Intelligent calibration method for microscopic parameters in the discrete element method based on ensemble learning.

The Block Discrete Element Method is widely used in engineering research because it can accurately model fractured rock masses. However, the accuracy of simulations depends on selecting appropriate microscopic parameters, which cannot be directly obtained from macroscopic rock tests. Therefore, calibrating microscopic parameters is essential to ensure that the model's macroscopic physical and mechanical states align with laboratory test results. Traditional trial-and-error calibration methods are highly inefficient and computationally demanding. To address this challenge, this study randomly generated microscopic parameters for discrete block elements and established computational models for uniaxial compression, Brazilian splitting, and triaxial compression tests. Maximum-edge length of the Voronoi Trigons and failure modes were analyzed to verify model reliability. Based on the results, a macroscopic-microscopic parameter dataset was constructed, and correlation analysis was performed to determine the relationship between microscopic parameters and macroscopic behavior. Subsequently, a Stacking ensemble learning model was developed, trained, and tested. The optimal Stacking ensemble model outperformed others across multiple tasks, achieving higher R