Prediction of Ethane/Ethylene Adsorption Selectivity of MOF Materials Based on Machine Learning
Year: 2026
Type: conference-paper
Abstract: Ethane/ethylene separation is a high-energy step in the petrochemical industry, and the conventional cryogenic distillation route remains capital-intensive and carbon intensive. Metal-organic frameworks (MOFs) are promising adsorptive media for this task, yet the very large structural design space makes purely experimental or simulation-only screening inefficient. In this work, a large MOF data set was constructed and benchmark selectivity values were generated by molecular simulation. Multidimensional descriptors covering geometry, chemistry, and topology were then extracted to train and compare several machine-learning models. Among the tested algorithms, XGBoost achieved the best predictive accuracy, reaching a test-set coefficient of determination of 0.92. SHAP analysis identified pore-limiting diameter, metal coordination number, and confinement-related descriptor ratios as the dominant factors governing selectivity. The optimized model was finally used for rapid virtual screening, from which five high-performance MOF candidates were selected and independently revalidated by simulation. The proposed workflow substantially shortens the materials discovery cycle and provides a practical data-driven route for the rational design of low-carbon adsorbents for C₂ hydrocarbon separations.
Author: Zheng Zhao
Language: English
Subfield: Inorganic Chemistry
Field: Chemistry
Domain: Physical Sciences
SDG: Climate action
Open Access status: hybrid
OpenAlex ID: W7204190868