this economic slowdown effect, this paper proposes a new SVR approach that further incorporates the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) as additional input variable to SVR model. The proposed SVR approach focuses on predicting island-wide hourly electricity load demand from year 2008 to 2009 in Taiwan. This paper also compares the results with conventional SVR approach, and reveals that the overall load forecasting accuracy, in the best condition, can be improved by 6.6% in average.