Wavelet time-frequency diagram and convolutional neural network-based fault diagnosis of commutation failure in HVDC transmission system

Abstract

Commutation failure is the most typical system failure of HVDC transmission system. Serious commutation failure will lead to subsequent continuous commutation failure, which will cause greater harm to the safe operation of power system. For the purpose of realizing the effective identification of the type about single commutation failure and continuous commutation failure in the HVDC transmission system, a new fault diagnosis method based on wavelet time-frequency diagram and convolutional neural network is proposed. Firstly, the collected inverter-side DC voltage and current original fault signals are continuously wavelet transformed to generate wavelet time-frequency diagrams as fault feature inputs; secondly, a convolutional neural network structure with dual softmax classifiers is proposed to realize the parallel judgment of six commutation failure fault types and single or continuous commutation failure problems. Finally, the CIGRE DC transmission standard test model is used for fault simulation test, and the results show that the method can identify the fault cause of single commutation failure and continuous commutation failure in HVDC systems.