Estimation and performance optimisation of blasting-induced ground vibration using LSO-RF, APO-RF, and PO-RF models

Blasting-induced ground vibration threatens the safety and optimisation of open-pit blasting. This study proposes hybrid machine learning models for peak particle velocity (PPV) prediction by integrating random forest (RF) and extreme gradient boosting (XGBoost) with three metaheuristic algorithms. The results show that light spectrum optimiser (LSO)-RF achieves the best accuracy. Shapley additive explanations (SHAP) and partial dependence plots (PDP) analyses identify distance from the blast face and stemming length as the dominant factors affecting PPV. A graphical interface was further developed for practical field application and blast design optimisation.