Intelligent prediction and visual optimization of surface settlement induced by earth pressure balance shield tunneling

Surface settlement (Ss) caused by earth pressure balance (EPB) shield tunneling is a threat to the safety of surface buildings and underground tunnel structures. To that end, a novel intelligent model named the beluga whale optimization-based kernel-extreme learning machine (BWO-KELM) model is proposed to predict the Ss. 148 monitored data with three categories of eight features collected from three tunnel projects are adopted to train the proposed models. The results of model evaluation indicate that the BWO-KELM model established by using eight features achieve the most satisfactory performance indices (determination coefficient (R²), root mean square error (RMSE), variance accounted for (VAF), and the prediction accuracy (U1)) in both training and testing phases, i.e., R² (0.9544 and 0.9481), RMSE (3.4948 and 4.3239), VAF (95.4380 % and 94.8875 %), and U1 (0.0833 and 0.0938). Then, the model interpretability is enhanced by using the feature importance (FI) and Shapley additive explanations (SHAP) method. The results show that the mean moisture content (MC) of the soil layers is the most important feature for the Ss prediction. Finally, a visualization program is established to improve the prediction efficiency of Ss, reliability of risk assessment, and rationality of tunnel designs. This paper provides a novel intelligent model paradigm and a visualization program with a strong application for solving the problems induced by EPB shield tunneling.