Research on identification of occasional ultra-high frequency partial discharge in ultra-high voltage GIS based on K-Means

The on-site live detection process will inevitably suffer from environmental interference and reduce the quality of the collected partial discharge signal. In this paper, based on the live detection of the ultra-high voltage(UHV) substation 1000kV GIS, the original UHF signal is collected as the original data of partial discharge(PD) evaluation, and the K-Means clustering algorithm based on time domain characteristics is proposed to identify the UHF PD of UHV GIS. Firstly, the characteristic parameters of different UHF sensor channel signals are extracted as the clustering eigenvalues to cluster the original UHF signal. According to the clustering results, the clustering boundary criterion and correction coefficient are proposed to effectively eliminate the interference signals of inclusions in the original signal, obtain high-quality partial discharge data, and locate the defects. The positioning results of PD data obtained by this method and the results of manual processing are compared with the positioning results of ultrasonic signal detected in the field. The positioning accuracy of this method is improved by 3.88%, reaching 99.83%, which verifies the effectiveness and accuracy of the algorithm.