A Lasso and Elastic-Net Regularized Generalized Linear Model for Predicting Blast-Induced Air Over-pressure in Open-Pit Mines

Abstract

Air overpressure (AOp) is one of the products of blasting
operations in open-pit mines which have a great impact on the environment
and public health. It can be dangerous for the lungs, brain, hearing and the
other human senses. In addition, the impact on the surrounding
environment such as the vibration of buildings, break the glass door
systems are also dangerous agents caused by AOp. Therefore, it should be
properly controlled and forecasted to minimize the impacts on the
environment and public health. In this paper, a Lasso and Elastic-Net
Regularized Generalized Linear Model (GLMNET) was developed for
predicting blast-induced AOp. The United States Bureau of Mines
(USBM) empirical technique was also applied to estimate blast-induced
AOp and compare with the developed GLMNET model. Nui Beo open-pit
coal mine, Vietnam was selected as a case study. The performance indices
are used to evaluate the performance of the models, including Root Mean
Square Error (RMSE), Determination Coefficient (R2), and Mean Absolute
Error (MAE). For this aim, 108 blasting events were investigated with the
Maximum of explosive charge capacity, monitoring distance, powder
factor, burden, and the length of stemming were considered as input
variables for predicting AOp. As a result, a robust GLMNET model was
found for predicting blast-induced AOp with an RMSE of 1.663, R2 of
0.975, and MAE of 1.413 on testing datasets. Whereas, the USBM
empirical method only reached an RMSE of 2.982, R2 of 0.838, and MAE
of 2.162 on testing datasets.