1. NARI Group Corporation(State Grid Electric Power Research Institute), Nanjing 211106, China;2. State Key Laboratory of Smart Grid Protection and Control, Nanjing 211106, China;3. School of Automation, Nanjing University of Science and Technology, Nanjing 210094, China; 4. Yunnan Electric Power Dispatching and Control Center, Kunming 650011, China; 5. Northwest China Branch of State Grid Corporation of China, Xi'an 710048 China;6. Electric Power Research Institute of State Grid Fujian Electric Power Corporation, Fuzhou 350007, China

Icing flashover fault of transmission line is one of important reasons leading to power grid failure. Current icing flashover researches mainly focus on the model study of flashover voltage of iced insulators. On one hand, the model of flashover voltage cannot fully reflect the insulator flashover voltage under the combined effects of all factors. On the other hand, due to the data acquisition error in the information of the transmission lines, the current characteristics model of icing flashover is difficult to be directly used in the forecasting of icing flashover fault. Considering the development of data mining technology, the partial mutual information(PMI)and support vector machine(SVM)are proposed to predict the icing flashover fault. Firstly, PMI is adopted to select the key factors for input variables. Secondly, SVM forecast model of icing flashover is established to train and predict the sample data. The simulation results show that the forecast method based on the PMI and SVM can predict the icing flashover more effectively, which can act as a reference for the ice defense of power grid.