Environment
Zaenal Zaenal; Noor Fauzi Isniarno; Delina Mutiara; Sofie Nur’aini; Hasyim Fadhilah; Elfida Moralista; Andrieanto Nurrochman
Abstract
Blasting is a fundamental open-pit mining operation necessary for rock breakage, but it also generates significant environmental noise pollution. Excessive noise from blasting not only endangers health but also poses problems to compliance with regulations, particularly in regions where acoustic standards ...
Read More
Blasting is a fundamental open-pit mining operation necessary for rock breakage, but it also generates significant environmental noise pollution. Excessive noise from blasting not only endangers health but also poses problems to compliance with regulations, particularly in regions where acoustic standards differ, such as Indonesia's use of both dBL and dBA standards. This research addresses the need for reliable and context-dependent predictive models for blasting noise, aiming to compare analytical and empirical formulas with machine learning techniques in dBA prediction. Measurements were conducted at 30 blasts at an open-pit coal mine in Indonesia, South Sumatra, using homogeneous acoustic sensors. The measured data points for frequency, dBL, and dBA were matched to calculated data using equations. Random Forest (RF) and Artificial Neural Network (ANN) predictive models using measured frequency and dBL as predictive variables were also derived. Results show that used Finn-derived equation has poor predictive accuracy, with errors exceeding 80%. Among the analytical and empirical models, Equation 3 performed the best, with an average error of 9%, while a site-spesific regression model based on measurements had an improved error rate of 5%. Machine learning models outperformed all models, with the RF model exhibiting an average error of 2% and demonstrating higher stability and consistency. The ANN model also did well, but with more variation and some overestimations.
