为避免盾构轴线偏离引发衬砌管片错台、开裂等质量与安全问题，提出一种基于机器学习算法的盾构姿态智能预测模型与控制方法。以盾构掘进施工的实测数据为驱动，通过贝叶斯优化（BO）与支持向量回归（SVR）构建盾构姿态预测模型，挖掘施工参数-地层信息-盾构姿态三者间的非线性关系。结合模拟退火算法（SA）形成可控施工参数动态调整的盾构姿态控制方法，并将其应用于福州滨海快线南—三区间隧道的工程实践。主要结论如下： 1）经数据预处理、特征筛选及BO超参数优化，基于SVR的盾构姿态预测模型具备优异的预测性能和泛化能力； 2）结合SA算法进行可控施工参数调整时，需设置合理的优化规则，以确保所推荐的可控施工参数具备可操作性； 3）将姿态控制方法应用于南—三区间后续掘进施工以辅助纠偏，盾尾垂直偏差在10环掘进过程中由45 mm减至18 mm，实现了连续稳定纠偏。

Axis deviation during shield tunneling may cause quality and safety problems, such as segment dislocation and cracking. Therefore, the shield attitude must be accurately predicted and effectively controlled. An intelligent prediction model and control method for shield attitude based on machine learning are proposed in this study. First, a shield attitude prediction model based on Bayesian optimization(BO) and support vector regression(SVR) is established using monitored construction data. The nonlinear relationship among construction parameters, stratum information, and shield attitude is revealed. Second, using the simulated annealing(SA) algorithm, the shield attitude control method for the dynamic adjustment of controllable construction parameters is proposed and applied in the engineering practice of the South Park stationSancha Street station section of the Binhai express in Fuzhou, China. Some conclusions are drawn: (1) After data preprocessing, feature screening, and BO hyperparameter optimization, the SVR-based shield attitude prediction model demonstrates excellent prediction performance and generalization ability. (2) It is important to set some optimization rules reasonably within the SA algorithm to ensure the operability of the recommended controllable construction parameters. (3) The application of the proposed control method in the case study shows that the vertical deviation of the shield tail is reduced from 45 mm to 18 mm during the successive 10-ring tunneling process, realizing a continuous and stable deviation correction.