Abstract
As tunnel engineering expands into complex geological areas, the stability and mechanical performance of the lining structure under complex loading conditions have become critical issues in engineering design. This paper takes the Yeheshan Tunnel as the engineering background and combines FLAC3D numerical simulation with the XGBoost ensemble learning model to systematically analyze the stress and deformation responses of single-layer steel fiber shotcrete linings under different surrounding rock conditions. Based on 32 sets of actual mix ratios and strength test data, a compressive strength prediction model was constructed. The numerical simulation results show that the radial displacement at key locations, such as the arch crown and invert, is significantly reduced in steel fiber concrete, with the maximum deformation controlled within 20 mm, an improvement of about 33% compared to ordinary concrete. At the same time, in the machine learning prediction model, the PSO-XGBoost model demonstrates the best performance. In the 28-day compressive strength prediction, it achieved R(2) = 0.979, RMSE = 1.237 MPa, and MAE = 1.047 MPa, significantly outperforming traditional models. The study results reveal the significant advantages of steel fiber concrete in improving the overall bearing capacity and ductility of the lining structure and validate the high accuracy and efficiency of intelligent optimization algorithms in predicting material mechanical properties. This provides reliable data support and methodological guidance for the design and material selection of supporting structures in complex tunnel engineering projects.