Identification of Typical Partial Discharge Defects of Distribution System Equipment Based on Classification Learner

In the operation and maintenance of smart distribution system equipment, in order to improve the intelligence level of inspection and significantly enhance the inspection efficiency and distribution network equipment management, it is critical to effectively identify and warn the partial discharge faults caused by insulation defects. In this paper, a feature extraction and identification method are proposed by experimentally simulating the partial discharges of typical defects inside the distribution network equipment. Firstly, different discharge models are built and the UHF sensor is used to obtain the partial discharge signals, and then the partial discharge statistical characteristics are calculated by analyzing features of the 2D and 3D partial discharge patterns. Based on various supervised machine learning algorithms, 3 typical defects are identified by the statistical characteristics using the classification learner. The results show that the selected statistical characteristics: the maximum amplitude of discharge, the average amplitude of discharge and the ratio of positive and negative half-period total discharge amplitude can well reflect the discharge characteristics of different defects and can distinguish between different types of discharges. At the same time, the identification accuracy of the three discharge types is high, which can achieve an accurate diagnosis of the fault types.