Fault Identification of Voltage Source Converter Valve Discharge Based on Mel-Spectrum and CNN

Le résumé fourni par la source

Voltage source converter valves are subjected to multiple internal and external stresses under long-term complex operating environments. There is a hidden danger of partial discharge caused by reduced insulation performance. Currently, there is a lack of effective means of discharge monitoring inside the converter valve hall. Usually, the occurrence of local discharge can only be investigated after the fault causes the converter valve to stop. Therefore, there is an urgent need to study the online monitoring technology to realize the active diagnosis of discharge faults. In this paper, a set of discharge fault identification methods based on key time-frequency feature extraction of mixed voiceprint signals is proposed. Firstly, the discharge signal is extracted from the mixed voiceprint pattern based on the fast independent component analysis method. Then the time-frequency spectral feature extraction is realized using Mel spectral analysis. Finally, a deep convolutional neural network is constructed to recognize the discharge fault type. In this paper, a voltage source converter valve discharge simulation test platform is built. The test results show that, compared with other algorithms, the method in this paper has superior accuracy while the computational speed is fast.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.