Predicting coal elemental components from proximate analysis: Explicit versus implicit nonlinear models

Abstract

In this study, the new explicit and implicit models for prediction of coal elemental components using the information of proximate analysis are developed. The nonlinear-based prediction models have inherently more ability than the linear-based models in identifying the complex relations between most of the elemental and proximate analysis components. Therefore, Nonlinear Parametric Regression (NPR) is considered as an explicit method while Artificial Neural Network (ANN) and Random Forest (RF) are taken into account as an implicit method. For modeling aim, a large coal database, including 5277 coal samples and representing wide range coal ranks, is utilized to build the nine novel nonlinear prediction models. One thousand coal samples are used to test the prediction performances and generalization capabilities of the developed models. The results indicate that the implicit based ANN and RF models can provide better results than the explicit-based NPR models, while the ANN-based models have the best predictive performances that R-2 is 0.9915, 0.9151 and 0.9817 for carbon, hydrogen and oxygen contents, respectively. According to the related literature, the developed models can predict successfully the coal elemental components. Therefore, they can be considered as valuable prediction tools in the design, operation, and analysis of the coal-related processes.