基于图表示学习与知识蒸馏的电缆故障快速识别方法  

A Fast Cable Fault Identification Method Based on Graph Representation Learning and Knowledge Distillation

作　　者：余盛灿 余涛[1] 陈鑫沛 杨家俊 潘振宁[1] YU Shengcan;YU Tao;CHEN Xinpei;YANG Jiajun;PAN Zhenning(School of Electric Power Engineering,South China University of Technology,Guangzhou 510641,Guangdong Province,China)

机构地区：[1]华南理工大学电力学院,广东省广州市510641

出　　处：《电力信息与通信技术》2024年第4期11-20,共10页Electric Power Information and Communication Technology

基　　金：国家自然科学基金项目(52207105)。

摘　　要：在牵引供电系统设备故障预警中,准确并快速识别电缆的早期故障是智能化运维的关键技术。为挖掘特征构建的深层信息和解决工程部署迭代速率问题,文章提出一种基于图表示学习和知识蒸馏的电缆故障识别方法。首先,对电缆的电流信号采样分析,将时间序列下的特征信息用图特征进行动态显示和更新,采用卷积自编码器对特征图像实现降噪重构;然后,利用基于知识蒸馏的图卷积神经网络识别算法,构建教师-学生网络故障识别模型,研究在PSCAD仿真环境中搭建电缆故障模型采集过电流扰动信号;最后,通过实验对比证明所提模型的有效性和准确性,所提模型大幅提升模型迭代速率,同时增强在噪声扰动下的鲁棒性,具有工程应用价值。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.

关 键 词：电缆早期故障 卷积自编码器 图表示学习 知识蒸馏 

分 类 号：TN915.853[电子电信—通信与信息系统]