High-speed railway seismic response prediction using CNN-LSTM hybrid neural network

In addressing the challenges of analyzing seismic response data for high-speed railroads, this research introduces a hybrid prediction model combining convolutional neural networks (CNN) and long short-term memory networks (LSTM). The model's novelty lies in its ability to significantly improve the precision of fiber grating monitoring for high-speed railroads. Employing quasi-distributed fiber optic gratings, seven grating monitoring points were strategically placed on each fiber to capture responses of the track plate, rail, base plate, and beam during seismic activities. Using data from peripheral gratings, the model predicts the central point's response. A continuous feature map, formed via a time-sliding window from the rail's acquisition location, undergoes initial feature extraction with CNN. These features are then sequenced for the LSTM network, culminating in prediction. Empirical results validate the model's efficacy, with an RMSE of 0.3753, MAE of 0.2968, and a R2 of 0.9371, underscoring its potential in earthquake response analysis for rail infrastructures.