Study Of Ultra-short-term Wind Power Forcasting Based On Support Vector Machine

Wind Generation represents the one of future energy development trends.The volatility of wind power to power grid is difficult to operate safely for power grid.An accurate forecast of wind power is an effective means of easing power peaking and frequency adjustment's pressure,reducing power system spare capacity,which can improve the ability to accept the wind.Therefore,it has important engineering significance to study how to predict the wind power output and the trend of it accurately.This paper was made a more detailed study of the ultra-short-term wind power forecasting based on SVM(Support Vector Machine).First,this paper analyzed the physical characteristics of wind power and introduced the basic concepts and methods of requirements of wind power prediction.Secondly,as far as be concerned ultra-short-term wind power prediction problem,this paper introduces the basic method of support vector machine(SVM)and key researches two problems of input variables in the process of predicting model and parameter selection.On the one hand,using experimental method analysis statistically to select the best types of input variables.To select input variable dimension adopt autocorrelation analysis after the normalized date.On the other hand,using traditional cross validation and the grid search method to optimize parameter values in the support vector machine(SVM).Then,in order to form a short-term wind power prediction algorithm,quantum particle swarm intelligence algorithm is used to optimize further appropriate parameters of support vector machine(SVM).Finally,a lot of testing research on wind power prediction method,the results show that when predicting scale within 1.5 h,using cross validation and grid search can achieve ideal accuracy;when predicting scale for 4 h,the method of quantum particle swarm optimization with support vector machine(SVM)parameters is better and to achieve the accuracy of“Function specification of wind power forecasting”.Try to add the empirical mode decomposition algorithm to quantum particle swarm optimization based on supported vector machine(SVM)model for short-term wind power prediction,although accuracy increased after decomposition,but combine the synthesis of new subsequence according to the decomposition entropy weight manually is necessary,the workload is greatly increasing,whether apply to super short-term power in actual engineering calculation need further research.At the same time,the research work of this article laid the method foundation and prediction program module for the wind power prediction software development.