A Framework for Parameter Calibration in Discontinuous Deformation Analysis Based on Interpretable Surrogate Models

Discontinuous deformation analysis (DDA) is a numerical method that is extensively utilized for simulating discrete blocks. Nevertheless, its implicit calculation approach brings in a multitude of control parameters that lack physical significance, and proper handling of these parameters is essential for obtaining accurate results. To tackle this issue, this study proposes a surrogate model‐driven parameter calibration framework that incorporates interpretability analysis. First, a Kriging surrogate model is constructed to establish an efficient substitute for DDA computations, thus accelerating forward calculations. Subsequently, the SHapley Additive exPlanations (SHAP) method is introduced to quantify global parameter sensitivity. Finally, an intelligent optimization algorithm is integrated to develop a parameter inversion mechanism, thereby forming a complete calibration system of “surrogate modeling–sensitivity analysis–parameter optimization.” Numerical examples demonstrate that this framework can effectively identify the optimal combination of key control parameters. The average errors are 1.39% in the two‐slider model and 1.63% in the elastic foundation model. This approach offers an automated parameter calibration process that doesn't require manual intervention, providing a reliable theoretical tool for DDA engineering applications in tunneling, slope stability, and rock engineering.