关键词: 隧洞, TBM, 岩体质量, 岩体软弱破碎概率, 极端梯度提升决策树, 自适应粒子群优化算法, 概率校准

Abstract: To achieve rapid and quantitative characterization of rock mass quality during TBM tunneling, a case study is conducted on the Chuo′er river to Xiliao river water-diversion project. A large amount of measured TBM tunneling data is collected to systematically analyze variation patterns of tunneling parameters under different geological conditions. By examining the statistical characteristics and fluctuation patterns of key parameters, four representative basic tunneling parameters, namely advance rate, cutterhead rotation speed, torque, and thrust, are selected. In addition, three physical fusion indices, namely the torque-penetration index, thrust-penetration index, and specific energy ratio, are constructed based on energy and mechanical response relationships. These parameters and indices are used as model inputs. Subsequently, an adaptive particle swarm optimization (APSO) algorithm and a probability calibration (PC) method are introduced to optimize and correct the model. On this basis, a hybrid APSO-PC-extreme gradient boosting (XGBoost) model integrating intelligent optimization and probabilistic correction mechanisms is established, enabling rapid and quantitative sensing of rock mass quality at the TBM tunnel face. The results show that (1) the advance rate, cutterhead torque, total thrust, and cutterhead rotation speed exhibit a clear decrease in mean values and a marked increase in fluctuation during the transition from intact rock zones to weak and fractured rock zones; (2) compared with the basic XGBoost model, the proposed model improves the F₁ score by approximately 0.069 and reduces the Brier score by 9.73%, indicating higher prediction accuracy and stability; and (3) a dynamic probability threshold adjustment strategy is proposed for different surrounding rock grades, and the corresponding early-warning thresholds for weak and fractured rock are determined to be 0.32, 0.46, and 0.69 for Grades Ⅰ, Ⅳ, and Ⅴ rock masses, respectively.

Key words: tunnel, TBM, rock mass quality, weak fracture probability of rock mass, extreme gradient boosting, adaptive particle swarm optimization, probability calibration