Sensitivity analysis and prediction of diesel particulate matter emissions in Indian underground metalliferous mines using regression and machine learning algorithms

Abstract

Understanding the accumulation and dispersion characteristics of diesel particulate matter (DPM) in the closed environment of the underground mines and identifying the factors affecting these concentrations enables the mining authorities to formulate effective mitigation techniques to provide a safe and healthy working environment for the miners. For achieving this, DPM concentrations of three Indian underground mines were studied with a focus on the prediction of DPM using regression and Machine Learning (ML) algorithms and comparing their prediction accuracy. Multivariate regression analysis (MVRA) and Artificial neural networks (ANN) were utilised to correlate DPM concentrations and site-specific factors including the distance from the exhaust, ventilating air velocity, and source emission concentration. The results indicate that the best correlations between DPM concentrations and site parameters could be established by ANN with an R2 value of 0.986, while MVRA exhibited an R2 value of 0.916. Additionally, DPM concentrations were found to be most sensitive to gallery air velocity with a sensitivity index of (-) 2.87. The critical wind velocity above which the DPM concentrations reached below the safe permissible limits was determined to be 1.65 m/s for given cross-sectional profiles.