A Quantitative Investigation on Pyrolysis Behaviors of Metal Ion-Exchanged Coal Macerals by Interpretable Machine Learning Algorithms

Abstract

Generalizing rules from a complex process like catalytic pyrolysis to guide its regulation is always a difficult but attractive task. The influences of ion-exchange of metal ions (Na+, K+, Ca2+, Mg2+, Co2+ and Ni2+) on the pyrolysis behavior of vitrinite and inertinite from Shendong coal were investigated by thermogravimetric analyzer-Fourier transform infrared spectrometer, fixed-bed reactor, gas chromatograph-mass spectrometer and X-ray diffractometer. The results indicate that ion-exchange always reduces the yield of char but increase the yield of gas and water. The alkali and alkaline earth metal cations reduce the tar yield while the transition metals increase it, especially the Co2+.  A group of machine learning models were successfully constructed basing on random forest, support vector machine and gaussian process regression algorithms, to quantify the relationships between pyrolysis behaviors with properties of metal and maceral. The leave-one-out cross validation showed considerable determination coefficients (R2>0.9) between predicted and experimental values for most responses (15 of 29). Genetic programming based symbolic regression was incorporated with black-box algorithm, and 23 symbolic regression expressions with high confidence were successfully found, which improve the models’ interpretability and open up a novel way for (multi-)optimization and mechanism study on small experimental scale.

Keywords: Coal pyrolysis, Maceral, Ion-exchange, Machine learning, Symbolic regression