Prediction of surface settlement caused by synchronous grouting during shield tunneling in coarse-grained soils: A combined FEM and machine learning approach

This paper presents a surrogate modeling approach for predicting ground surface settlement caused by synchronous grouting during shield tunneling process. The proposed method combines finite element simulations with machine learning algorithms and introduces an intelligent optimization algorithm to invert geological parameters and synchronous grouting variables, thereby predicting ground surface settlement without conducting numerous finite element analyses. Two surrogate models based on the random forest algorithm are established. The first is a parameter inversion surrogate model that combines an artificial fish swarm algorithm with random forest, taking into account the actual number and distribution of complex soil layers. The second model predicts surface settlement during synchronous grouting by employing actual cover-diameter ratio, inverted soil parameters, and grouting variables. To avoid changes to input parameters caused by the number of overlying soil layers, the dataset of this model is generated by the finite element model of the homogeneous soil layer. The surrogate modeling approach is validated by the case history of a large-diameter shield tunnel in Beijing, providing an alternative to numerical computation that can efficiently predict surface settlement with acceptable accuracy.