Predictive modeling of blast induced ground vibration using RF-PSO-SVM

In the mining industry, blasting operation is an effective method for rock fragmentation. Blast-induced is one of the most harmful undesirable outcomes produced by blasting operations. When rising to a certain level, it can cause discomfort for humans living in surrounding communities, and damage to civil constructions. Therefore, the prediction and control of ground vibration gain attention in the literature. Peak particle velocity (PPV) is widely used as the indicator of the magnitude of ground vibration. Thus, the main objective of this paper is to establish an accurate and universal tool for PPV prediction. This study implemented a novel non-linear model based on a support vector machine (SVM), particle swarm optimization (PSO) and random forest (RF) (the proposed model is so called RF-PSO-SVM). Multivariate linear regression (MLR) and three popular empirical models are implemented and compared with the proposed model. The dataset utilized in this paper was collected in a mine located in Istanbul. The results indicate that PSO-SVM has a competitive accuracy with the highest 𝑅2 (R-squared value) of 0.7716, least MSE (Mean squared error) of 13.5979 and least MAE (Mean absolute error) of 2.7550. In terms of stability analysis, 10-fold cross-validation was applied. The results reveal that RF-PSO-SVM had the optimal ability of generalization as it had the steadiest prediction variance over different validation groups. The proposed non-linear model presents the feasibility and superiority of blast-induced ground vibration prediction and control.