Intelligent Inversion of Multi‐Stratum Parameters in Shield Tunnels and Reliability Analysis of Tunnel Deformation Under Spatial Variability Conditions

Owing to deposition, weathering, and historical loading variations, the mechanical properties of underground rock and soil masses demonstrate significant spatial variability and stratified distribution. This study investigates the influence of multi‐layer soil spatial variability on ground settlement and tunnel reliability during shield tunnel construction by developing a refined stochastic finite element model. The CPSO‐TLOOA‐Stacking hybrid intelligent algorithm optimizes the inversion of multi‐stratum mechanical parameters based on the measured surface settlement data from tunnel engineering and the Conditional Tabular GAN (CTGAN) data extension framework. Utilizing the Karhunen–Loève (K‐L) series expansion method and random field theory, a joint analysis framework of stochastic finite element and probability statistics is constructed to evaluate the impact of spatial random field parameters of different soil layers on formation deformation and failure probability. Coupled with the Hamiltonian Monte Carlo‐Subset Simulation algorithm, the reliability of tunnel deformation under conditions of cross‐correlated random fields with multiple surrounding rock parameters is effectively assessed. The results indicate that the R2 value of the expanded dataset fitted by the CPSO‐TLOOA‐Stacking hybrid intelligent algorithm is 99.46%, and the relative error between the dataset and the measured value is 0.7%. The Hamiltonian Monte Carlo‐Subset Simulation algorithm significantly enhances the calculation efficiency of tunnel deformation reliability and provides valuable guidance for shield tunnel construction and design.