Partial discharge type recognition based on support vector data description

Traditional methods of insulation defect recognition often perform not well due to the changes in defect development influenced by changing environmental conditions, as well as the scattered and complex partial discharge (PD) data obtained. Therefore, we proposed a support vector data description (SVDD) algorithm for PD pattern recognition of gas insulated switchgear (GIS). Based on the principle of maximum interval of support vector machine (SVM) and the one-to-multiple principle of multiple classification methods, the optimal radius SVDD (OR-SVDD) algorithm was proposed to solve the problems of traditional recognition methods, including low recognition rate, recognition error, recognition miss, and long recognition time. Simulation and experiments prove that the OR-SVDD algorithm for identification performs better than the traditional SVM algorithm: in a comparatively shorter time and with higher recognition rate, all data objects are described correctly, while the outlying objects are recognized in the target data objects effectively. Therefore, it is concluded that the OR-SVDD algorithm has a good application prospects in both on-line monitoring of power equipment and PD pattern recognition.