The fault prediction of power transformer can realize the early warning of transformer fault

which is vital to ensure the normal operation of the electric power system. A transformer fault prediction model based on time series and support vector machine（SVM） is presented in this paper. The model is based on the autoregressive integral moving average（ARIMA） model in time series analysis

and the parameters p and q of ARIMA model are determined by genetic algorithm（GA）. The dissolved gas in transformer oil is predicted by the time series model after parameter determination. Then the SVM model optimized by grid search algorithm（GS） is used to predict the dissolved gas in oil. The running results show that the accuracy of transformer fault prediction model can reaches 89.66%

while the prediction accuracy of GM-SVM

ARIMA-SVM and GA-ARIMA-ANN is 58.62%

79.31% and 75.86%

respectively. So the method proposed in this paper has the higher prediction accuracy.