Abstract: Blasting vibration will cause harm to surrounding residents and buildings (structures).However,blasting vi bration prediction technology has problems such as complex process and premature convergence.In order to improve the predic tion efficiency and accuracy,an improved GS-SVR prediction model is proposed.By changing the search step size of the grid search algorithm (improved GS),the support vector machine parameters are optimized,and the combination of the optimal ker nel function parameter gamma and the optimal penalty parameter c is applied to the support vector regression model (SVR). Based on the monitoring data of an open-pit mine,the random forest method was used to process the original data,and four pa rameters were selected as the input parameters of the model,including the distance between the explosion center,the total a mount of single blasting,the delay time of the hole distance meter and the delay time of the row distance meter.The model was used to predict the peak velocity and the main vibration frequency of the blasting vibration.The results show that the conver gence speed of the improved GS-SVR model is 10 s,and the accuracy can reach 99%.Compared with the prediction results of SVR,GA-SVR,PSO-SVR and Tabnet models,the training efficiency and prediction accuracy are significantly improved,indica ting that the improved GS-SVR model has better generalization ability.The prediction model of blasting vibration velocity and dominant frequency proposed in this study provides a effective prediction method for similar blasting projects.

Key words: blast vibration,improved grid search,SVR,blast velocity,vibration frequency