Enhancing Substation Equipment Fault Diagnosis with Multi-Channel Voiceprint Detection

Substation equipment fault is an important concern in the whole power system. Usually different voiceprint signals are produced by the substation equipment under different faults, such as winding faults, overload operations, etc. How to correctly diagnose the operating status of substation equipment in advance based on these voiceprint signals is of great practical significance for avoiding power safety accidents. Although currently artificial intelligence technology is widely utilized in power equipment detection, it is not easy to detect substation faults in a timely and accurate manner. On the one hand, there are a lot of interfering sounds which are superimposed with the target voiceprint signals in the substation, resulting in a very low signal-to-noise ratio of the collected signal to affect the accuracy of fault identification. On the other hand, training an effective fault diagnosis model requires collecting and labeling a large amount of voiceprint data from complex power substations, which is a time-consuming and laborious work. To overcome above these challenges, this paper proposes a novel substation fault diagnosis model that takes advance of multi-channel voiceprint separation technology to reduce the interference of background noise on key voiceprint signals. At the same time, the self-supervised learning method is utilized to reduce the dependence on labeled data and improve the model’s ability to diagnose substation equipment faults.