Research On Support Vector Regression In Prediction Of The Short-term Load Of Power System

With the rapid development of power industry and the advance of the power management technology, research on load forecasting of power system raises more and more attention. How to predict load forecasting of power system effectively has become an important issue. This thesis introduces support vector regression (SVR) into the short-term load of power system.Based on statistical learning theory, support vector machine (SVM) not only has the advantages of simple structure, but also the generalization ability, so it can solve small sample learning problems well. The theoretical study and the practical application of SVM are growing rapidly currently, and SVM has become focus in the field of machine learning. The main work of this thesis is summarized as followings:(1) The thesis summarizes the current study of load forecasting of power system and SVM, and lists some representative models of load forecasting of power system. Then we describe the theory of support vector regression in details.(2) There are two difficulties in the practical application of SVM: feature selection and parameters optimization. In order to obtain better learning performance and prediction accuracy, the model we design relates to two ideas :①feature selection: we choose effective feature based on genetic algorithm;②parameters optimization: we improve the basic PSO algorithm, and use the improved algorithm for parameters optimization.(3) Taking the advantages of GARCH model in dealing with the volatility problems and excellent generalization ability of SVR model. This thesis proposes a new combination model of SVR and GARCH to improve the prediction accuracy.The thesis presents a simulation example on the basis of the modeling process of combined prediction model above, then we verify the validity of the model through the analysis of experimental results.