Assessment of blast-induced peak particle velocity using random forest optimized by multi-strategy greylag goose algorithm

To overcome the challenges arising from strong nonlinearity, complex multi-factor coupling, and the limited interpretability of conventional methods for blast-induced peak particle velocity (PPV) estimation, a hybrid model integrating a multi-strategy improved greylag goose optimization (MSGGO) algorithm with random forest (RF) is proposed. The MSGGO algorithm is developed by incorporating multiple enhancement strategies into the standard greylag goose optimization (GGO) framework, which enhances its ability to perform global exploration while improving convergence behavior. MSGGO is subsequently used to optimize the RF hyperparameters to enhance PPV prediction performance. Comparative experiments on 12 benchmark functions demonstrate that MSGGO generally achieves favorable convergence accuracy and stability compared with GGO, SOA, GJO, WMA, and RBMO. Using 119 sets of field-measured data from mining operations, a series of RF models optimized by different algorithms are constructed. Further comparisons with conventional models, including RF, BPNN, and SVR, as well as the USBM and Ambraseys empirical formulas, indicate that the MSGGO-RF model yields the most favorable results on the test dataset (R² = 0.9641, RMSE = 0.3100, MAE = 0.2574), demonstrating high prediction accuracy. SHAP analysis reveals that DBM is the most influential variable governing PPV response, accounting for 59.5% of the model contribution, and it shows a strong inverse relationship with PPV. The remaining variables mainly exhibit nonlinear effects. Overall, the MSGGO-RF model achieves high predictive accuracy while preserving good model interpretability, offering a reliable basis for precise predictive modeling and safety management in blasting operations.