Research On Power System Load Forecasting Based On Copula Related Structure

Load forecasting plays an important role in grasping the trend of load development,active distribution network optimization,promotion of renewable energy consumption,and ensuring the safe and economical operation of the power grid.Ultra-short-term load forecasting plays an important role in power system preventive control and emergency treatment.Therefore,the improvement of ultra-short-term load forecasting accuracy and efficiency has always been the focus of scholars worldwide.The correlation analysis between the input and output of ultra-short-term]oad forecasting model in the power system,the traditional linear correlation coefficient method can only reflect the linear correlation between variables,which can’t characterize the nonlinear correlation between them,and the application has some limitations.In this paper,the Copula related structure is proposed to describe the correlation.Copula related structures are mainly composed of the establishment of the marginal distribution function,the estimation of unknown parameter in the Copula function and the selection of optimal Copula functions.In the aspects of forecasting model and method,combining the entropy weight theory and grey correlation theory,this paper proposes the method of weighted gray relational degree method to choose the similar day based on the daily characteristic meteorological data and day-week type.In view of the fact that the correlation analysis in the existing prediction methods is only used for the model input selection and has not been used in the prediction model,a new prediction method(Copula prediction method)is proposed based on the Copula related structure and.the mixed penalty function method.The causes of the prediction error and the applicable scene of the new method are also explained.At the same time,this paper implements support vector regression machine and gray model as a comparison of Copula prediction method.And this paper aims to minimize the K-fold cross-validation error for the parameters selection of SVR based on Libsvm toolbox.Based on the control of insensitive loss parameters,grid search,genetic algorithm and adaptive weight particle swarm optimization algorithm are used respectively to select the penalty factor C and kernel parameter g of support vector regression machine.The results of numerical examples show that the forecasting accuracy of the Copula prediction method and the improved SVR based on Libsvm toolbox(APSO-SVR and GA-SVR)method proposed in this paper are significantly higher than those of the traditional SVR based on Libsvm toolbox(grid search method)and GM(1,1)model.The online prediction speed is consistent in the prediction efficiency,but the offline training time of the Copula prediction method is much smaller than the improved SVR based on Libsvm toolbox(mainly because the optimization algorithm takes longer time to select parameters and runs slowly.But setting the parameters arbitrarily will affect the forecasting accuracy).What’s more,Copula prediction method has more advantages in the analysis of the prediction error for its model mathematical structure is clearer.