Enhancing blasting vibration prediction accuracy using hybrid ML models

Blasting is the main method of rock excavation in mining and civil engineering. However, ground vibration induced by blasting often causes cracks and even collapses of surrounding structures. Traditional empirical methods struggle to provide a reliable explosion prediction model to design blasting parameters due to the complexity of explosion physics and mining area topography. The PSO-GBDT, PSO-XGboost, and Random forest(RF) hybrid ensemble machine learning (ML) models are given and compared in this research. To prevent overfitting, K-cross validation is used in hyper-parameters tuning. A hybrid stacking ML model based on PSO-RF-XGboost and a data-augmented method is developed with the goal of increasing the accuracy of blasting prediction. To shed light on the viability and interpretability of the model, a variety of interpretability analyses—including sample required numbers analysis and dimension contribution analysis based on SHAP—are discussed.