Multisource information fusion for real-time prediction and multiobjective optimization of large-diameter slurry shield attitude

Abnormal large-diameter slurry shield attitudes (SA) lead to safety and quality problems in shield construction. To achieve intelligent prediction and optimization of the large-diameter slurry SA, a combination of Bayesian optimization categorical boosting (BO-CatBoost) and a multiobjective evolutionary algorithm based on decomposition (MOEAD) is proposed for the prediction and optimization of the tunnel large-diameter slurry SA, as well as the corresponding digital twin framework. Using the Yangtze River large-diameter slurry shield project in Wuhan as an example, data were collected for three soil conditions, namely, silty fine sand, angular gravelly soils and chalky clays, and gravelly soil. The results show that: (1) The R2 values of the three predicted types of soils are all above 0.9, the RMSE values are less than 1.2, and the MAE values are less than 0.9. (2) Based on the results of the SHapley Additive exPlanations (SHAP), the propulsive thrust had the greatest influence on the large-diameter slurry SA. (3) The large-diameter slurry SA optimization improves when the number of adjusted construction parameters increases, with a maximum optimization rate of 23.436%. In addition, a digital twin framework based on BO-CatBoost-MOEAD is proposed, which can effectively control large-diameter slurry SAs and improve overall project safety and quality.