Research Of Short-Term Load Forecasting Models Based On Least Squares Support Vector Machines

Short-term load forecasting is an important part of load prediction, which has great influence on the optimal combination, economic dispatching, optimum power flow, power market trade. Short-term load forecasting, which is in units of months, weeks, days, hours, is mainly used in power system dispatching. Accurate results of short-term load forecasting are not only beneficial to make an appropriate planned power trading volume, put forward a proper operation plan and bidding strategy, formulate reasonable power construction planning, improve the economic benefit and social benefit of power system, but also conductive to electricity plan management, saving coal and oil, and reduction of the generating cost. So finding an appropriate load forecasting method to improve the accuracy of prediction has important application value.The paper first introduces the present study of the load forecast at home and abroad, analyzes the characteristics of short-term load forecasting and its influencing factors, summarizes the common method of short-term load forecasting, and illustrates the pros and cons of various methods; Then details the theoretical basis and principles of support vector machine (SVM), derives the SVM regression model; According to historical load data and meteorological data of a certain area of Luo Yang, adopting the least squares support vector machine (LS-SVM) model, analyzing various factors related to the prediction, this paper has expounded the periodic law of load changes and amended the abnormal data in historical load, and normalized the related elements in load forecasting. However, two parameters in LS-SVM have great influence on the model, and the parameter selection is still based on experience which easily leads to errors in prediction. Finally, in this paper, particle swarm optimization is adopted in optimizing the model parameters, and regards the support vector machine parameter selection problem as a global search problem in a given space, and the average error of the test sample set as a judgment to the end of the algorithm, which has achieved automatic optimization of the parameters of support vector machines.This paper established a least squares support vector machine model based on particle swarm optimization, the optimization of the model parameters makes the accuracy of prediction increase. Practical calculation examples show that the forecast method in this paper is good in convergence, high accuracy in prediction and fast in training speed.