Uncertainty quantification in predicting seismic response of high-speed railway simply-supported bridge system based on bootstrap

Abstract

Reliable and rapid prediction of seismic-induced response is crucial for post-earthquake repair or rescue operations. In this paper, a method for quantifying uncertainty in rapid seismic response prediction for high-speed railway simply-supported bridge system (HRSBS) was developed based on a Bi-LSTM neural network surrogate model and Bootstrap resampling to address the challenge of acquiring timely seismic responses for HRSBS and the inability to determine confidence intervals from a single prediction result. Epistemic and aleatory uncertainties were quantified in rapid prediction of seismic-induced responses for HRSBS. The applicability of Bi-LSTM model based on a single seismic time series for predicting seismic-induced responses of HRSBS was identified. The results indicated that the prediction intervals with the 95% confidence level obtained by the proposed method encompass the actual values. The misjudgment rates of component damage states are effectively reduced. The Bi-LSTM model employing a single seismic time series input is suitable for predicting the time-history curves of seismic responses of components but not suitable for predicting seismic-induced residual displacement of rail.