机器学习预测筛选MOF气体分离膜的研究进展  

Research progress of machine learning in MOF gas separation membranes prediction and screening

作　　者：田凯[1] 李东洋 段锟[1] 王景 张亚涛[1] TIAN Kai;LI Dongyang;DUAN Kun;WANG Jing;ZHANGYatao(School of Chemical Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学化工学院,郑州450001

出　　处：《膜科学与技术》2023年第6期149-158,共10页Membrane Science and Technology

基　　金：国家自然科学基金青年项目(22108258);河南省自然科学基金优秀青年基金项目(22300420085);河南省高校科技创新人才支持计划(24HASTTT004)。

摘　　要：金属有机框架(MOFs)气体分离膜因其优异的分离性能在碳捕集、能源气体分离等领域得到广泛关注。尽管采用高通量分子模拟计算可以实现高性能MOF膜的筛选,但随着MOFs数量的激增,逐一计算MOF膜的气体分离性能需要消耗大量的计算资源,而基于机器学习方法可以快速进行MOF膜的性能预测与筛选,进而加速高性能MOF膜的设计制备流程.本综述系统介绍了机器学习预测筛选MOF膜的方法与流程,总结了当前的研究进展,分析了机器学习预测筛选MOF膜未来的方向与挑战.Metal-organic frameworks(MOFs)based gas separation membranes have attracted great attention in the field of carbon capture and hydrocarbon separation due to their excellent performance.Undoubtedly,high-throughput molecular simulation helps to screen high-performance MOF membranes,but calculating gas separation performance for numerous MOF membranes requires significant computational resources.Lately,Machine learning-based methods can rapidly predict and screen the performance of MOF membranes,thereby accelerating the membrane design process.In this paper,the methods and procedures of machine learning prediction and screening of MOF membranes are systematically presented in four aspects:data preparation,feature engineering,model training and selection,and model evaluation and interpretation.Additionally,the current research advances in machine learning screening of pure MOF membranes and MOF mixed matrix membranes are expressed in detail.Also,the challenges and future directions of machine learning screening of MOF membranes are analyzed.

关 键 词：MOF膜 机器学习 气体分离 性能预测与筛选 分子模拟 

分 类 号：O604[理学—化学] TQ028.8[化学工程]