Seismic response prediction models of high-speed railway track-bridge coupling system considering varying pier heights

High-speed railway (HSR) track-bridge systems are complicated and difficult to model. When HSR lines traverse a vast region with varying elevations, the heights of the bridge piers vary greatly, which leads to a significant reduction in the finite element calculation efficiency. To rapidly predict the seismic responses of HSR trackbridge systems considering different pier heights, two efficient prediction models (a multi-input model and a multi-output model) were respectively proposed based on long short-term memory networks. The models adopted the sliding prediction form of stacked sequences. Combined with the random search optimization algorithm, the stacked datasets and ideal models were obtained dynamically. A case study on a CRTSII-type trackbridge system of 5-span HSR with different pier heights was then conducted, in which, the prediction abilities of the two prediction models were verified. The results show that the models possess strong generalization capability and stability, together with a simple hyperparameter tuning process and high computing efficiency. The multi-input model is more flexible in response prediction considering pier heights whereas the multi-output model has higher prediction efficiency. The architectures of the proposed models can be analogously applied to various scenarios, especially in the rapid prediction of seismic responses of structures considering multiple influence factors.