Predicting the Concentrations of Rare Earth Elements and Yttrium in Coal Using Self-Organizing Map

Abstract

Several coals and coal by-products around the world have been identified as important alternative sources for rare earth elements and yttrium (REY) recovery, as these are considered crucial. However, many pre-existing coal chemical data and coal samples do not contain REY data, and in many cases, it is not possible to re-determine the REY concentrations in these samples. In this investigation, 528 coal samples collected from 36 coal mines of China were used to train a self-organizing map (SOM) model and the trained model was subsequently used to predict the REY concentrations in coal. The results were compared with the results of three other existing machine leaning methods, and the SOM model exhibited the highest accuracy in predicting REY concentrations. The trained SOM model was successfully used to predict REY concentrations in coal from the Fuqiang Mine, Hunchun Coalfield, northeastern China. The results were mostly consistent with those determined by an analytical technique. This work not only allows geologists to predict large-scale analysis of REY potential in coals but also improves our understanding to predict geochemical data using machine learning methods.