基于深度学习的轨道不平顺估计车体横向加速度研究

Research on estimating lateral car-body acceleration with track irregularity based on deep learning

针对轨道不平顺影响行车安全和乘坐舒适性的问题,开展轨道不平顺和车体振动响应之间关联关系的研究.根据轨道不平顺与车体横向加速度数据之间的时空传递特性,基于卷积神经网络(Convolutional Neural Network,CNN)与长短期记忆网络(Long Short-Term Memory,LSTM),建立一种CNN-LSTM组合模型.首先,对轨道不平顺与车体横向加速度之间的相干性进行分析,同时探究列车运行速度对车体横向加速度的影响,进而确定模型的输入数据;然后,对模型进行优化,确定CNN-LSTM组合模型参数细节;最后,利用CNN-LSTM组合模型、LSTM模型和多体动力学仿真模型对车体横向加速度进行估计,并与实测的车体横向加速度进行对比分析.研究结果表明:CNN-LSTM组合模型预测的平均均方根误差值为0.046,相比于LSTM模型、多体动力学仿真模型的平均均方根误差值分别减少了0.006和0.026,说明了CNN-LSTM组合模型能够对车体横向加速度进行有效估计.

To address the problem of track irregularity affecting driving safety and ride comfort,this study investigates the correlation between track irregularity and vehicle vibration response.Based on the spatiotemporal data transmission characteristics between track irregularity and lateral car-body ac⁃celeration,a combination of Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)is proposed,forming a CNN-LSTM combined model.Firstly,the coherence analysis of track irregularity and lateral car-body acceleration is performed,and the influence of train speed on lat⁃eral car-body acceleration is explored to determine the input data of the model.Then,the lateral car-body acceleration is estimated using the CNN-LSTM combined model,LSTM model,and multi-body dynamics simulation model,and the results are compared with the measured lateral car-body ac⁃celeration.The results show that the CNN-LSTM combined model achieves an average root mean square error of 0.046,which is 0.006 and 0.026 lower than that of the LSTM model and the dynamics simulation model,respectively.This demonstrates the effective estimation capability of the CNN-LSTM combined model for lateral car-body acceleration.