Insight into Adsorptive Desulfurization by Zeolites: A Machine Learning Exploration
Liberty L. Mguni, Andrew Ndhlovu, Xinying Liu, Diane Hildebrandt, Yali Yao
Abstract
Adsorptive desulfurization (ADS) of hydrocarbon fuels using zeolite-based adsorbents holds great promise due to the mild conditions required to remove sulfur, thus addressing the energy and environmental concerns. However, screening of the ever-increasing number of potential ADS zeolites for adsorptive capacity is increasingly intractable. Furthermore, there is no consensus on the parameters with a dominating influence; hence, adsorbent synthesis design has remained an art. Machine learning (ML) has gained popularity as a powerful tool for understanding the catalytic mechanism and providing insights into catalytic design. In this study, we used multiple linear regression (MLR) and random forest (RF) regression to explore the process of ADS by zeolites using data from the literature. We found better predictive performance under the RF model (R2 = 0.93) than the MLR model (R2 = 0.88), which violated the assumption of linearity. The initial adsorbate concentration showed the highest relative importance of the variables, followed by zeolite properties (metal ion, mesoporous volume, pore size, Si/Al ratio, and surface area) for ADS activity. Our RF prediction model may be used in place of experimental ADS zeolite screening, cutting down on time and resource requirements. This work demonstrates the utility of ML and literature survey data as an inexpensive alternative to experimentation when doing research to obtain mechanistic insight into the complex process of ADS.