Railway Passenger Ride Comfort Evaluation Based on Deep Learning and Multibody Dynamics Simulation

The heavy reliance of many of the previously researched data-driven models on experimental data significantly limits their generalizability and robustness to the prediction of passenger ride comfort under similar vehicle-track system conditions. Consequently, this study attempts to demonstrate that rich data obtained from validated Multi-Body System (MBS) simulation models can achieve better generalizability to accurately predict passenger ride comfort under different vehicle-track system conditions. By varying the railway vehicle-track parameters iteratively, rich data is obtained and used for training and testing of a CNN-LSTM-SVM hybrid Deep Learning (DL) model which is proposed for the evaluation and prediction of railway passenger ride comfort. The proposed model is compared with CNN and CNN-LSTM models to demonstrate its robustness and high accuracy in the prediction of railway passenger ride comfort.