Predicting earth pressure balance (EPB) shield tunneling-induced ground settlement in compound strata using random forest

The prediction and control of the ground settlement induced by shield tunneling in compound strata is a crucial challenge for tunnel excavation. In this regard, a finite element model for the analyses of ground surface set-tlement induced by earth pressure balance (EPB) tunneling was established and validated by the field test. The influence of single influence factors, such as rock proportion in the face, support pressure, and tunneling angle, on the surface settlement above the tunnel face was systematically studied. To investigate the joint influence of different factors on ground deformation mechanism, the multi-factor analysis based on field data and the mul-tiple regression (MR) method was conducted. However, the prediction accuracy of the MR method was limited. Therefore, a prediction approach of shield tunneling-induced settlement in compound strata based on a random forest (RF) was developed. Tunnel geometry, geological features, and operation parameters were investigated as input variables in the RF model to achieve this goal. The applicability effect of the RF-based method was proved by utilizing two data sets (i.e., the settlement ahead of tunnel face and final settlement). The results indicated that the prediction accuracy of shield tunneling-induced settlement in compound strata by the RF-based model is higher than the MR method. The relative importance analysis of variables indicates that advance rate, the cover depth of the tunnel ring, and pressure in the chamber were the most influential features for the settlement ahead of the tunnel face, while the thrust of the cutter head and the grout quantity were the most significant variables for the settlement after the segment assembling is completed. Overall, the conducted study could provide a helpful reference for the efficient excavation of EPB tunnels across compound strata.