<p>Shield tunnel launching adjacent to a river is a typical high-risk problem in urban underground engineering. As a novel reinforcement method for launching shafts, TRD (Trench cutting Re-mixing Deep wall) has been widely applied in shield launching projects, while its stability evolution and failure mechanism remain insufficiently understood. To address this issue, a coupled theoretical model based on Mindlin’s elastic solution and the rigorous limit equilibrium method is proposed for stability analysis. Considering the high computational complexity and time consumption of conventional coupled analysis methods, the XGBoost machine learning algorithm is further introduced based on the proposed theoretical model to establish a nonlinear mapping relationship between input parameters and the safety factor, thereby enabling rapid intelligent assessment of excavation stability under complex working conditions. Furthermore, the SHAP method is employed to improve the interpretability of the prediction model by identifying the contribution of different parameters to the safety factor and validating the results against theoretical sensitivity analyses. The results indicate that the shear strength reduction coefficient of the TRD-reinforced zone (<i>k</i>_<i>e</i>) has the most significant influence on the safety factor, with a contribution ratio of 65%. The horizontal vibration coefficient (<i>k</i>_<i>c</i>) contributes approximately 20%, while the influences of shield thrust (<i>P</i>) and advancement distance (<i>L</i>) are relatively minor. During shield advancement, the evolution of the safety factor can be divided into three stages: rapid adjustment, high-impact dominance, and low-impact stabilization. The system gradually reaches a stable state after the shield advances approximately 15&#xa0;m. The proposed method can provide theoretical support for the optimal design of TRD reinforcement, safety control during shield launching, and rapid on-site construction safety management.</p>