Objective Train body vibration acceleration is a core indicator for evaluating the operational safety and stability of trains. Accurately predicting the lateral and vertical vibration accelerations of the train body is the key to realizing active safety early warning for trains. Therefore, it is necessary to develop a method for predicting train body vibration status that integrates multi-source data.

Method A train body vibration acceleration prediction model based on an LSTM （long- and short-term memory） neural network is proposed. First, a dataset covering multi-dimensional features such as operating speed, track irregularities, curve parameters, wheel profile parameters, and suspension system parameters is constructed. A genetic algorithm is utilized to obtain the optimal values of the hyperparameters for the LSTM neural network. A high-precision train body vibration status prediction model is established, taking multi-dimensional features as inputs and train body vibration acceleration as the output. Measured data are adopted to verify the accuracy of the prediction model.

Result & Conclusion  The RMSE （root mean square error） values of the proposed prediction model for the lateral and vertical vibration accelerations of the train body are 0.007 m/s² and 0.013 m/s², respectively. The prediction performance of the proposed model for both lateral and vertical vibration accelerations of the train body is superior to that of the comparison models. Especially in predicting the lateral vibration acceleration of the train body, the RMSE is reduced by 95.9%, which verifies the effectiveness and advanced nature of the proposed prediction model.