Efficient seismic response modeling for train-bridge systems: A time series mixer approach

This paper introduces an innovative surrogate modeling approach for predicting seismic responses in train-bridge coupled (TBC) systems, utilizing the Time Series Mixer (TSMixer) neural network. The model features a multi-layered architecture with sliding time windows, ensuring continuous, real-time analysis. We refined key performance metrics to address limitations in traditional response evaluation, particularly in handling phase discrepancies. The TSMixer model demonstrates high accuracy across a variety of seismic waves and system parameters, making it a valuable tool for rapid prediction in urban seismic scenarios. By offering swift and precise assessments, this approach has significant implications for improving the design efficiency of high-speed rail systems under seismic conditions. This research advances the field by introducing a robust method capable of capturing the nonlinear dynamics of TBC systems, fulfilling modern engineering demands for speed and precision in seismic analysis.