Short-term Load Forecasting Based On Support Vector Machine

The daily operation and planning activities of an electric utility requires the prediction of the electrical demand of its customers. In general, the required load forecasts can be categorized into short-term, mid-term, and long-term forecasts. The quality of short-term load forecasts has a significant impact on the economic operation of the electric utility since many decisions based on these forecasts have significant economic consequences. The importance of accurate load forecasts will increase in the future because of the dramatic changes occurring in the structure of the utility industry due to deregulation and competition. This environment compels the utilities to operate at the highest possible efficiency, which as indicated above requires accurate load forecasts. Conventional short term load forecasting models includes statistical models, autoregressive and moving averages, spectral expansion techniques and Kalman filters. Because these models can not adapt easily to rapid changes of the load variation pattern, they have deficiencies. In the past decades, artificial neural networks (ANN) which is based on empirical risk minimization (ERM) principle have been successfully used for short term load forecasting. However, the selection of training data and network architecture significantly effects the performance of the neural network approach, and a trained network may become obsolete with abrupt changes in load conditions thereby making the forecaster inaccurate. Recently, a novel learning technique called support vector machine (SVM) has been put forward. SVM is based on structural risk minimization (SRM) principle and has shown many advantages compared with ANN. This thesis concentrates in implementing SVM to short term load forecasting. The statistical learning theory which is the basis of SVM is introduced firstly, then the detailed procedures of using SVM for classification problem and for regressive problem are given. The effects and selection of some important parameters of SVM are discussed. After that an online training algorithm for short term load forecasting based on SVM is proposed. In the online training algorithm, the regression function is updated by inputting new load data. Simulation results on practical power system show that the online training algorithm results in both smaller number of support vectors and faster convergence speed compared with the conventional SVM regression algorithms.