Prediction of rock blasting induced air overpressure using a self-adaptive weighted kernel ridge regression

Abstract

Blasting operations are widely recognized as the most frequently used rock breakage approach in the field of Civil and Mining Engineering. However, the induced air-overpressure (AOp) can result in structural damages to surrounding buildings and therefore needs to be well predicted and subsequently minimized. In this study, a novel self-adaptive weighted kernel ridge regression (Sa-WKRR) was proposed to predict blast induced AOp, which used a self-adaptive weighting strategy to improve the performance of traditional kernel ridge regression (KRR). The model was developed and validated on two blasting datasets. Subsequently, the performance of the proposed Sa-WKRR was compared with 9 other machine learning models, i.e., KRR, kernel random vector functional link (KRVFL), Sa-WKRVFL, transductive KRR (TKRR), ensemble deep RVFL (edRVFL), artificial neural network (ANN), random forest (RF), support vector machine (SVM) and multivariate adaptive regression splines (MARS). The optimal performance of these models were obtained using grid search method and compared by three evaluation indices, root mean squared error (RMSE), mean absolute percentage error (MAPE) and correlation coefficient (R). The results demonstrated that the proposed Sa-WKRR has the best performance in two datasets, with RMSE of 0.46/1.98, MAPE of 0.30% and 1.20%, and R of 0.9991/0.9235 in Case study 1 and RMSE of 1.03/3.22, MAPE of 0.76%/2.74%, and R of 0.9965/0.9373 in Case study 2. Findings revealed that the proposed Sa-WKRR emerged as the most powerful and stable technique in predicting blast induced AOp compared with other machine learning models.