Abstract: Accurate prediction of tunneling-induced ground surface settlement is crucial for ensuring safe construction and decision-making in tunneling projects. In this study, a physics-informed neural network (PINN) model is established for predicting shield tunneling-induced stratum deformation. This model is constructed by incorporating the relationship between tunnel convergence deformation and tunneling position into a deep neural network (DNN) framework. Considering the geological characteristics of multiple strata, a multi-physics-informed neural network (MPINN) model is proposed to represent the physical information of different strata in a unified framework. The results show that the MPINN model can highly reproduce the results by the finite difference method, and can accurately predict the tunneling-induced ground surface settlements considering the complex geological information of the composite strata. Due to the integrated physical mechanism, the MPINN model is applicable to the problem of tunnel-induced ground surface settlement, and it can be employed to predict the tunneling-induced ground surface settlement under different geological and geometric conditions. Based on the measured data, the proposed MPINN model accurately predicts the ground surface settlement curve of the monitored cross-section, thus it can provide a reference for the prediction and early warning of ground surface settlement during tunneling process.

Key words: physics-informed neural network (PINN), shield tunnel, ground surface settlement, machine learning, data-driven and physics-informed model