【摘要】 渣土改良是解决土压平衡盾构施工过程中刀盘“结泥饼”、刀具磨损等施工难题的有效措施.采用机器学习模型预测随地质条件变化的改良剂用量不仅可以降低上述施工风险，还弥补了试验法确定改良剂用量的滞后性.依托沈阳地铁四号线区间盾构项目，对1 396环砂土地层掘进数据进行预处理，将扭矩切深指数（TPI）和场切深指数（FPI）作为渣土改良效果判据并选择出优良数据集，建立Optuna-XGBoost改良剂预测模型.研究结果表明，Optuna算法在超参数优化上与其他算法相比有明显的优势；Optuna-XGBoost与其他5种预测模型相比，在地质条件变化的情况下可实现更高精度预测.更多还原

【Abstract】 Soil conditioning is an effective measure to solve the problems in the construction process of earth pressure balance(EPB) shield, such as cutter clogging and cutter abrasions. The use of machine learning models to predict soil conditioners varying with geological conditions can not only reduce the aforementioned construction risks, but also make up for the lag in determining the amount of modifier by test method. Based on the shield project of Shenyang Metro Line 4, the excavation data of 1 396 ring are preprocessed, and torque penetration index(TPI) and field penetration index(FPI) are used as the criteria for the soil conditioning effect to select good datasets. The Optuna-XGBoost model is established to predict the soil conditioners. The results show that Optuna algorithm owns obvious advantages over other methods in hyperparameters optimization. Compared with the other five prediction models, Optuna-XGBoost model owns higher accuracy under changeable geological conditions.更多还原