Peak electricity load forecasting using online support vector regression

Abstract

Load forecasting is essential in planning and operation of smart grid systems. Short term load forecasting (STLF) plays an important role from the generation perspectives. Existing methods of STLF are needed to remodel each time when new training data are included in the training set. This degrades overall efficiency of the system. In this paper we propose a method of STLF to update the trained model without remodeling by using online support vector regression (SVR) algorithm. In online SVR, changes of model parameters due to new training samples are updated in finite number of steps so that it meets the SVR optimization criteria. Real world data set of residential buildings in an area of Surrey, British Columbia is used to verify the system's performance and the proposed system showed promising results.