Electric equipment breakdown is mainly caused by Partial Discharge (PD), so identification and diagnosis of PD are of great importance to the maintenance of electric equipment. This research analyzes the field TEV data of PD using SVM theory and establishes a model to detect and analyze PD, with related experiments verifying the validity of the model. According to the experience on site, we redesign the evaluation criteria and build a new model to guarantee that the identification rate reaches over 80%. The comparative experiment indicates that the SVM method proposed in this paper performs better on PD detection than BP Neural Network, laying foundation for further research.