Identification of Partial Discharge Fault Type and Sensitivity Analysis of Joint Detection Based on Clustering Algorithm

Abstract

Partial discharge (PD) is the initial manifestation of insulation deterioration in power equipment. When PD accumulates to a certain extent, it will lead to equipment damage. PD detection in advance is an effective method to prevent insulation deterioration. In this paper, AE method, TEV method and UHF method are used to study the internal discharge, suspension discharge and surface discharge respectively, pointing out the limitations of a single detection method, and using clustering method to extract the discharge characteristic parameters of different defect types to classify. The results of K-means clustering show that the accuracy of internal discharge is 83%, surface discharge is 65%, and suspension discharge is 66%. FCM clustering results show that the accuracy of internal discharge is 63%, surface discharge is 60%, and suspension discharge is 50%. The results of hierarchical clustering show that the accuracy of internal discharge is 73%, surface discharge is 70%, and suspension discharge is 67%. From the perspective of sensitivity, TEV method has the highest sensitivity, AE method and UHF method have their advantages in the sensitivity of different discharge defects. In order to overcome the limitation of a single detection method, combined with the experimental results and the actual noise interference and local location requirements, the application of multi-means combined detection is analyzed to provide some reference for the selection of joint detection method. UHF/AE method can avoid electromagnetic interference. UHF/TEV method can avoid acoustic interference; TEV/AE method can improve the localization accuracy.