基于NARX神经网络的轨道垂向不平顺估计

Assessment of vertical track irregularity based on NARX neural network

轨道不平顺是影响车辆平稳性和安全性的关键因素,因此及时掌握轨道不平顺的状态对保障列车运营具有重要意义。针对单个惯性量较难达到对不同波段不平顺的检测,提出一种采用非线性自回归神经网络(Nonlinear Auto-Regressive with exogenous Input Neural Networks,NARX)的轨道不平顺估计方法。以实测高铁轨道不平顺数据作为输入,通过车辆—轨道垂向耦合动力学模型仿真得到多个惯性量数据,再将归一化的多个惯性量仿真数据作为神经网络的输入,轨道不平顺作为输出,并用均方根误差和相关系数进行网络性能评价。仿真结果表明,NARX神经网络模型估计结果的均方根误差为0.028 9,相关系数为0.939 5,优于反向传播(BP)神经网络模型的均方根误差0.086 8和相关系数0.641 8,NARX神经网络拟合效果更好,表明本文所提方法能精确有效地实现轨道垂向不平顺估计。

Track irregularities are the key factors which influence vehicle stability and safety,it is important to grasp the state of track irregularity for guaranteeing the train operation safety. Due to it is difficult to detected different bands irregularities with a single inertial amount,a method was proposed to assess the vertical track irregularities based on Nonlinear Auto-Regressive with exogenous input Neural Networks（ NARX）. A coupling dynamics model of vertical vehicle-track interactions was developed with the actual measured track irregularity data from high-speed line as input to obtain the simulation a plurality inertia data. Then,NARX neural network,with the normalization simulation a plurality inertia data as the input and track irregularity as the output,was built to assessvertical track irregularity. The root mean square error（ RMSE） and correlation coefficients was applied to evaluate the network performance. Simulation results show that the RMSE of assessed results is 0. 028 9,the correlation coefficient is 0. 938 7 with NARX neural network model,which is higher than the 0. 086 8 and 0. 641 8 with Back Propagation（ BP） neural network model,respectively. It shows that the proposed method can accurately and efficiently assess the track vertical irregularity.