Random forest model for determination of the lower heating value of TPP “Kolubara A” coal

Abstract

Heating value is an important indicator for assessment of the coal quality. Machine learning models are powerful computational tools that allow for the analysis of various heat and mass transfer phenomena in energy systems. In this paper, Random forest model for determining the lower heating values of coal from the thermal power plant “Kolubara A” is developed based on proximate and ultimate fuel analysis. A database of the proximate and ultimate fuel analysis values and lower heating value of coal was created by experimental measurements in the accredited test laboratory of the Department of Thermal Engineering and Energy (“VINČA” Institute of Nuclear Sciences). The developed Random forest models, applied to a relatively small database, showed acceptable predictions for the lower heating value based on both the proximate analysis (RMSE = 0.22 MJ/kg and MAPE = 2.26%) and the ultimate analysis (RMSE = 0.64 MJ/kg and MAPE = 6.12%), with better accuracy achieved by the model whose input data consisted of the values of technical fuel analysis.