Based on Bayesian theory and online learning SVM for short term load forecasting

The paper adopts Bayesian theory to input feature selection for short term load forecasting (STLF). It makes use of the information from both samples and prior knowledge. In this way, not only can the over-fitting problem be effectively solved but also the model of forecasting can be simplified. Simultaneously, an online learning support vector machine (SVM) method for short-term load forecasting model is presented here. The method comprises incremental algorithm and decrement algorithm, which efficiently updates a trained regression function whenever a sample is added to or removed from the training set. So it is favorable for applications like online learning or leave-one-out cross-validation. The practical examples show that online learning support vector machine with input feature selection based on Bayesian theory outperforms other methods in both forecasting accuracy and computing speed.