Estimating Air Over-pressure Resulting from Blasting in Quarries Based on a Novel Ensemble Model (GLMNETs-MLPNN)

In this study, a coupling of generalized linear modeling (GLMNET) and nonlinear neural network modeling with multilayer perceptrons (MLPNN), called GLMNETs-MLPNN modeling, was conducted for predicting air over-pressure (AOp) induced by blasting in open-pit mines. Accordingly, six GLMNET models were developed first. Then, their predictions were bootstrap aggregated as the new predictors, and an optimal MLPNN model was developed based on these new predictors. To prove the improvement of the proposed GLMNETs-MLPNN model, the conventional models, such as GLMNET, support vector machine, MLPNN, random forest, and empirical, were considered and developed based on the same dataset. The results of the proposed model then were compared with that of the conventional models in terms of accurate prediction and modeling. The findings revealed that the bootstrap aggregating of six generalized linear models (i.e., GLMNET models) by a nonlinear model (i.e., MLPNN) could enhance the accuracy in predicting AOp with a root-mean-squared error (RMSE) of 2.266, determination coefficient (R-2) of 0.916, and mean squared error (MAE) of 1.718. In contrast, the other stand-alone models provided poorer performances with RMSE of 2.981-4.686, R-2 of 0.597-0.860, and MAE of 3.156-1.990. Besides, the sensitivity analysis results indicated that burden, stemming, distance, spacing and maximum explosive charge per delay were the most important parameters in predicting AOp.

Automated summary

The study successfully developed a GLMNETs-MLPNN model for predicting air over-pressure induced by blasting in open-pit mines, and compared it with conventional models. The results demonstrated that the GLMNETs-MLPNN model outperformed other standalone models in accuracy, and conducted sensitivity analysis on the parameters affecting AOp prediction.