Prediction of gross calorific value as a function of proximate parameters for Jharia and Raniganj coal using machine learning based regression methods

Abstract

ABSTRACT Rising incidents of grade slippage of coal among coal suppliers and user companies are an increasing concern in Indian coal industries. This study aims to investigate the relation between Gross calorific value (GCV) and coal proximate parameters. Support vector, random forest, extreme gradient boosting based regression methods (SVR, RFR, and XGBoost, respectively) were used to establish the relationship between proximate data and GCV of coal. The predictive model was generated by introducing proximate variables (moisture, volatile matter, ash, fixed carbon) as independent variables and GCV as a dependent variable. The performance of these machine learning regression-based prediction models and some of the published empirical models is compared with measured GCV values. The result shows that the XGBoost regression model works best to estimate the GCV from proximate data with R2 of 99.87%, RMSD of 0.1280 MJ/kg, and average absolute error (MAPE) of 0.32%.