基于CNN-ADABOOST的车载设备故障诊断  

Fault diagnosis of on-board equipment based on CNN-ADABOOST

作　　者：宋鹏飞 陈永刚[1] 王海涌[2] SONG Pengfei;CHEN Yonggang;WANG Haiyong(School of Automation and Electrical Engineering,Lanzhou Jiaotong University,Lanzhou 730070,P.R.China;School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,P.R.China)

机构地区：[1]兰州交通大学自动化与电气工程学院,兰州730070 [2]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《重庆邮电大学学报（自然科学版）》2023年第6期1174-1182,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金项目(52062028)。

摘　　要：列控车载设备故障量大,其诊断依赖于专家经验且故障信息数据分布不平衡。以CTCS3-300T型列控车载设备中CTCS-3控制单元(ATPCU)记录的故障词条为样本,提出一种卷积神经网络(convolutional neural networks,CNN)与自适应增强(adaptive boosting,Adaboost)算法结合的车载设备故障诊断方法。考虑到列控车载日志为半结构化文本,采用skip-gram模型对故障词条进行处理生成词向量,通过CNN进行特征提取,将CNN作为基分类器,通过Adaboost更新样本权重与基分类器权重生成强分类器对分布不平衡的故障数据进行分类,实现列控车载设备的故障诊断。实验数据为某铁路局电务段原始车载日志,研究结果表明,基分类器迭代步数与数目是影响模型性能的关键因素,通过确定基分类器数目与单个基分类器训练步数,可显著提升模型对于不平衡数据样本的分类能力。研究成果为车载设备故障诊断提供了一种智能且有效的方法。The train control on-board equipment has a large number of faults,and its diagnosis relies on expert experience and the distribution of fault information data is unbalanced.Taking as a sample the fault entries recorded by the CTCS-3 control unit(ATPCU)in the CTCS3-300T train control vehicle equipment,a fault diagnosis method for on-board equipment combining convolutional neural network and adaptive enhancement algorithm is proposed.Considering that the train control vehicle log is semi-structured text,the skip-gram model is used to process the faulty entry to generate the word vector,and then the feature extraction is carried out through CNN,and the CNN is used as the base classifier,and then the sample weight and base are updated through Adaboost.The weight of the classifier generates a strong classifier to classify the unbalanced fault data to realize the fault diagnosis of the train control on-board equipment.The experimental data is the original on-board log of an electric service section of a railway bureau.The research results show that the number of iterations and the number of base classifier are the key factors affecting the performance of the model.By determining the number of base classifiers and the number of training steps for a single base classifier,the classification ability of the model for unbalanced data samples can be significantly improved.The research results provide an intelligent and effective method for fault diagnosis of on-board equipment.

关 键 词：车载设备 故障诊断 词向量 卷积神经网络 自适应增强算法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术] U284[自动化与计算机技术—计算机科学与技术]