基于神经网络的轨道车辆振动预测

Rail Vehicle Vibration Prediction Based on Neural Network

摘要:

车体振动加速度是反映车辆振动状态及轮轨接触品质的重要参数；基于NARX神经网络,结合车辆动力学模型,构建了轨道不平顺激励下的车辆振动加速度神经网络预测模型；为提高模型的预测精度,运用遍历法确定了网络的时延阶数、隐节点等模型参数；基于车辆系统SIMPACK模型仿真数据的验证结果表明,模型输出与目标输出具有较高的相关性以及较小的均方根误差值,模型能够较好的预测出车体振动加速度的变化趋势；最后采用实测数据进一步验证了模型的预测性能,证明了构建的神经网络模型可以准确预测车体振动加速度输出,并有良好的泛化效果。

Abstract:

The vehicle vibration acceleration is an important parameter which reflecting the state of vehicle vibration and the quality of wheel/rail contact. A NARX neural network prediction model of vehicle system was established combined with the vehicle dynamics model,which reflecting the relationship between the track input parameters and the vehicle vibration. In order to improve the prediction precision,the number of time delay and hidden nodes are determined through traversing method. The validation results based on simulation data from SIMPACK model,show that the model output and target output is of high relevance and smaller root mean square error values,the model can well predict the change trend of vehicle vibration acceleration. The model validation based on the measured data,further proved that the proposed method can predict the output of vehicle vibration acceleration with high precision and perfect generalization.