Optimization of small spacing tunnel excavation method and prediction model for deformation and stress

As China’s railway network expands, the complexity of tunnel engineering has increased, particularly for large cross section and small spacing tunnels. These tunnels exhibit significant excavation spans, multiple construction stages, and interdependent construction processes. However, the effects of different tunnel excavation methods on the deformation and stress distribution during the excavation process are still not clear. Therefore, based on a small spacing railway tunnel in Chongqing, this study employs monitoring, numerical simulation, and machine learning methods [gray wolf optimizer (GWO), particle swarm optimization (PSO), and genetic algorithm (GA)] to analyze tunnel deformation and stress. A method for automatic parameter optimization was proposed, which improved the accuracy of the machine learning prediction model (the error has decreased from 9.38% to 0.67%). The results indicate that the center cross diagram and double side drift methods reduce deformations and stress compared to the bench method (reduced by 55.03% and 54.36%, respectively). The GWO model demonstrates superior predictive performance for vault deformations and stress; the R2 of GWO was increased by 0.06 compared to that of PSO and by 0.047 compared to that of GA.