摘要
高铁列车关键部件的智能化诊断是当前电气化铁路智能检测与控制的关键一环,如何通过深度学习方法更精确、灵敏地诊断出高铁列车关键部件的故障已成为当前研究的热点。该文在卷积神经网络架构的基础上,提出了融合注意力机制与多尺度网络的故障诊断方法,选择4种乙丙橡胶(ethylene propylene rubber,EPR)电缆终端典型缺陷模型作为研究对象,通过所提出的融合注意力机制与多尺度网络的方法进行了终端典型缺陷的诊断。结果表明:所提出的方法对于EPR电缆终端故障的预测准确率达95.9%,相较于传统方法,预测准确率提高15%;同时,相较于其他深度学习方法,该方法对终端故障的预测准确率提高了5%,且所构建的诊断模型的训练迭代步数减少约40%。该文还提出了一种在极值寻优能力、网络稳定性、特征学习能力优于多尺度网络的网络模型,但相较于传统识别方法,深度学习方法的模型训练时间过长,如何缩短网络的模型训练时间还需要进一步深入研究。
Intelligent diagnosis of key components of high-speed railway train is a key part of intelligent detection and control of electrified railway.How to diagnose the faults of key components of high-speed railway train more accurately and sensitively through deep learning method has become a hot spot of current research.On the basis of convolutional neural network architecture,a fault diagnosis method combining attention mechanism and multi-scale network is proposed in this paper.Four typical defect models of ethylene propylene rubber(EPR)cable terminal are selected as the research object,and the proposed method of integrating attention mechanism and multi-scale network is used to diagnose the typical defects of terminal.The results show that the prediction accuracy of the proposed method for EPR cable terminal faults reaches 95.9%,which is 15%higher than that of the traditional method;at the same time,the prediction accuracy of terminal fault is improved by 5%compared with that of other deep learning methods,and the training iteration steps of the diagnosis model are reduced by about 40%.In this paper,a network model which is superior to multi-scale network in extreme value optimization,network stability and feature learning ability is proposed.However,compared with the traditional recognition methods,the model of deep learning method has an excessive long training time.How to shorten the network model training time needs further research.
作者
郭蕾
曹伟东
白龙雷
唐惠玲
项恩新
周利军
GUO Lei;CAO Weidong;BAI Longlei;TANG Huiling;XIANG Enxin;ZHOU Lijun(College of Electrical Engineering,Southwest Jiaotong University,Chengdu 611756,China;College of Physics and Optoelectronic Engineering,Guangdong University of Technology,Guangzhou 510006,China;Electric Power Research Institute,Yunnan Power Grid Co.,Ltd.,Kunming 650217,China)
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2021年第11期3872-3880,共9页
High Voltage Engineering
基金
四川省科技计划项目(青年科技创新研究团队项目)(2020JDTD0009)。
作者简介
郭蕾,1981-,女,博士,副教授,硕导,主要从事牵引供电系统建模和电能质量分析,E-mail:guolei_mail@swjtu.com;曹伟东,1995-,男,硕士生,主要从事电缆局部放电特性及故障诊断研究,E-mail:541046507@qq.com;白龙雷,1991-,男,博士生,主要从事电缆局部放电特性及故障诊断研究,E-mail:longlei0107@163.com;通信作者:周利军,1978-,男,博士,教授,博导,主要从事电气设备状态检测与故障诊断方面的教学与研究,E-mail:ljzhou10@163.com。
