Applications of support vector machine and improved k-Nearest neighbor algorithm in fault diagnosis and fault degree evaluation of gas insulated switchgear

摘要

Gas Insulted Switchgear (GIS) plays an important role in switch, control and protection and its safe and reliable operation is vital to the power system. However, its partial discharge failure usually cause serious consequences. It is necessary to monitor SF6 insulated power equipment operation state by detecting and analyzing the gas decomposition components with the aid of pattern recognition algorithm. The method of fault diagnosis and condition evaluation for GIS by measuring and analyzing SF6 decomposition components possess the ability to resist the electromagnetic interference and high sensitivity. Many existing algorithms proposed realize fault diagnosis and prediction of SF6 insulated power equipment mainly by calculating the ratio of the gas decomposition products which easily lead to low accuracy. This paper proposes a new diagnosis method based on the support vector machine (SVM) and improved k-NN algorithm which can accurately classify various partial discharge faults and have made predictions aimed at estimating their fault state levels and demonstrates its effectiveness and superior performance by applying an example of experiment.