PSO-SVM Machine Learning for Blasting Vibration Velocity Prediction in Open Pit Mines

The peak particle velocity (PPV) of blasting vibration is a primary indicator to evaluate the explosion effect in an open pit mine. In the blasting scenario of an open pit mine, existing methods for predicting the peak velocity of blasting vibration are difficult to achieve ideal outcomes, leading to inappropriate designs of blasting parameters and detonation network. As a result, the peak velocity of blasting vibration cannot be accurately forecasted. Aiming at improving the prediction accuracy of blasting vibration peak velocity, based on the monitoring data of blasting vibrations at Yuanbaoshan open-pit mine with different coal storage conditions, distance from the explosive center, maximum charge per delay, elevation difference and longitudinal wave speed are chosen as the input parameters. The key parameters C and g of the support vector machine (SVM) algorithm are optimized through the global optimization of the particle swarm optimization (PSO) algorithm, so that the prediction performance of SVM is at the optimal state, and then the PSO-SVM model for predicting the peak velocity of blasting vibration is constructed. By analyzing the relationship among the input parameters and the blasting vibration peak velocity, it is concluded that the longitudinal wave speed, which represents the site conditions, is also a significant factor influencing the propagation of blasting vibration velocity. Comparing to the test results of the RS-dept and the improved Sadovsky formula, it is found that the maximum error of blasting vibration prediction by the PSO-SVM model is reduced by 48.19% and 53.6% compared with that by the RS-dept model and the improved Sadovsky formula, respectively. Also, the average error of blasting vibration prediction by the combined PSO-SVM algorithm is 3.37%, which is 29.94% and 17.86% lower than that by the improved Sadovsky formula and the RS-dept model, respectively. The predicted values of the improved PSO-SVM model match best with the measured values and the predicted results are most reliable. The proposed research method can provide theoretical guidance and practical reference for the blast design of an open pit mine.