首页期刊在线 2022年（第25卷）第10期

基于Hopfield神经网络的磁浮列车悬浮系统本征非线性特性辨识

宋一锋1,2倪菲2林国斌2陈琛1,2刘永红3陶清宝3

Intrinsic Nonlinear Characteristics Identification of Maglev Train Levitation System Based on Hopfield Neural Network

SONG YifengNI FeiLIN GuobinCHEN ChenLIU YonghongTAO Qingbao

作者信息:

1.1.同济大学道路与交通工程教育部重点实验室, 201804, 上海；
2.2.同济大学磁浮交通工程技术研究中心, 201804, 上海；
3.3.茂盟工程技术股份有限公司, 203201, 上海
Affiliation:

The Key Laboratory of Road and Traffic Engineering of Ministry of Education， Tongji University, 201804, Shanghai, China
关键词:

磁浮列车;悬浮系统非线性模型;Hopfield神经网络;系统辨识;耦合振动
Key words:

maglev train;nonlinear model of levitation system;Hopfield neural network;system identification;coupled vibration
DOI:

10.16037/j.1007-869x.2022.10.026
中图分类号/CLCN:

U270.1+1∶U237
栏目/Col:

研究报告

摘要:

磁浮列车的悬浮系统具有本征非线性特性，在轨道不平顺以及非线性负载的作用下，非线性项的辨识和处理能力对于列车的悬浮稳定性具有重要影响。尤其是在轨道刚度较小时，在微小形变的作用下更易影响悬浮稳定性，而发生掉点、砸轨现象。从磁浮列车悬浮系统的非线性动力学建模出发，重点分析非线性项的系统参数辨识，基于Hopfield神经网络来构建误差函数和网络辨识方案。结合参数辨识误差函数和Hopfield网络的标准能量函数，得到相应的网络权值矩阵，并进一步给出相应辨识结果。通过数值仿真分析可以发现，基于本辨识方法所得到的结果与非线性项输出结果误差较小，可较好地拟合状态方程的含参非线性项，验证了辨识模型的可靠性。

Abstracts:

The levitation system of maglev train has intrinsic nonlinear characteristics. Under the action of track irregularities and nonlinear loading, the identification and processing ability of nonlinear term have an important influence on the suspension stability of the train. Especially when the track stiffness is weak, it is easier to affect the suspension stability under the action of small deformation, resulting in the phenomenon of point dropping and rail smashing. Starting from the nonlinear dynamic modeling of maglev train levitation system, the system parameter identification of nonlinear term is emphatically analyzed. Based on Hopfield neural network, the error function and network identification scheme are constructed. Combined with the parameter identification error function and the standard energy function of Hopfield network, corresponding network weight matrix is obtained, and the corresponding identification results are further derived. Through numerical simulation analysis, it can be found that the error between the results of the proposed identification method and the output results of the nonlinear term is very small, this identification method can better fit the parametric nonlinear term contained in the state equation, thus the reliability of the identification model is verified.