Prediction of distribution network malfunction based on meteorological factors

Abstract

Distribution network malfunction often causes serious economic losses and social negative impact. If we can effectively predict the numbers of distribution network malfunction, it would provide reliable data basis for the promptly maintenance and power repair. In this paper, three kinds of analysis algorithms, stepwise regression analysis, zero-inflated Poisson regression and support vector regression (SVR), are used to fit the numbers of malfunction. We utilized the lightning data and meteorological factors as independent variables, and utilized the external malfunctions and natural malfunctions as the dependent variables to establish the prediction models. At the end of this paper, the accuracy of these three methods is discussed. The relative root mean square error(R-RMSE) of each prediction method is calculated. We found that the external malfunctions using support SVR to obtain the best results, and natural malfunctions are better with the zero-inflated Poisson regression model.