Simultaneous prediction of coal rank parameters based on ultimate analysis using regression and artificial neural network

Results from ultimate analysis, proximate and petrographic analyses of a wide range of Kentucky coal samples were used to predict coal rank parameters (vitrinite maximum reflectance (R_max) and gross calorific value (GCV)) using multivariable regression and artificial neural network (ANN) methods. Volatile matter, carbon, total sulfur, hydrogen and oxygen were used to predict both R_max and GCV by regression and ANN. Multivariable regression equations to predict R_max and GCV showed R²=0.77 and 0.69, respectively. Results from the ANN method with a 2-5-4-2 arrangement that simultaneously predicts GCV and R_max showed R² values of 0.84 and 0.90, respectively, for an independent test data set. The artificial neural network method can be appropriately used to predict R_max and GCV when regression results do not have high accuracy.