Influence Of Enhanced Production and Pit Dimensions of Opencast Coal Mines on Dust Concentration

Abstract

Opencast coal mines play a crucial role in meeting the energy demands of a country. With significant availability of indigenous coal reserves and its affordability, coal is likely to continue as primary source of energy to meet the developmental needs of rising economy. However, the operations will result in deterioration of ambient air quality, particularly due to emission of particulate matter (PM), PM10 and PM2.5. PM10 consists of PM size less than 10 microns and PM2.5 consists of PM size less than 2.5 microns. The PM disperses to different directions with varying concentrations based on the quantum of mining operations and local meteorological parameters. Monitoring of PM dispersion is essential due to its associated health impacts. These particles can penetrate deep into lungs, potentially triggering respiratory and cardiovascular illnesses. Prediction of ambient dispersion of PM helps in taking mitigation measures for reducing the dispersion and thus exposure. Development of region-specific dispersion models with respect to changing mine parameters and local meteorological conditions for PM10 and PM2.5 emanating from opencast coal mines are necessary for accurate prediction of PM dispersion. Though number of techniques exists for prediction of dust concentration due to opencast mining, machine learning offers several advantages over traditional modeling techniques in terms of data driven insights, non-linearity, flexibility, handling of complex interactions, anomaly detection, etc. An approach is made in this research to analyze the concentration of PM in and around 6 opencast coal mines of the Singareni Collieries Company Limited located in Godavari Valley Coal Field, Telangana State of India and to predict the PM concentrations in core and buffer zones using machine learning techniques. The study involved collection of 10 years historical data comprising of 31,680 observations related to mine operating parameters, meteorological and PM data for processing and subsequent dust predictions. Data has been analyzed using multivariate regression and different machine learning techniques like Decision Tree, Bagging and Random Forest (RF). i The performance metrics of test data compared in order to find the best fit model among these techniques. It has been found that RF method has given better accuracy. A software program PMPOM (Particulate Matter Prediction in Opencast Mines), has been developed using Python with RF algorithm to predict PM10 and PM2.5 values at select locations in each mine. Further, contour plots have been developed using Quantum Geographic Information System (QGIS), a geospatial modelling application, for visualization and spatial analysis of PM dispersion patterns within and surrounding the mine. This comprehensive approach has been instrumental in creating more accurate and robust dust dispersion model for the entire region. The study also aimed at determining the ratio of PM2.5/PM10 which is helpful in arriving at the concentration of PM2.5 from PM10 or vice-versa in case of non measurement of any one parameter. It has been concluded that a relationship is not possible to establish between PM2.5 and PM10 in the core zone, whereas in the buffer zone is feasible.