Based on the analysis of gases dissovled in transformer oil, it can disscover the fault of transformer.In order to fully reflect the relationship between the internal fault of transformer and the characteristic of gas,it put forward a pre input witch uesd the characteristic of gas of 5 concentration ratio for a total of 15 sets of characteristic and adopted RFE algorithm on the 15 feature selection,using the selected feature as the input for the SVM model.In the SVM model,it optimized the parameters by using the SA algorithm and provided the GUI interface.Finally,the RFE-SA-SVM model fault diagnosis rate is better than that of single model through data validation.