GIS Breakdown Defect Diagnosis Based on EPPS-BT-SVM-RFE

During on-site GIS withstand voltage tests, accurate identification of defect types is essential for efficient troubleshooting. This study proposed a diagnostic model based on recursive feature elimination, a binary tree support vector machine, and the evolutionary predator and prey strategy. RFE removes redundant variables and retains effective features related to different defect states. The binary tree structure divides the original multi-class task into hierarchical binary classification tasks. This approach reduces decision complexity. To further improve classifier performance, EPPS optimizes SVM hyperparameters globally. The results showed that the proposed model achieves an overall classification accuracy of 94%. It outperforms several other methods. These results demonstrated its effectiveness for defect diagnosis during on-site GIS testing.