Seismic acceleration response prediction method of the PSCFST bridge based on TCN

Precast segment self-centering concrete filled steel tube (PSCFST) bridge has become a research hotspot in the field of infrastructure because of its excellent seismic performance and self-centering resilience. However, the highly nonlinear response of bridge structures due to their size and complexity poses a serious challenge to the accurate prediction of their seismic response, especially under extreme conditions such as earthquakes. This study presents a prediction method utilizing a temporal convolutional neural network prediction method to accurately forecasting the acceleration response of PSCFST bridge under seismic actions. The dataset was constructed by integrating data obtained from PSCFST single-span bridge shaking table tests and finite element model simulations. The Temporal Convolutional Network (TCN) model is employed as the training architecture, using acceleration time histories from diverse ground motions as inputs and the acceleration response of the bridge's superstructure as the training output. The TCN model employs causal expansion convolution to effectively capture long-term dependence in the time series data of the bridge structure's acceleration response. Furthermore, superposition of residual blocks enables the extraction of more profound nonlinear features at each data layer, thereby facilitating precise forecasting of acceleration responses in the bridge superstructure. The TCN model ensures capturing longer-span temporal correlations while reducing the model parameters, thereby achieving accurate and efficient prediction of bridge seismic response. Detailed comparative experiments were conducted among various algorithmic models, including Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Random Forest Regression (RFR). The results validate that the TCN model demonstrates higher prediction accuracy, better generalization capability, faster training speed, and fewer model parameters. These findings comprehensively demonstrate the superiority of the TCN model in predicting bridge vibration responses, offering an effective prediction methodology for enhancing the safety and reliability of bridge structures under seismic actions.