Prediction and Control of Existing High-Speed Railway Tunnel Deformation Induced by Shield Undercrossing Based on BO-XGboost

Abstract

The settlement of existing high-speed railway tunnels due to adjacent excavations is a complex phenomenon influenced by multiple factors, making accurate estimation challenging. To address this issue, a prediction model combining extreme gradient boosting (XGBoost) with Bayesian optimization (BO), namely BO-XGBoost, was developed. Its predictive performance was evaluated against conventional models, such as artificial neural networks (ANNs), support vector machines (SVMs), and vanilla XGBoost. The BO-XGBoost model showed superior results, with evaluation metrics of MAE = 0.331, RMSE = 0.595, and R2 = 0.997. In addition, the BO-XGBoost model enhanced interpretability through an accessible analysis of feature importance, identifying volume loss as the most critical factor affecting settlement predictions. Using the prediction model and a particle swarm optimization (PSO) algorithm, a hybrid framework was established to adjust the operational parameters of a shield tunneling machine in the Changsha Metro Line 3 project. This framework facilitates the timely optimization of operational parameters and the implementation of protective measures to mitigate excessive settlement. With this framework’s assistance, the maximum settlements of the existing tunnel in all typical sections were strictly controlled within safety criteria. As a result, the corresponding environmental impact was minimized and resource management was optimized, ensuring construction safety, operational efficiency, and long-term sustainability.