Coal higher heating value prediction using constituents of proximate analysis: Gaussian process regression model

ABSTRACT

This study aims to develop a globally valid prediction model for coal higher heating value (HHV). For the first time, the Gaussian process regression (GPR) method is performed to build the prediction model. For this purpose, a large dataset (as received basis) composed of a wide range of coal ranks is gathered from different geographic locations throughout the world countries in the related literature. The predictor variables for the prediction model include proximate analysis constituents that are moisture, volatile matter, fixed carbon, and ash. Furthermore, multiple linear regression (MLR) method is employed to predict coal HHV. To evaluate the performances of the developed models, the results obtained from each model are compared with each other and the results of the models given in the related literature by prediction performance criteria. The results prove that the prediction capability of the GPR model is superior to the MLR model and the models reported in the literature. For the testing stage, the attained coefficient of determination (R²), mean absolute percentage error (MAPE), root mean square error (RMSE) are 0.9833, 2.5%, 0.7672, respectively. It can be concluded that the proposed GPR model is a powerful tool to achieve high precision coal HHV prediction.

KEYWORDS:

• Coal
• gross calorific value
• estimation
• proximate analysis
• machine learning