A Fast Cable Fault Identification Method Based on Graph Representation Learning and Knowledge Distillation
In the early warning of equipment failures in traction power supply systems,accurate and rapid identification of early cable failures is a key technology for intelligent operation and maintenance.In order to mine the deep information of feature construction and solve the problem of engineering deployment iteration rate,this paper proposes a cable fault identification method based on graph representation learning and knowledge distillation.First,the current signal of the cable is sampled and analyzed,and the feature information under the time series is dynamically displayed and updated with graph features.The convolutional autoencoder is used to reconstruct the feature image with noise reduction,and then the graph convolution neural network based on knowledge distillation is used.The network identification algorithm builds a teacher-student network fault identification model.The study builds a cable fault model in the PSCAD simulation environment to collect overcurrent disturbance signals,and proves the effectiveness and accuracy of the model through experimental comparisons,and greatly improves the model iteration rate,and at the same time enhances the robustness under noise disturbances,and has engineering application value.
cable early faultconvolutional auto-encodergraph representation learningknowledge distillation
国家科技期刊平台
NSTL
万方数据
