A new machine learning-basd fault diagnosis method of high voltage shunt reactor using spectral residual

High voltage shunt reactor is a primary electric power apparatus and plays a significance role in electric power system. In term of diagnosing the fault of high voltage shunt reactor, vibration signal is an easy acquired information. However, in the initial fault stage of high voltage shunt reactor, the characteristic information of vibration signal is weak and the noise interference is large. In this paper, we collect vibration signal from 24 sampling position on the four sides of high voltage shunt reactor under different kinds of state. We put forward a way which is based on spectral residual and machine learning to diagnosis high voltage shunt reactors fault. This method not only can effectively remove the weak direct current component in the frequency component but also can highlight the fundamental frequency component and frequency doubling component. In the experiment, we set up the fault diagnosis models by Support Vector Machine (SVM) and Convolutional Neural Network (CNN) respectively to compare the residual signals of raw vibration signal spectrum. The results show that compared to the raw vibration signal, the spectrum residual algorithm improved accuracy state by 9% and 10.75% respectively. Therefore, spectral residual can improve the fault diagnosis accuracy of high voltage shunt reactors.