Physics-Informed Neural Networks for Settlement Analysis of the Immersed Tunnel of the Hong Kong–Zhuhai–Macau Bridge

Abstract

In this study, we propose a physics-informed neural networks algorithm that integrates a simplified physical model and neural networks for the settlement analysis and prediction of the immersed tunnel of the Hong Kong–Zhuhai–Macau Bridge (HZMB). The proposed method has high flexibility and generalizability because it integrates physical information into the loss function as a soft penalty constraint for neural network models. The uncertainty quantification is also realized with the Bayesian theorem and Markov chain Monte Carlo algorithm. A synthetic case study shows that the newly proposed method has high feasibility and efficiency for the inverse analysis of the tunnel settlement. The analysis of field data on the HZMB tunnel shows that the proposed method is applicable to practical engineering. The effect of the postconstruction settlement on the settlement prediction is discussed.