基于高压XLPE电缆逸出气体与多尺度卷积特征融合的电缆缺陷评估方法研究  

A cable defect assessment method based on high-voltage XLPE cable evolved gas and multi-scale convolutional features fusion

作　　者：孙韬 叶良鹏 张帆 张佳庆 过羿 周凯[2] 缪煦扬 SUN Tao;YE Liangpeng;ZHANG Fan;ZHANG Jiaqing;GUO Yi;ZHOU Kai;MIAO Xuyang(State Grid Anhui Electric Power Co.,Ltd.Electric Power Research Institute,Hefei 230601,China;College of Electrical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]国网安徽省电力有限公司电力科学研究院,安徽合肥230601 [2]四川大学电气工程学院,四川成都610065

出　　处：《绝缘材料》2025年第3期117-124,共8页Insulating Materials

基　　金：国网安徽省电力有限公司科技项目资助(52120523001C);安徽省自然科学基金资助项目(2208085UD13)。

摘　　要：本文提出了一种基于多尺度相关特征融合卷积神经网络的高压XLPE电缆缺陷评估方法。该方法基于数据驱动,通过训练卷积神经网络,建立特征气体浓度与缺陷类型之间的潜在关系模型,从而根据特征气体浓度诊断电缆缺陷。首先采用基于均值漂移的数据增强技术获取模拟数据,接着设计一种基于多尺度相关特征融合的1D卷积神经网络,最后利用该卷积神经网络基于模拟数据进行训练并进行缺陷识别。结果表明:该方法在模拟数据测试集和真实基础数据上的缺陷识别准确率分别为92%和88%,表明该方法能够有效地利用特征气体浓度实现电缆缺陷的诊断。This paper proposed a defect assessment method for high-voltage XLPE cable based on a multi-scale correlation feature fusion convolutional neural network.On the basis of a data-driven approach,this method established the potential relationship model between characteristic gas concentration and defect type by training a convolutional neural network,thereby diagnosing the cable defects based on characteristic gas concentration.Firstly,simulated data were obtained using a data augmentation technique based on mean shift.Then,a 1D convolutional neural network based on multi-scale correlation feature fusion was designed.Finally,the training and defect identification were carried out on the basis of simulation data by using the convolutional neural network.The results show that the method on the synthetic data test set and the real basic data achieves defect recognition accuracies of 92%and 88%,respectively.It is indicated that the proposed method can effectively utilize characteristic gas concentration to diagnose cable defects.

关 键 词：电缆早期故障诊断 特征气体分析 卷积神经网络 数据驱动 均值漂移 

分 类 号：TM247[一般工业技术—材料科学与工程]