Prediction of Hardgrove Grindability Index of Afsin-Elbistan (Turkey) Low-grade Coals Based on Proximate Analysis and Ash Chemical Composition by Neural Networks

Abstract

ABSTRACT The aim of the study is to provide insights about the Hardgrove grindability index (HGI) value of Piocene aged Afsin-Elbistan low-grade coal determination without human or experimental errors, timely and low cost. For this reason, the HGI for Afsin-Elbistan low-grade coal based on twelve coal parameters was predicted through linear regression (LR), feed forward neural network (ANN), generalized regression neural network (GRNN) and elman network (EN) approaches in this study.. The obtained results show that LR and EN approaches were unsatisfactory because of high differences determined between actual and predicted HGI. However, ANN and GRNN approaches make quite reliable predictions on HGI with a high accuracy. The R-value of the models was 0.93 for ANN and 0.98 for GRNN approaches. The percentage of predicted HGI within ±3 deviation through ANN and GRNN approaches was found to be 89.53 and 94.19, respectively. The sensitivity of approaches to predictors was evaluated by an enterremove selection method to determine the individual effect of each parameter on the prediction of HGI. The best prediction approach for the Afsin-Elbistan Pliocene aged coal was GRNN that can be applied to predict HGI for the similar aged coals.