-
二氧化碳(CO2)等温室气体的过量排放对生态环境造成严重危害,中国为此提出2060年前实现“碳中和”的目标[1],美国也有类似的计划,包括到2030年将温室气体污染从2005年的水平减少50%,以及2050年前实现零排放经济[2]. 尽管各地企业极力节能减排,但仍处于起步阶段,减少碳排放进展缓慢,高效处理烟气中的CO2为实现这一目标提供了一种新思路.
烟气中CO2的捕获与封存(carbon capture and storage,CCS)已经成为减少碳排放的主要手段[3],其中物理吸收法、化学吸收法、薄膜法等较为常用. 化学溶剂吸收法使用氨水、醇胺等来分离回收CO2,是目前工艺最成熟、应用最广泛的技术,但它也存在许多缺点,如腐蚀问题、溶剂易损失和成本较高等[4]. 近10年来,高性能固体吸附剂因其高效率、低能耗和适用于不同温度和压力范围等优点被广泛关注. 固体吸附剂包括多孔碳材料、沸石、金属-有机框架等,其中多孔碳材料(porous carbon materials,PCM)因其丰富可调的多孔结构、性质稳定、易于制备、可再生、价格低廉,在CO2吸附方面具有独特优势[5].
PCM的比表面积(SBET)、微孔容(Vmicro)、中孔容(Vmeso)等结构特性和C、H、O、N等元素组成是影响CO2吸附的重要因素。微孔被证明是PCM吸附CO2的主要影响因素[6],含氮官能团可以作为固定酸性CO2分子的Lewis碱性位点,增强CO2吸附能力. PCM的制备经过碳化和活化过程,不同碳化温度、活化剂(例如KOH,ZnCl2)、活化温度得到的PCM差异很大。为提高吸附性能,传统方法是改变活化温度、活化剂(类型和比例)制备PCM,根据吸附等温线判断其对CO2的吸附性能,选择最佳的制备条件. 这种方法周期长、工作量大[7],无法深入了解PCM的结构性质、元素组成对CO2吸附的影响[8].
为解决这个问题,越来越多的研究使用机器学习(machine learning,ML)进行高性能PCM的筛选,以计算为指导进行材料设计,减少了材料开发的时间和成本[9]. Yuan等利用3种基于决策树的ML模型,以结构特性、元素组成、吸附参数为输入特征,对常规多孔碳和杂原子掺杂多孔碳的CO2吸附量进行预测,其中梯度提升树(gradient boosting decision tree,GBDT)模型具有最佳的预测性能[10];Wang等利用卷积神经网络,直接将氮气(N2)吸附等温线作为输入来预测PCM的N2/CO2吸附选择性[11];Zhu等开发了一个随机森林模型来预测PCM对CO2的吸附能力,同时利用特征重要性量化了结构特性和吸附条件在吸附过程中的作用[12]. 机器学习作为一种数据挖掘技术,利用数据派生模型从给定数据集中提取复杂且隐蔽的相关性[13],通过机器学习算法建立CO2吸附量与材料性质和吸附条件(温度、压力)之间的模型,可以量化各变量与吸附性能之间的非线性关系,以及对具有不同结构特性PCM的CO2吸附量做出预测.
基于机器学习的多孔碳材料吸附CO2的关键因素
Study on the key factors of CO2 adsorption by porous carbon materials based on machine learning
-
摘要: 多孔碳材料具有性质稳定、高孔隙率和高比表面积等优点,被广泛应用于气体分离和存储. 目前,针对多孔碳材料吸附CO2的优化主要通过“炒菜”的模式,但对于各种结构特性、元素组成对吸附的作用尚不明确. 机器学习算法通过构建特征与标签之间的模型从而量化其中的非线性关系,被逐渐运用于高性能多孔碳材料的筛选. 为了探究多孔碳材料吸附CO2的关键因素,建立了一个包含65个碳样本、
1591 个数据点的数据集,以7∶3的比例随机拆分为训练集和测试集,采用线性回归、反向传播神经网络、支持向量机、随机森林等5种不同机器学习模型训练和测试. 结果表明:随机森林模型在测试集上具有最佳的泛化能力(R2=0.954、RMSE=0.286). 压力是多孔碳材料吸附CO2最重要的影响因素,但随着压力和温度的提高,结构特性和元素组成逐渐成为主导因素. 结构特性方面,比表面积和微孔是结构特性中影响CO2吸附能力的关键因素;常压下CO2吸附主要受微孔控制,高压下中孔对吸附也有重要作用. 元素组成方面,氮元素是最重要的因素,其特征重要性在273 K、0.5—1 bar时可以达到15.27%.Abstract: Porous carbon materials are widely used for gas separation and storage because of their stable properties, high porosity and high specific surface area. Currently, the optimization of CO2 adsorption on porous carbon materials is mainly through the "frying" mode, but the role of various structural properties and elemental composition on adsorption is not clear. Machine learning algorithms, which quantify the nonlinear relationships by constructing models between features and labels, are gradually being applied to the screening of high-performance porous carbon materials. In order to find the key factors of CO2 adsorption by porous carbon materials, a dataset containing 65 carbon samples with1591 data points is established and randomly split into training and testing sets in the ratio of 7:3. Five different machine learning models, including linear regression, backward propagation neural network, support vector machine, and random forest, are used for training and testing. The results show that the random forest model has the best generalization ability on the test set (R2=0.954, RMSE=0.286). Pressure is the most important influencing factor for CO2 adsorption by porous carbon materials, but with increasing pressure and temperature, structural properties and elemental composition gradually become the dominant factors. As for the structural properties, specific surface area and micropores are the key factors affecting CO2 adsorption capacity in structural properties; CO2 adsorption at atmospheric pressure is mainly controlled by micropores, mesopores also play an important role in adsorption under high pressure. As for elemental composition, nitrogen is the most important factor, and its characteristic importance can reach 15.27% at 273 K and 0.5—1 bar.-
Key words:
- porous carbon materials /
- pore structure /
- carbon dioxide adsorption /
- machine learning.
-
表 1 各模型的参数和参数空间
Table 1. Parameters and parameter space of each model
模型
Models参数
Parameters搜索空间
Search spaceSVR(L) gamma [0.001, 0.005, 0.01, 0.02, 0.05, 0.1, 0.2, 0.3,0.4,0.5] C [0.1, 0.2, 0.25, 0.5, 1, 1.5, 2, 5, 10, 15, 16] SVR(RBF) gamma [0.001, 0.005, 0.01, 0.02, 0.05, 0.1, 0.2, 0.3,0.4,0.5] C [0.1, 0.2, 0.25, 0.5, 1, 1.5, 2, 5, 10, 15, 16] BP max_iter [10, 30, 50, 100, 200] learning rate ['constant', 'invscaling', 'adaptive'] solver ['lbfgs', 'sgd', 'adam'] RF n_estimators Range between 200 and 2000 with increments of 10 maximum tree depth [2, 3, 4, 5, 6, 7, 8, 9, 10] min_samples_leaf [1, 2, 4] 表 2 5种不同模型的拟合性能
Table 2. Fitting performance of the five different models
模型
Model均方误差
MSE均方根误差
RMSE平均绝对误差
MAE决定系数
R2LR 0.402 0.634 0.471 0.775 SVR(L) 0.414 0.643 0.468 0.768 SVR(RBF) 0.089 0.299 0.192 0.95 BP 0.112 0.335 0.252 0.937 RF 0.082 0.286 0.215 0.954 表 3 不同吸附条件下结构特性和元素组成的特征重要性
Table 3. Characteristic importance of structural properties and elemental composition under different adsorption conditions
273 K 298 K 0—0.5 bar 0.5—1 bar 0—0.5 bar 0.5—1 bar SBET 2.95% 18.27% 3.07% 10.69% Vtotal 1.95% 7.31% 3.27% 7.12% Vmicro 2.18% 9.61% 2.93% 6.04% Vmeso 6.22% 5.89% 4.64% 7.92% C 3.47% 6.31% 10.26% 10.1% H 3.2% 5.44% 5.55% 12.85% O 2.63% 8.28% 5.02% 12.22% N 6.87% 15.27% 9.07% 11.36% -
[1] CAO L, LEI S, GUAN Y X, et al. CCUS industry under target of carbon-peak and carbon-neutrality: Progress and challenges[J]. Frontiers in Energy Research, 2022, 10: 860665. doi: 10.3389/fenrg.2022.860665 [2] CHAPMAN A, ERTEKIN E, KUBOTA M, et al. Achieving a carbon neutral future through advanced functional materials and technologies[J]. Bulletin of the Chemical Society of Japan, 2022, 95(1): 73-103. doi: 10.1246/bcsj.20210323 [3] RAHMAN F A, AZIZ M M A, SAIDUR R, et al. Pollution to solution: Capture and sequestration of carbon dioxide (CO2) and its utilization as a renewable energy source for a sustainable future[J]. Renewable and Sustainable Energy Reviews, 2017, 71: 112-126. doi: 10.1016/j.rser.2017.01.011 [4] 王燕. 烟气中二氧化碳的化学处理技术及应用研究[J]. 科技信息, 2011(4): 114. WANG Y. Chemical treatment technology and application research of carbon dioxide in flue gas[J]. Science & Technology Information, 2011(4): 114 (in Chinese).
[5] CREAMER A E, GAO B. Carbon-based adsorbents for postcombustion CO2 capture: A critical review[J]. Environmental Science & Technology, 2016, 50(14): 7276-7289. [6] ZHANG Z H, SCHOTT J A, LIU M M, et al. Prediction of carbon dioxide adsorption via deep learning[J]. Angewandte Chemie International Edition, 2019, 58(1): 259-263. doi: 10.1002/anie.201812363 [7] 杨榛, 顾幸生, 梁晓怿, 等. 基于PCA-GA神经网络模式识别的炭纤维复合材料导电综合性能优化及预测的研究[J]. 计算机与应用化学, 2008, 25(12): 1543-1548. YANG Z, GU X S, LIANG X Y, et al. Optimization and prediction research on carbon fiber composite integrated conductive performance based on PCA-GANN pattern recognition[J]. Computers and Applied Chemistry, 2008, 25(12): 1543-1548(in Chinese).
[8] MAULANA KUSDHANY M I, LYTH S M. New insights into hydrogen uptake on porous carbon materials via explainable machine learning[J]. Carbon, 2021, 179: 190-201. doi: 10.1016/j.carbon.2021.04.036 [9] CAI J Z, CHU X, XU K, et al. Machine learning-driven new material discovery[J]. Nanoscale Advances, 2020, 2(8): 3115-3130. doi: 10.1039/D0NA00388C [10] YUAN X Z, SUVARNA M, LOW S, et al. Applied machine learning for prediction of CO2 adsorption on biomass waste-derived porous carbons[J]. Environmental Science & Technology, 2021, 55(17): 11925-11936. [11] WANG S, LI Y, DAI S, et al. Prediction by convolutional neural networks of CO2/N2 selectivity in porous carbons from N2 adsorption isotherm at 77 K[J]. Angewandte Chemie International Edition, 2020, 59(44): 19645-19648. doi: 10.1002/anie.202005931 [12] ZHU X Z, TSANG D C W, WANG L, et al. Machine learning exploration of the critical factors for CO2 adsorption capacity on porous carbon materials at different pressures[J]. Journal of Cleaner Production, 2020, 273: 122915. doi: 10.1016/j.jclepro.2020.122915 [13] HAGHIGHATLARI M, HACHMANN J. Advances of machine learning in molecular modeling and simulation[J]. Current Opinion in Chemical Engineering, 2019, 23: 51-57. doi: 10.1016/j.coche.2019.02.009 [14] 刘楠. 煤基多孔碳的制备、改性及其对CO2/CH4吸附分离研究[D]. 焦作: 河南理工大学, 2020. LIU N. Study on preparation and modification of coal-based porous carbon and its adsorption and separation for CO2/CH4[D]. Jiaozuo: Henan Polytechnic University, 2020(in Chinese).
[15] 张涛, 王彬彬, 李瑶. 基于玉米秸秆的氮掺杂多孔碳制备及其对CO2吸附和CO2/N2分离性能研究[J]. 河南理工大学学报(自然科学版), 2022, 41(6): 174-180. ZHANG T, WANG B B, LI Y. Preparation ofstalk-based nitrogen-doped porouscarbon and itsCO2 adsorption and CO2/N2 separation properties[J]. Journal of Henan Polytechnic University (Natural Science), 2022, 41(6): 174-180(in Chinese).
[16] 王娅鸽, 王彬彬, 杨德威, 等. 氮掺杂柔性块体多孔碳的制备及其对CO2/CH4吸附分离研究[J]. 材料导报, 2023, 37(22): 22050326. WANG Y G, WANG B B, YANG D W, et al. Preparation of nitrogen-doped flexible bulk porous carbon and its study on CO2/CH4 adsorption and separation[J]. Materials Reports, 2023, 37(22): 22050326(in Chinese).
[17] 熊龙. 氮掺杂多孔碳纤维的制备及其二氧化碳吸附性能研究[D]. 昆明: 昆明理工大学, 2020. XIONG L. Preparation of nitrogen doped porous carbon fiber and its carbon dioxide adsorption performance[D]. Kunming: Kunming University of Science and Technology, 2020 (in Chinese).
[18] 刘康恺, 韩云龙, 孟龙月, 等. Zeolites模板法制备氮掺杂多孔碳材料及其CO2吸附性能[J]. 实验室研究与探索, 2018, 37(9): 29-32. LIU K K, HAN Y L, MENG L Y, et al. Preparation and performance of CO2 adsorption of N-doped porous carbons by zeolites-template method[J]. Research and Exploration in Laboratory, 2018, 37(9): 29-32(in Chinese).
[19] SONG C C, LIU M H, YE W Y, et al. Nitrogen-containing porous carbon for highly selective and efficient CO2 capture[J]. Energy & Fuels, 2019, 33(12): 12601-12609. [20] QIN F F, GUO Z Y, WANG J S, et al. Nitrogen-doped asphaltene-based porous carbon nanosheet for carbon dioxide capture[J]. Applied Surface Science, 2019, 491: 607-615. doi: 10.1016/j.apsusc.2019.06.194 [21] KIM Y K, KIM G M, LEE J W. Highly porous N-doped carbons impregnated with sodium for efficient CO2 capture[J]. Journal of Materials Chemistry A, 2015, 3(20): 10919-10927. doi: 10.1039/C5TA01776A [22] WANG J C, SENKOVSKA I, OSCHATZ M, et al. Highly porous nitrogen-doped polyimine-based carbons with adjustable microstructures for CO2 capture[J]. Journal of Materials Chemistry A, 2013, 1(36): 10951-10961. doi: 10.1039/c3ta11995e [23] FU N, YANG B, WANG Y, et al. Micromeso hierarchical porous ultrathin nitrogen-doped carbon nanosheets with rough surfaces for efficient gas adsorption and separation[J]. Energy & Fuels, 2022, 36(22): 13705-13712. [24] SHARMA M, SNYDER M A. Facile synthesis of flower-like carbon microspheres for carbon dioxide capture[J]. Microporous and Mesoporous Materials, 2022, 335: 111801. doi: 10.1016/j.micromeso.2022.111801 [25] KOU J H, SUN L B. Fabrication of nitrogen-doped porous carbons for highly efficient CO2 capture: Rational choice of a polymer precursor[J]. Journal of Materials Chemistry A, 2016, 4(44): 17299-17307. doi: 10.1039/C6TA07305K [26] PARSHETTI G K, CHOWDHURY S, BALASUBRAMANIAN R. Biomass derived low-cost microporous adsorbents for efficient CO2 capture[J]. Fuel, 2015, 148: 246-254. doi: 10.1016/j.fuel.2015.01.032 [27] RAO L L, LIU S F, WANG L L, et al. N-doped porous carbons from low-temperature and single-step sodium amide activation of carbonized water chestnut shell with excellent CO2 capture performance[J]. Chemical Engineering Journal, 2019, 359: 428-435. doi: 10.1016/j.cej.2018.11.065 [28] THRUN M C, GEHLERT T, ULTSCH A. Analyzing the fine structure of distributions[J]. PLoS One, 2020, 15(10): e0238835. doi: 10.1371/journal.pone.0238835 [29] 戚兴怡, 胡耀峰, 王若愚, 等. 机器学习在新材料筛选方面的应用进展[J]. 化学学报, 2023, 81(2): 158-174. doi: 10.6023/A22110446 QI X Y, HU Y F, WANG R Y, et al. Recent advance of machine learning in selecting new materials[J]. Acta Chimica Sinica, 2023, 81(2): 158-174 (in Chinese). doi: 10.6023/A22110446
[30] FERNÁNDEZ-DELGADO M, SIRSAT M S, CERNADAS E, et al. An extensive experimental survey of regression methods[J]. Neural Networks, 2019, 111: 11-34. doi: 10.1016/j.neunet.2018.12.010 [31] 李炜, 梁添贵, 林元创, 等. 机器学习辅助高通量筛选金属有机骨架材料[J]. 化学进展, 2022, 34(12): 2619-2637. LI W, LIANG T G, LIN Y C, et al. Machine learning accelerated high-throughput computational screening of metal-organic frameworks[J]. Progress in Chemistry, 2022, 34(12): 2619-2637 (in Chinese).
[32] TANVEER M, RAJANI T, RASTOGI R, et al. Comprehensive review on twin support vector machines[J]. Annals of Operations Research, 2022: 1-46. [33] SITEK W, TRZASKA J. Practical aspects of the design and use of the artificial neural networks in materials engineering[J]. Metals, 2021, 11(11): 1832. doi: 10.3390/met11111832 [34] PALLE K, VUNGUTURI S, GAYATRI S N, et al. The prediction of CO2 adsorption on rice husk activated carbons via deep learning neural network[J]. MRS Communications, 2022, 12(4): 434-440. doi: 10.1557/s43579-022-00197-2 [35] AURET L, ALDRICH C. Interpretation of nonlinear relationships between process variables by use of random forests[J]. Minerals Engineering, 2012, 35: 27-42. doi: 10.1016/j.mineng.2012.05.008 [36] ARASHI M, LUKMAN A F, ALGAMAL Z Y. Liu regression after random forest for prediction and modeling in high dimension[J]. Journal of Chemometrics, 2022, 36(4): e3393. doi: 10.1002/cem.3393 [37] MIZUMOTO A. Calculating the relative importance of multiple regression predictor variables using dominance analysis and random forests[J]. Language Learning, 2023, 73(1): 161-196. doi: 10.1111/lang.12518 [38] ZHANG X, ZHANG S H, YANG H P, et al. Influence of NH3/CO2 modification on the characteristic of biochar and the CO2 capture[J]. BioEnergy Research, 2013, 6(4): 1147-1153. doi: 10.1007/s12155-013-9304-9 [39] OUYANG L K, XIAO J F, JIANG H S, et al. Nitrogen-doped porous carbon materials derived from graphene oxide/melamine resin composites for CO2 adsorption[J]. Molecules, 2021, 26(17): 5293. doi: 10.3390/molecules26175293 [40] QI S C, LIU Y, PENG A Z, et al. Fabrication of porous carbons from mesitylene for highly efficient CO2 capture: A rational choice improving the carbon loop[J]. Chemical Engineering Journal, 2019, 361: 945-952. doi: 10.1016/j.cej.2018.12.167