Significance and formulation of ground loss in tunneling-induced settlement prediction: a data-driven study

Ground loss is the volume of soil excavated in excess of designed excavation. It defines the boundary conditions for ground deformation field and thereby dominates the magnitude of tunneling-induced ground settlement. In practice, it is generally quantified by a parameter called ground loss parameter. To date, there is no unanimously agreed formulation for ground loss parameter despite its indispensable role in developing both empirical and analytical solutions for tunneling-induced ground settlement. Herein, a comprehensive field database is utilized to quantitatively assess existing formulations of ground loss, and to unravel its role in settlement prediction via inverse analysis. It reveals that remarkable errors can be generated by classical solutions for tunneling-induced ground settlement. This implies that the reliability of classical solutions for tunneling-induced settlement can be potentially improved with a more accurate formulation of ground loss. A data-driven formulation for ground loss is developed with aid of the random forest algorithm, and it can well capture the target value with an R-value equaling 0.84. The developed formulation is further implemented in the O’Reilly and New solution, yielding a hybrid model for settlement prediction. The hybrid model can accurately predict the actual settlement with an R-value of 0.84, outperforming the purely data-driven model and further confirming the accuracy of the proposed formulation of ground loss.