化工进展 ›› 2023, Vol. 42 ›› Issue (1): 148-158.DOI: 10.16085/j.issn.1000-6613.2022-0539

• 化工过程与装备 • 上一篇    下一篇

机器学习高效筛选用于CO2/N2选择性吸附分离的沸石材料

王璐(), 张磊(), 都健   

  1. 大连理工大学化工学院化工系统工程研究所,辽宁 大连 116024
  • 收稿日期:2022-04-01 修回日期:2022-05-12 出版日期:2023-01-25 发布日期:2023-02-20
  • 通讯作者: 张磊
  • 作者简介:王璐(1997—),女,硕士研究生,研究方向为化工系统工程。E-mail:877907576@qq.com
  • 基金资助:
    国家自然科学基金(22078041);中央高校基本科研业务费(DUT20JC41)

High-throughput screening of zeolite materials for CO2/N2 selective adsorption separation by machine learning

WANG Lu(), ZHANG Lei(), DU Jian   

  1. Institute of Chemical Process Systems Engineering, School of Chemical Engineering, Dalian University of Technology, Dalian 116024, Liaoning, China
  • Received:2022-04-01 Revised:2022-05-12 Online:2023-01-25 Published:2023-02-20
  • Contact: ZHANG Lei

摘要:

目前,针对气体吸附性能的测定及材料设计筛选,传统的实验法耗时耗力,因此分子力学方法中的巨正则蒙特卡洛(GCMC)方法已被广泛应用于该领域中,但日益增长的材料数目使得GCMC方法的计算量越来越高。为解决这一问题,本文提出了一种基于机器学习(ML)方法的吸附材料的筛选框架,包含ML模型的建立、理想化PSA工艺模型筛选材料及GCMC方法的验证三个阶段。首先,建立人工神经网络模型,提出了沸石材料的结构描述符“天然构造单元(NBU)”对特定条件下的气体吸附量进行预测。对于CO2和N2气体,分别构建了两个拓扑结构不同的多层前馈神经网络。其次,通过理想吸附溶液理论(IAST)将纯组分的吸附等温线转化为摩尔分数为0.14/0.86的CO2/N2二元混合物吸附等温线,并根据一系列吸附材料评估指标筛选出11种最佳沸石材料,并从中选出4种沸石(MON、ABW、NAB和VSV)计算其GCMC的吸附数据。结果表明,它们对N2的吸附能力远低于CO2,因此对两种气体的吸附选择性较高,能够很好地从二元混合物中分离CO2

关键词: 机器学习, 神经网络, 二氧化碳捕集, 沸石, 巨正则蒙特卡洛

Abstract:

At present, in the determination of gas adsorption performance and material design screening, the traditional experiments consume time and effort, so Grand Canonical Monte Carlo (GCMC) method in molecular mechanics has been widely used. However, the growing number of materials makes the GCMC method more and more computationally intensive, and a framework for screening adsorption materials based on machine learning (ML) method was proposed to solve this problem. The framework included three stages: the building of ML model, material selection of idealized PSA process model and validation using GCMC method. Firstly, artificial neural network models were established, and the structure descriptors "natural building unit (NBU)" of zeolite materials was proposed to predict the adsorption capacity under certain conditions. For CO2 and N2, two multi-layer feed-forward neural networks with different topological structures were built. Secondly, the ideal adsorbed solution theory (IAST) can predict the mixture adsorption isotherms of CO2/N2 (mole fractions is 0.14/0.86) from pure-component adsorption isotherms, and then 11 "best" zeolite materials were selected by some adsorbent evaluation metrics. Four zeolite materials (MON, ABW, NAB and VSV) were selected and calculate their adsorption data using GCMC method. The results proved that their adsorption capacity for N2 was much lower than that for CO2, so they had high adsorption selectivity for the two gases and can well separate CO2 from binary mixtures.

Key words: machine learning, neural networks, CO2 capture, zeolite, grand canonical Monte Carlo

中图分类号: 

京ICP备12046843号-2;京公网安备 11010102001994号
版权所有 © 《化工进展》编辑部
地址:北京市东城区青年湖南街13号 邮编:100011
电子信箱:hgjz@cip.com.cn
本系统由北京玛格泰克科技发展有限公司设计开发 技术支持:support@magtech.com.cn