A novel optimized deep learning framework for acceleration prediction of cross-railway bridges under train-induced vibrations

This is the authors' abstract. We don't add key points for this paper.

Train operations will inevitably affect the operation of cross-railway bridges, and acceleration prediction plays an important role in early warning and abnormal state monitoring of the cross-railway bridges. However, train-induced and environmental noise interference result in the low signal-to-noise ratio of the measured acceleration; thus the complexity and randomness of the measured data present challenges in improving the accuracy of predictions. Finite element model-based prediction methods are complex to construct and struggle to accurately simulate the cross-railway bridge vibrations induced by train loads. To overcome the problem, a novel prediction framework for accurate acceleration prediction of cross-railway bridges based on time-varying filter-based empirical mode decomposition (TVFEMD), sparrow search algorithm (SSA), and long short-term memory (LSTM) is proposed. The TVFEMD technique is employed to decompose the measured vibration acceleration into multiple simple and stationary intrinsic mode functions (IMFs), which is helpful in better capturing the changing characteristics of acceleration. Furthermore, the SSA algorithm is introduced to adaptively determine the learning rate, iterations, and hidden layers of the LSTM model, thereby enhancing the acceleration predicting capability of the TVFEMD-LSTM. To verify the feasibility and accuracy of the proposed framework, a comparative study has also been carried out based on the LSTM and TVFEMD-LSTM models. The results demonstrate that the proposed method accurately predicts bridge vibration acceleration, providing a reliable basis for practical cross-railway bridge engineering.