The Researching Of Power Load Forecasting System Based On Support Vector Machine

The Short-term Load Forecasting (STLF) is a daily routine in the operations of electric power system. Accurate STLF plays an important rate both in production planning, unit maintenance scheduling, economical and safe running, which directly influences the profit of the electric utility enterprises, and in the construction of electric power market, which need competition and economic ensure between the electric power corporations.The core of the STLF is the forecasting technology, which needs construct the mathematic model of the real load at the basis of the analysis of its speciality, and then design the effective algorithm to obtain accurate results.In recent years, with the rapid development of technology, new forecasting methods appeared, they offered well support to the researching of power load forecasting. The main forecasting methods include BP, fuzzy logic, ES, wavelet analysis etc. These forecasting methods are not only considering the shape of load, but also many external factors that affecting power load, for example, weather, date character and so on, as a result, the precision of forecasting has been increased, which have better development.Based on the project of unit maintenance scheduling of HeBei Province electric power network, this paper research on the daily average load of main power network, and a STLF model based on the support vector machine (SVM) is proposed.The content of STLF is to forecast the load of next day's even to next several months', and the most typical one is to forecast the curve of daily load of next day's, which is also the research object of this paper.In this paper, firstly the theoretical basis are introduced, which include the basic theories of STLF and the basic theories of SVM. Secondly, characteristics of short term load of main power network is analyzed, based of which the main influence factors are choosed, and the SVM forecasting model is established. To the SVM network, several kernel functions and parameters are tried, and by the way of statistic, the most suitable kernel and related parameters are choosed. Then the SVM model is trained and finally established by a improved training algorithm named Sequential Minimal Optimization (SMO). At last the model is used in STLF of HeBei power network. By the error analysis of the result and compare with BP network, the SVM model is proved to be more excellence, and fit for HeBei power system's STLF.