High-Throughput and Machine Learning for Predicting the Separation Performance of Metal–Organic Frameworks for SF6/N2

In view of the greenhouse effect of SF 6 and its recycling benefits, the development of highly efficient SF 6 /N 2 separation is extremely urgent. In this study, high-throughput screening and machine learning methods were used to analyze the relationship between features and performance, and a machine learning model capable of predicting materials with high SF 6 /N 2 adsorption and separation performance was trained. By implementing a database cleaning strategy, unreasonable structural data were systematically removed, improving the prediction accuracy of the model. Feature construction was carried out to obtain characteristic information covering aspects of structure, chemistry, force field, and energy. The adsorption data of the SF 6 /N 2 mixed gas were calculated based on the RASPA2 software. The importance analysis shows that the Henry’s coefficient of SF 6 plays a decisive role in the adsorption performance score. And cross-database prediction was conducted on the Tobacco database. The coefficient of determination of the logarithmic selectivity reaches 0.969.