Explicit Prediction and Uncertainty Analysis of Slurry Density during Slurry Shield Tunneling in Sand

The maintenance of soil–water pressure equilibrium and operational stability during slurry shield tunneling in saturated sandy soils critically depends on the formation of a low-permeability filter cake at the tunnel face, where slurry density emerges as the dominant controlling parameter. The study, which leverages multigene genetic programming (MGGP) and monitors data from the Greenheart Tunnel, develops two predictive models for slurry density during drilling and standstill phases, unified via a binary variable mechanism. Comparative analysis with a physical model demonstrates the MGGP’s superior accuracy (R2 = 0.841 for drilling; R2 = 0.997 for standstill). To further assess the predictive uncertainty, quantile regression is applied, and two statistical measures of uncertainty are introduced [mean prediction interval (MPI) = 0.075 and prediction interval coverage probability (PICP) = 0.893 for drilling; MPI = 0.031 and PICP = 0.901 for standstill], confirming the robustness of the model. Global sensitivity analysis identifies tunnel boring machine (TBM) velocity and initial slurry density as primary levers governing chamber density dynamics. Consequently, meticulous management of TBM velocity and initial slurry density is essential to maintain chamber slurry density within the optimal operational range.