Short-term Load Forecasting Research Based On SVM

The Electronic Power System acts as the core department of the nation, takes a veryimportant role on the country’s economic development and the people’s daily life, thereforethe Electronic Power System should run on the secure and economic operation, and it needsthe basics of the short-term load forecasting provided for, and the short-term loadforecasting determines the Electronic Power System will run correctly or not. The SupportVector Machine (SVM) is a novel machine learning method with many merits such as thesimple structure, the global optimum and strong generalization ability, which shows uniqueadvantages in solving the small sample, non-linear and high dimensional problems. This paper,using the SVM’s advantages of non-linear processing and generating ability, is proposed toaccomplish the improved short-term load forecasting of the power system, which is comparedwith the traditional methods. And the result shows that both the forecasting precision andexecuted speed are improved. Consequently the study is significant in reality and is valuablein practice. This paper is organized as follows:Firstly the research status of the short-term load forecasting is describedcomprehensively and systematically, and both the advantages and the disadvantages of themethods are analyzed. Secondly, the basic SVM theory and SVM regression are introduced indetails. Thirdly, the principle and feature of the Sequential Minimal Optimization (SMO)algorithm is elaborated, and the traditional and improved SVM regression models are putforward. And the following part is the implementation of the short-term load forecastingbased on SVM regression algorithm, including data pre-processing, the construction andselection of kernel function, and parameter optimization method, and the current solutions areprovided respectively. In particular, for a series of SVM-based improvements and some mixedforecasting methods consisting of SVM with other algorithms, a comprehensive summary isgiven, from the perspective of the mechanism about SVM algorithm being applied into loadforecasting to the elevation of prediction accuracy and speed, and the results under differentsamples situation are analyzed. Finally, some key issues about the short-term load forecastingbased on SVM are summarized, and some recommendations are given.