Multi-Objective Intelligent Optimization of Excavation and Support for Super-Large Cross-Section Small Clear Spacing Tunnels in Urban Sensitive Areas: An Integrated Study of Numerical Simulation and Machine Learning

Abstract

Coastal subsea tunnels often traverse shallow heterogeneous strata, creating significant deformation risks. Deformation control is therefore essential for safe urban construction. This study integrates FLAC3D, SSA-LSSVM, and NSGA-II for multi-objective excavation-support optimization. The Qingdao Jiaozhou Bay Second Subsea Tunnel serves as the case study. The results show that the large-span double central diaphragm method with vertical temporary supports provides the best deformation control. Ground surface grouting provides better control of ground and building settlements than the advanced pipe roof scheme, while the combined reinforcement scheme provides only marginal additional benefit. Sensitivity analysis identifies shotcrete strength, thickness, bolt length, bolt spacing, and steel-arch spacing as key parameters. The SSA-LSSVM surrogate model achieves R2 > 0.99 in reproducing the numerical simulation results of crown deformation, ground surface settlement, and building settlement within the studied parameter space. The optimized scheme reduces crown deformation, ground surface settlement, and building settlement by 25.27%, 17.09%, and 15.69%, respectively, with a 22.39% cost increase. Field monitoring of the construction section, which adopted the recommended excavation and reinforcement strategy but not all the exact optimized support parameters, indicates that the deformation remains within acceptable limits and may provide preliminary engineering verification.