Short-term prediction of distribution network faults based on support vector machine

As the network end of power transmission, the distribution network (DN) directly determines the reliability of electricity energy supply. To predict failure accurately is important for increasing the repair efficiency of DN. Based on the failure data from DN in Beijing, the paper researches short-term DN failures prediction and proposes a fault judgment program based on weather and season factors. Failure is analyzed to determine its most important factors. Through support vector machine (SVM) algorithm and considering the relative meteorological factors, using the classification model predicts the number of failures in DN weekly, and establishes sub region classification forecasting model in week frequency with meteorological influence for DN failures prediction. Through the analysis for the number of DN failures data, we find the main influence factors are temperature, precipitation, wind and other meteorological factors. A short-term prediction program is tested lots of times with the data of DN failure. The practical data in Tongzhou district, Beijing, China, proved the effectiveness, precision and feasibility of the proposed method. The paper software used Matlab2014 and LIBSVM.