The Fuzzy Rough Set Reduction Properties Support Vector Machine Short-term Load Forecasting Method

With the deepening of the process of electricity market, to meet the safe operation of the power system, the prerequisite for reliable power system operation economy put forward higher requirements. As a short-term load forecasting system to order electric power dispatching department plans and arrangements important reference operating mode, the system is improved operating economy plays an important role. Because of the many factors that affect short-term load, coupled with the interference of random factors, to short-term load forecasting has brought no small difficulty. So, how reasonable and effective use of these factors, how to characterize the complex nonlinear relationship between different factors and load are two important issues currently facing.Article detailed analysis of fuzzy rough sets and support vector machines (SVM) characteristics of these two methods of data mining, given the former is able to extract data from a large number of implicit, there is the potential value of the decision-making information to effectively deal with the problem of information redundancy, the latter due to a strong non-linear fitting ability and good generalization performance and is widely used in load forecasting. This paper proposes a hybrid data mining method with fuzzy rough sets and SVM, and used to power the short-term load forecasting; This method uses fuzzy rough set theory attribute reduction algorithm to solve the power load of the many factors information expansion problems, eliminate irrelevant factors and decision information, and after the reduction factor as SVM input. The proposed method in the overall consideration of many factors affecting the load at the same time compressing the appropriate input variables eliminating the previous modeling process input variables selected based on the impact of subjective experience.In addition, the article also in all aspects of the work load forecasting for clues pretreatment on historical data, select SVM model and kernel function SVM model parameter optimization of several aspects were discussed. Pretreatment of historical data including correction of outliers, to fill the missing values of discrete and continuous data standardization sample data processing, these steps provide support for accurate data on load forecasting; SVM model nuclear select the function to predict the impact on the performance of the model is large, the analysis of the kernel function is relatively common, choose the RBF kernel function with good analytical nature; of issues that need to optimize the model parameters when creating SVM model, using genetic algorithm SVM model parameter optimization, prediction model based on SVM establish optimal parameters. Numerical example of the predictive effect of combining fuzzy rough set reduction SVM model and the conventional SVM model was not carried reduction compared to verify the effectiveness of the proposed method.