Prediction of Ground Vibration Induced by Rock Blasting Based on Optimized Support Vector Regression Models

Accurately estimating blasting vibration during rock blasting is the foundation of blasting vibration management. In this study, Tuna Swarm Optimization (TSO), Whale Optimization Algorithm (WOA), and Cuckoo Search (CS) were used to optimize two hyperparameters in support vector regression (SVR). Based on these methods, three hybrid models to predict peak particle velocity (PPV) for bench blasting were developed. Eighty-eight samples were collected to establish the PPV database, eight initial blasting parameters were chosen as input parameters for the prediction model, and the PPV was the output parameter. As predictive performance evaluation indicators, the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and a10-index were selected. The normalized mutual information value is then used to evaluate the impact of various input parameters on the PPV prediction outcomes. According to the research findings, TSO, WOA, and CS can all enhance the predictive performance of the SVR model. The TSO-SVR model provides the most accurate predictions. The performances of the optimized hybrid SVR models are superior to the unoptimized traditional prediction model. The maximum charge per delay impacts the PPV prediction value the most.