Fault Diagnosis Method of Disconnector Based on CNN and D-S Evidence Theory
Qi Wang, Kaipu Zhang, Sheng Lin
Abstract
Most of the faults that occur on high-voltage disconnectors are non-self-announcing hidden failures that are challenging to detect and diagnose. In this article, a deep learning-powered methodology for faults diagnosis of the disconnector is proposed. The methodology involves obtaining vibration and torque signals through a field test platform. Followed by the extraction of 2D features from these signals using wavelet packet transform and time-domain analysis, the latent characteristics of the signals are successfully identified and explored. A convolutional neural network (CNN) based model is subsequently established to endow the method with high accuracy in the fault classification tasks. Additionally, the Dempster-Shafer (D-S) evidence theory is employed to synthesize the decision and further enhance the model's performance. Test results have demonstrated the noteworthy superiority of the method in comparison to several existing technologies, thereby indicating its potential for extensive field application.