作者：方明宽, 宁静, 陈春俊

作者单位：1. 西南交通大学机械工程学院，四川 成都 610031;
2. 轨道交通运维技术与装备四川省重点实验室，四川 成都 610031

关键词：高速列车;小幅蛇行;蛇行失稳;EEMD;固有模态;LSTM

摘要：

高速列车运行时出现横向蛇行失稳严重威胁到列车运行安全。目前大多数方法主要为蛇行失稳的在线识别，而忽略从正常到蛇行失稳过程中的小幅蛇行阶段。为此，提出一种EEMD-CNN-LSTM方法来预测小幅的演变趋势，进而分析是否会发生蛇行失稳。并且基于相关系数与能量特征提出一种新的指标来挑选EEMD的最优模态。该方法首先通过集合经验模态分解(ensemble empirical mode decomposition, EEMD)对原始信号分解得到n阶固有模态(intrinsic mode functions, IMF)，然后挑选m阶最优的固有模态分量，最后将最优模态分量输入到构建的CNN-LSTM神经网络模型，并输出结果。根据在线实测数据实验结果：提出的方法能够准确预测小幅蛇行的变化趋势，预测准确率达到100%，且计算速度优于EEMD-LTSA方法，证明该方法的有效性。

Lateral hunting instability prediction model of high-speed train based on EEMD-CNN-LSTM
FANG Mingkuan, NING Jing, CHEN Chunjun
1. School of Mechanical Engineering, Southwest Jiaotong University, Chengdu 610031, China;
2. Technology and Equipment of Rail Transit Operation and Maintenance Key Laboratory of Sichuan Province, Chengdu 610031, China
Abstract: The safety of train operation is seriously threatened by the lateral hunting instability of high-speed trains. At present, most of the methods focus on the identification of hunting instability, but ignore the small amplitude hunting state from normal to hunting instability. Therefore, a method based on EEMD-CNN-LSTM is proposed to predict the trend of small amplitude hunting, and then to analyze whether the hunting instability will occur. An optimal IMF selection index of EEMD based on correlation coefficient and energy features is proposed. Firstly, the original signal is decomposed by  EEMD to obtain the IMFs, then the optimal IMFs is selected. Finally, the m-order optimal IMFs are input into the CNN-LSTM neural network model. The experimental results show that the evolution state of small amplitude hunting can be accurately predicted by EEMD-CNN-LSTM, and the prediction accuracy of the proposed method is 100%, which shows effectiveness of the method.
Keywords: high-speed train;small amplitude hunting;hunting instability;EEMD;intrinsic mode functions;LSTM
2021, 47(7):79-83  收稿日期: 2020-07-28;收到修改稿日期: 2020-08-09
基金项目: 国家自然科学基金项目（51975486，51975487）；四川省科技计划资助（2020JDTD0012）
作者简介: 方明宽(1994-),男,湖北孝感市人,硕士研究生,专业方向为智能化状态监测及故障诊断
参考文献
[1] 王晓东, 宁静, 陈春俊. 1D CNN 和 LSTM 高速列车横向稳定性状态识别研究[J]. 中国测试, 2020, 46(11): 25-30
[2] 王福天. 车辆动力学[M]. 北京: 中国铁道出版社, 1981.
[3] 铁路工程动态验收技术规范: TB10761—2013[S]. 北京: 中国铁道出版社, 2013.
[4] 铁道客车行车安全监测诊断系统技术条件: TB/T 3188—2007[S]. 北京: 中国铁道出版社, 2007.
[5] SOUZA A F D. Influence of the wheel and rail treads profile on the hunting of the vehicle[J]. Transact of the ASME, 1985, 107(1): 167-174
[6] SOUZA A F D, CARAVAVATNA P, 汪廷椿. 转向架蛇行振动的非线性分析[J]. 国外铁道车辆, 1986, 4(1): 5-14
[7] 崔万里, 宁静, 种传杰, 等. 基于EMD-ISOMAP高速列车小幅蛇行异常特征提取[J]. 中国测试, 2016, 42(12): 105-110
[8] 叶运广, 宁静, 种传杰, 等. 高速列车转向架蛇行失稳的MEEMD-LSSVM预测模型[J]. 铁道学报, 2018(1): 38-43
[9] 叶运广, 宁静, 种传杰, 等. 基于MEEMD香农熵-LSSVM的高速列车蛇行失稳诊断方法[J]. 计算机应用研究, 2017(4): 1097-1100
[10] ZHAO M H, ZHONG S S, FU X Y, et al. Deep residual shrinkage networks for fault diagnosis[J]. IEEE Transactions on Industrial Informatics, 2010, 16(7): 4681-4690
[11] HOCHREITER S, SCHMIDHUBER J. Long short-term mmemory[J]. Neural Computation, 1997, 9(8): 1735-1780
[12] 冉伟, 宁静, 陈杨, 等. 基于EEMD-SVD-LTSA的高速列车蛇行演变特征提取框架[J]. 电子测量技术, 2019, 42(5): 1-5