基于LSTM的脱轨系数间接测量方法研究  

Indirect Measurement of Derailment Coefficient Based on Long Short⁃Term Memory

作　　者：孙昭意 陈建政[1] 吴越[1] 乔笑冉 SUN Zhaoyi;CHEN Jianzheng;WU Yue;QIAO Xiaoran(State Key Laboratory of Rail Transit Vehicle System,Southwest Jiaotong University Chengdu,610031,China)

机构地区：[1]西南交通大学轨道交通运载系统全国重点实验室,成都610031

出　　处：《振动.测试与诊断》2025年第1期57-64,200,共9页Journal of Vibration,Measurement & Diagnosis

基　　金：四川省科技计划资助项目(2023YFH0049)。

摘　　要：为了对脱轨系数进行准确而高效的在线监测,基于长短时记忆(long short-term memory,简称LSTM)网络算法,对脱轨系数间接测量问题展开研究。首先,选取实测地铁车辆上易测得的物理信号作为输入,研究不同输入信号与脱轨系数的相关性;其次,建立网络模型的样本数据集并对脱轨系数划分相应标签,通过网络训练实现标签的准确划分;最后,筛选标签中脱轨系数临近安全限值的样本数据作为网络的输入,实现脱轨系数具体数值的间接测量,并在另一条试验线路上验证了该方法的有效性。研究结果表明:将轴箱横、垂向振动加速度信号、车体摇头角速度信号、一系悬挂垂向压缩位移量以及车辆运行速度全部作为网络输入信号,进行脱轨系数间接测量的效果最好;基于LSTM方法可以有效地对脱轨系数标签进行分类,且能够对危险区域内脱轨系数进行精确测量,满足实际线路上脱轨系数间接测量及评价的要求。该研究成果有助于地铁车辆长期运营时脱轨系数间接测量以及安全监测等。With the continuous increase of train speed,the requirement of the running safety of the vehicle is also improving.The derailment coefficient is an important indicator for evaluating the running safety,it is par⁃ticularly important to monitor online accurately and efficiently.To achieve the goal,the indirect measurement of derailment coefficient is studied by long short-term memory(LSTM).Firstly,the easily measurable signals from the vehicle are selected,and the correlation between different input signals and derailment coefficient is verified.The sample datasets of network are established and the labels of derailment coefficient are divided cor⁃rectly through network training.The sample data of derailment coefficient close to the safety limit is chosen and the data is substituted into model to predict,which achieves the indirect measurement of derailment coefficient.Then verifying the efficiency of the method on another test line.The results show that axle box lateral and verti⁃cal acceleration,the yaw angular velocity of the car body,the vertical displacement of primary suspension and vehicle running speed are all used as input signals to indirectly measure the derailment coefficient with the best effect.Based on the LSTM method,the labels can be effectively classified,and the derailment coefficient in the dangerous area can be accurately measured,which meet the requirements of indirect measurement of derailment coefficient on the actual lines.This research result is helpful for indirect measurement of derailment coefficient in long-term operation of the safety monitoring of metro.

关 键 词：脱轨系数 振动加速度 长短时记忆网络 间接测量 

分 类 号：U270.1[机械工程—车辆工程]