Peak particle velocity (PPV) serves as a critical metric in assessing the appropriateness of blasting design parameters. However, existing methods for accurately measuring PPV remain insufficient. To develop a robust PPV prediction model, this study integrates the Extreme Gradient Boosting (XGBoost) algorithm with four distinct optimization techniques: Runge Kutta Optimizer (RUN), Equilibrium Optimizer (EO), Gradient-Based Optimizer (GBO), and Reptile Search Algorithm (RSA). Real-time blasting data from open-pit mines are employed to predict PPV, utilizing parameters including Charge Quantity per Hole (CQH/kg), Total Charge Quantity (TCQ/kg), Distance from Bursting Point to Measuring Point (DBM/m), Drilling Depth (DP/m), Borehole Diameter (BD/mm), Spacing (S/m), Row Spacing (RS/m), Minimum Burden (MB/m), and Depth Displacement (DD/m). The predictive outcomes of the XGBoost model, optimized by various algorithms, are benchmarked against the Sadovsky empirical formula, the conventional XGBoost model, and several traditional machine learning models (Ridge, LASSO, SVM, SVR) using performance metrics including R2, RMSE, VAF, MAE, and MBE. Additionally, the Shapley Additive Explanations (SHAP) method is employed to assess the impact of various factors on PPV prediction outcomes. The findings reveal that the GBO-optimized XGBoost model surpasses the RUN, EO, and RSA-optimized XGBoost models, along with other machine learning models and traditional empirical formulas, in predicting PPV. This study further corroborates that the XGBoost model, when enhanced with various optimization algorithms, effectively manages the non-linear characteristics of multiple factors, resulting in a reliable, straightforward, and efficient PPV prediction model. Moreover, the SHAP sensitivity analysis identifies DBM, TCQ, and CQH as the primary factors influencing PPV, enabling engineers to mitigate the impact on nearby structures, equipment, and personnel through the careful adjustment of explosive quantities.
Read full abstract