Parameter Calibration of Discrete Element Simulation for Unstructured Pavements Using Method‐Fused Screening and a Polynomial Kernel‐Based Support Vector Regression Model

ABSTRACT Accurate calibration of discrete element simulation parameters (characteristic parameters) for unstructured pavements is a prerequisite for model application, particularly in analyses such as tire‐pavement contact mechanics. Focusing on soil pavements, experiments such as the soil angle of repose were conducted, using the angle of repose as the response value. A fusion screening method for significant characteristic parameters of unstructured pavements was proposed using Plackett–Burman experiments, the Spearman correlation coefficient, and the gradient boosting decision tree algorithm (GBDT). The steepest ascent experiment determined the optimal range of the significant characteristic parameters, and a polynomial kernel‐based support vector regression (poly‐SVR) model was introduced to obtain their optimal values. Based on this, a discrete element model for the unstructured pavement was established, and its validity was demonstrated through compaction tests. Results showed an average relative error of 0.55% between simulated and measured soil repose angles, demonstrating the feasibility of the significant characteristic parameter fusion screening method and the poly‐SVR model for identifying optimal parameter values. Additionally, the mean relative error between the measured and simulated soil compaction values was determined to be 7.40%, confirming the validity of the established pavement model. The presented calibration method for discrete element simulation parameters of unstructured pavements can serve as a theoretical reference for related research.