GIS设备局部放电类型表观相似性及模式识别准确性的探讨

In order to reduce the occurrence of partial discharge recognition misjudgment and improve the accuracy of pattern recognition of GIS equipment, we discussed the resemblance of different partial discharges. Adopted two common classifiers to diagnose typical faults by the hierarchical identification method, generalized the types of partial discharge of GIS existing in the misjudgment phenomenon, and analyzed the resemblanc of the types of partial discharge. Moreover, we summarized the causes of misjudgment, for example, the instability of free metal particles is the causes of misjudgment, and the position of the sensor distance from the discharge source determines whether the misjudgment of the floating discharge can occur. At the same time, we studied the rationality of the traditional partial discharge type division of GIS. Finally, the accuracy of pattern recognition of the two classifiers was compared by the recognition rate and decision reliability. The research shows that the recognition confidence of protrusion discharge is highest, the identification rate of air bubbles in insulator discharge is highest, and the recognition effect of SVM is better than BP neural network. The research results lay a foundation for the high efficiency of the follow-up work, such as, partial discharge pattern recognition, risk assessment of partial discharge, etc.