摘要
随着我国碳达峰、碳中和战略的实施,煤电、钢铁、水泥等产业CO_(2)捕集迫在眉睫.为了替代高能耗、易挥发的有机胺CO_(2)吸收剂,不挥发、高稳定、可设计的离子液体CO_(2)吸收剂受到广泛关注.相对于传统实验试错方法,人工智能机器学习成为实现快速预测离子液体CO_(2)捕集性能、选择最佳离子液体吸收剂的重要途径.本文在简要介绍机器学习算法和模型的基础上,从离子液体的描述符出发,综述了近年来基于人工智能和机器学习的定量结构-性能关系(quantitative structure-property relationship,QSPR)方法用于离子液体CO_(2)捕集性能研究的进展,重点介绍离子液体的分子结构描述符的种类及相应QSPR模型建立方法,对机器学习应用于离子液体CO_(2)捕集性能研究中存在的问题和进一步的研究工作提出了建议.本文有助于数据与化学化工交叉学科研究人员开发高效离子液体CO_(2)吸收剂.
With the implementation of China’s carbon peak and carbon neutrality strategy,CO_(2) capture in industries such as coal-fired power,steel,and cement is urgent.To develop alternative absorbents without high energy consumption and volatility of amine-based absorbents,nonvolatile,highly stable,and designable ionic liquid-based CO_(2) absorbents have received widespread attention.Machine learning(ML)is the core of artificial intelligence(AI)and a powerful tool that can search for patterns in high-dimensional data.In recent years,ML has been widely applied in fields such as energy and materials,chemistry and chemical engineering,and life science and medicine.By applying ML to the study of CO_(2) capture by ionic liquids(ILs),computers can simulate the linear or nonlinear relationship between the structure of ILs and CO_(2) capture performance,as well as predict the CO_(2) capture performance by ILs with specific structures under specific conditions.Compared to traditional trial and error methods,AI and ML technology have become an important way to quickly predict the CO_(2) capture performance of ILs and select the optimal IL-based absorbents.Based on a brief introduction to algorithms and models of ML including artificial neural network(ANN),support vector machine(SVM),tree-based models,multivariate linear regression(MLR),and gaussian processes regression(GPR),this article reviews the progress of AI and ML in the study of CO_(2) capture by ILs in recent years from a viewpoint of descriptors of ILs.This review focuses on the types of molecular structure descriptors of ILs and the establishment of corresponding quantitative structure-property relationship(QSPR)or quantitative structure-activity relationship(QSAR)models.Thereby,ML models can be divided into two categories based on the types of descriptors used to describe the CO_(2) dissolution performance of ILs.One is the ML models based on physical property descriptors,and another is the ML-based QSPR models based on molecular structure descriptors.The descriptors of former include molecular weight(Mw),acentric factor,(ω),critical temperature(Tc),critical pressure(Tp),critical compressibility factor(Zc),operating temperature(P),and operating pressure(T).The descriptors of latter are based on the concept of“structure determines properties”,and the structure of anions and cations in ILs has a significant impact on their CO_(2) capture properties.The descriptors of latter are generated based on atoms,ionic fragments contribution(IFC),group contribution(GC),fingerprint(FP),simplified molecular input line entry system(SMILES),molecular graph,quantum chemical descriptors.The development of ML models is gradually evolving from a single model to an integrated model.Generally,ML models have improved from“simple learning”to“deep learning”.MLbased QSPR models using molecular structural descriptors are more accurate and robust than ML models using physical property descriptors.Therefore,ML-based QSPR methods can be used to search and identify the optimal structures of ILs with the CO_(2) absorption,solubility,or Henry constant as the aim.Finally,the perspective for further researches on the design of ML models for predicting CO_(2) capture by ILs was also provided.It should be considered in the ML models that flue gas containing SO_(2),NOx,and moisture will affect the capture performance of ILs.Besides,functional groups in the functionalized ILs have a significant impact on capture performance,suggesting different ML models from models for physical solubility.In summary,the application of ML models in the prediction of CO_(2) capture by ILs will expand the application of AI technology in the fields of energy and environmental protection.
作者
张瑞娜
田媛
葛春亮
张威
卢晗锋
崔国凯
Ruina Zhang;Yuan Tian;Chunliang Ge;Wei Zhang;Hanfeng Lu;Guokai Cui(Institute of Industrial Catalysis,College of Chemical Engineering,State Key Laboratory Breeding Base of Green Chemistry Synthesis Technology,Zhejiang University of Technology,Hangzhou 310014,China;Beijing Key Laboratory of Ionic Liquids Clean Process,CAS Key Laboratory of Green Process and Engineering,State Key Laboratory of Multiphase Complex Systems,Institute of Process Engineering,Chinese Academy of Sciences,Beijing 100190,China;Zhejiang Zheneng Technology&Environment Group Co.,Ltd.,Hangzhou 310012,China)
出处
《科学通报》
北大核心
2025年第26期4462-4472,共11页
Chinese Science Bulletin
基金
浙江省重点研发项目(2024C03108,2023C03127)
国家自然科学基金(22378353,22078294)
浙江省自然科学基金(LTGS24E080008)
浙江浙能科技环保集团股份有限公司项目(ZNKJ-2024-077)资助。
关键词
人工智能
机器学习
离子液体
二氧化碳捕集
定量结构-性能关系
artificial intelligence
machine learning
ionic liquid
carbon dioxide capture
quantitative structure-property relationship
