Online-learning support vector machine approach for short term load forecasting

A new approach for short term load forecasting based on online-learning support vector machine (SVM) algorithm is presented. The conventional implementations of support vector machine are usually inefficient for online learning because one must retrain from scratch as the training set is modified to ensure the forecasting accuracy. An accurate online-learning support vector machine algorithm, which efficiently updates a trained regression function whenever a sample is added to or removed from the training set, is proposed. The practical examples show that the online-learning support vector machine algorithm outperforms the conventional SVM with higher computing rate as well as better generalization.