A Modified Inverse Analysis Method for Soil Parameters Considering the Spatial Structure of Soil Layers

When evaluating the impact of foundation pit excavation on adjacent existing metro structures through numerical simulation, it is essential to adopt a reasonable constitutive model and parameters. This study proposes an innovative inverse analysis method for soil parameters that considers the spatial structure of soil layers. Unlike traditional inverse analysis methods, this approach reconstructs the target parameters based on the spatial distribution characteristics of the soil layers, transforming discrete and independent parameters into a sequential structure (sequence). Using the reconstructed sequence as input and structural deformation as output, a surrogate model is established and combined with an optimization algorithm for inverse analysis of soil parameters. The proposed method was validated through numerical approaches, and the results demonstrate that, compared to conventional methods, the proposed method reduces the mean absolute error (MAE) of surrogate model predictions for structural deformations by up to 29.7% and decreases the MAE of numerical simulation results based on inversely analyzed parameters by up to 48.9%. Subsequently, the method was applied to an actual engineering project to verify the reliability of the soil parameters it provides. Finally, SHapley Additive exPlanations (SHAP) analysis was conducted to explore the interpretability of the machine learning model. The results indicate that the soil parameter not only contributes the most to the structural deformation but also interact strongly with other parameters. Furthermore, for the soil layers, the contribution of deeper soil layers to structural deformation is generally higher than that of the upper layers.