Machine Learning-Assisted\nHigh-Throughput Screening\nof Metal–Organic Frameworks for CO<sub>2</sub> Separation from\nCO<sub>2</sub>‑Rich Natural Gas

Under the appeal of carbon peaking and carbon neutrality\ngoals,\nit is highly advisable to develop green chemical technologies. Based\non this, it is even more attractive to synthesize methanol with the\nH<sub>2</sub> generated from water electrolysis by offshore wind power\nand the CO<sub>2</sub> separated from offshore CO<sub>2</sub>-rich\nnatural gas. Therefore, the separation and adsorption of CO<sub>2</sub>-rich natural gas in this context is of great socioeconomic significance.\nHowever, the conventional high-throughput screening methods for metal–organic\nframeworks (MOFs) in separating natural gas components and CO<sub>2</sub> suffer from great challenges such as high model complexity\nand long computation time. To address the aforementioned problems,\na machine learning-assisted modeling and screening strategy is proposed\nherein for the rapid and efficient separation of CO<sub>2</sub> from\nthe actual natural gas of six components (N<sub>2</sub>, CO<sub>2</sub>, CH<sub>4</sub>, C<sub>2</sub>H<sub>6</sub>, C<sub>3</sub>H<sub>8</sub>, and H<sub>2</sub>S). First, structural analysis is used\nto eliminate the MOFs that cannot adsorb CO<sub>2</sub> from the Computation-Ready\nExperimental Metal–Organic Frameworks (CoRE-MOFs) database.\nSix structural and 17 chemical descriptors of the remaining MOFs were\ncalculated. Grand Canonical Monte Carlo (GCMC) simulations were applied\nto evaluate the separation performance metrics of the randomly selected\ntraining and testing MOF samples. By combining 23 descriptors and\nseparation performance metrics, a Random Forest (RF) regression model\nwas obtained with <i>R</i><sup>2</sup> exceeding 0.92 on\nthe test samples, which was employed to predict the separation performance\nof the remaining MOFs. As a result, 10 MOF candidates with the best\nCO<sub>2</sub> separation performance were obtained. Furthermore,\na structure–property relationship of MOFs with satisfactory\nregenerability was conducted. Three design strategies were proposed\nto guide the development of high-performance novel MOFs for CO<sub>2</sub> separation. This study offers a high-throughput screening\nframework for MOFs to facilitate the separation of CO<sub>2</sub> from\na CO<sub>2</sub>-rich natural gas.