Intelligent prediction of mechanical properties of backfill materials using a PSO-optimized XGBoost model

Backfill mining technology has been established as a primary method for achieving green and low-carbon coal mining. However, the mechanical properties of backfill materials are influenced by multiple parameters, accompanied by issues such as prolonged production cycles and high costs. To achieve rapid matching between the mechanical characteristics of backfill materials and actual working conditions, an intelligent prediction model was developed in this study using the XGBoost algorithm to predict the mechanical performance of backfill materials. First, a multifactor dataset was constructed based on laboratory experiments, and the optimal solution was identified through hyperparameter optimization via Particle Swarm Optimization (PSO). Accordingly, a PSO-XGBoost prediction model was designed to investigate the properties of backfill materials under various mixing ratios. Second, based on experimental data of backfill material proportions, the input features were selected as gangue content, gangue particle size, fly ash, cement, and fiber, while the uniaxial compressive strength, elastic energy index, and impact energy index were set as outputs. Finally, the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) were employed as evaluation metrics to assess the predictive performance of the model. Factors influencing the mechanical properties of backfill materials were also analyzed. The results demonstrated that the PSO-XGBoost model achieved coefficients of determination of 0.96846, 0.97814, and 0.96403 for predicting uniaxial compressive strength, elastic energy index, and impact energy index, respectively, exhibiting superior predictive performance compared to other models and enabling accurate prediction of the mechanical properties of gangue-cemented backfill materials. The findings contribute to facilitating the advancement of intelligent and green mining practices.