Prediction and calibration of black soil modeling parameters based on response surface methodology and machine learning algorithms

Five machine learning algorithms Decision Tree, Random Forest, Support Vector Machine (SVM), KNN, and XG Boost were used to calibrate the discrete element contact parameters of the black soil by combining the measured data on the black soil and the simulated pile load test. Firstly, the physical parameters of the black soil and the angle of stacking were determined based on physical tests. Next, Plackett-Burman tests were carried out and the following important parameters were obtained: black soil-black soil static friction coefficient, black soil-black soil rolling friction coefficient, and black soil-stainless steel rolling friction coefficient. The simulation parameters that significantly influenced the black soil stacking angles were designed for the steepest climbing tests to optimize a range of values of the significant parameters. Machine learning was performed to determine the optimal model based on the results of the response surface index results. The results show that the decision tree model has better predictive ability and stability for the stacking angle compared to Random Forest, SVR, KNN, and XG Boost models. The best combination of parameters for the black soil-black soil static friction coefficient was 0.956, the black soil-black soil rolling friction coefficient was 0.499, and the black soil-stainless steel rolling friction coefficient was 0.221. The simulation parameters can provide a reference for optimizing the simulation parameters for the subsequent soil particles.