Indirect Gross Calorific Value prediction using Random Forest

During operation of coal-based power plants, frequent calorific Value measurement is necessary. Previously, Artificial Intelligence based models have been developed for instant calorific value calculation based on proximate analyses or ultimate analyses or combination of both. In this paper, random forest was used for comparison of all the three methods and computing relative analyses parameters importance. This study uses well known USGS coal qual dataset. In this work, 10-fold validation strategy and R-squared was used as validation strategy and performance metric respectively. Ultimate analyses (R-squared = 0.9984) performed slightly better than proximate analyses (R-squared = 0.9861) or combination of both (R-squared = 0.9982). Lastly, carbon was found to be the most important feature in all models.