基于时序数据的列车牵引系统故障预测方法  

Method of train traction system fault prediction based on timeseries data

作　　者：贺鑫来 孙庚 汪敏捷 翟逸男 陈岩霖 尹娴 冯艳红 HE Xinlai;SUN Geng;WANG Minjie;ZHAI Yinan;CHEN Yanlin;YIN Xian;FENG Yanhong(College of Information Engineering,Dalian Ocean University,Dalian 116023,China)

机构地区：[1]大连海洋大学信息工程学院,辽宁大连116023

出　　处：《现代电子技术》2025年第4期57-62,共6页Modern Electronics Technique

基　　金：大连海洋大学科研项目:轨道列车智能运维管理平台(2023001)。

摘　　要：牵引系统作为列车动能转换的关键模块,如果发生故障会给整车正常运行带来重大安全隐患,所以对其进行故障预测具有重要意义。然而,传统预测方法存在高度依赖人工经验判断、不能包含大量故障特征、预测精度不足等问题。为此,文中提出一种基于时序数据的故障预测方法。利用XGBoost算法对列车牵引变流器系统的故障特征进行计算和筛选,确定与变流器故障相关性较强的关键特征;采用贝叶斯优化的LSTM模型自适应地学习多源变量数据特征,利用时间窗对特征变量数据进行截取,实现对不同类型故障的预测。实验结果表明,所提方法在预测变流器场景下的6种故障时准确率可达到91%以上。As a key module for the conversion of train kinetic energy,traction system will bring great safety risks to the normal operation of the vehicle if it fails,so it is of great significance to predict its failure.However,traditional prediction methods have problems such as high dependence on manual experience judgment,inability to include a large number of fault features,and insufficient prediction accuracy.On this basis,a method of fault prediction based on timeseries data is proposed.The XGBoost algorithm is used to calculate and screen the fault features of the train traction converter system to determine the key features that are strongly correlated with the converter faults.The LSTM model optimized by Bayes is used to adaptively learn the multi-source variable data features,and the time window is used to intercept the feature variable data to realize the prediction of different types of faults.The experimental results show that The accuracy of the proposed method can reach more than 91%when predicting 6 kinds of faults in converter scenario.

关 键 词：牵引系统 故障预测 时序数据 XGBoost算法 LSTM 时间窗 

分 类 号：TN911.23-34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]