Predicting the blast-induced ground vibration with support vector regression optimized by five swarm algorithms

Blast-induced vibration is one of the primary adverse effects associated with blasting operations, typically evaluated by peak particle velocity (PPV). Accurate prediction of PPV is critical for ensuring blasting safety. This study addresses this issue by introducing a support vector regression (SVR) model optimized with five swarm intelligence (SI) algorithms. The input parameters of the dataset are selected, and 53 data sets are measured in a case study. The dataset is divided into training (70%) and testing (30%) subsets. Moreover, tenfold cross-validation is employed to assess the potential occurrence of overfitting in this study. The results indicate that, for the overall dataset, the GWO-SVR model demonstrates strong predictive performance evaluated by six statistical metrics of mean square error (MSE), mean absolute error (MAE), coefficient of determination (R²), root mean square error (RMSE), Adjusted-R² and Taylor skill score (TSS). In the testing dataset, the PSO-SVR model has the best predictive performance based on MAE, R² and Adjusted-R², showing significant differences from the other models. The HHO-SVR model achieves the lowest uncertainty in both the training and testing datasets, with U₉₅ values of 0.1631 and 0.3185, respectively. Furthermore, the partial dependence plot (PDP) analysis identifies the importance of charge length (L), powder factor (Pf), maximum charge per delay (W) and the distance from the measuring point to blasting face (D) in predicting PPV. An interactive graphical user interface (GUI) has been developed to facilitate the prediction of PPV during blasting. A more comprehensive selection of input parameters and prediction models, developed based on small datasets, holds significant practical importance for determining the safety warning zone for blasting in Dahuangshan open-pit mine.