文献类型：期刊文献

英文题名：Machine-learning-guided reaction kinetics prediction towards solvent identification for chemical absorption of carbonyl sulfide

作者：Chen, Yuxiang[1];Liu, Chuanlei[1];Guo, Guanchu[1];Zhao, Yang[1];Qian, Cheng[1];Jiang, Hao[1];Shen, Benxian[1,2];Wu, Di[3,4,5,6];Cao, Fahai[1];Sun, Hui[1,2]

机构：[1]East China Univ Sci & Te chnol, Sch Chem Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Int Joint Res Ctr Green Energy Chem Engn, Shanghai 200237, Peoples R China;[3]Washington State Univ, Alexandra Navrotsky Inst Expt Thermodynam, Pullman, WA 99163 USA;[4]Washington State Univ, Gene & Linda Voiland Sch Chem Engn & Bioengn, Pullman, WA 99163 USA;[5]Washington State Univ, Dept Chem, Pullman, WA 99163 USA;[6]Washington State Univ, Mat Sci & Engn, Pullman, WA USA

年份：2022

卷号：444

外文期刊名：CHEMICAL ENGINEERING JOURNAL

收录：;EI(收录号：20221912081430);WOS:【SCI-EXPANDED(收录号:WOS:000800419500005)】；

基金：This work is financially supported by the National Natural ScienceFoundation of China (Grant 21878097 and 22178109) and the Natural Science Foundation of Shanghai (Grant 21ZR1417700) . D.W. acknowledges the institutional funds from the Gene and Linda Voiland School Chemical Engineering and Bioengineering, and the Alexandra Navrotsky Institute for Experimental Thermodynamics at Washington State University.

语种：英文

外文关键词：Reaction kinetics; Machine learning; Carbonyl sulfide; Chemical absorption

摘要：Exploring solvents for chemical absorption of carbonyl sulfide (COS) is largely hindered because of challenging study of reaction kinetics by means of either experimental or computational methods. Machine learning (ML) has been proven to greatly accelerate information identification and function achievement in academic and industrial fields. Herein, we report a machine-learning-guided reaction kinetics prediction for intelligent identification of promising compounds for chemical absorption of COS. The rate-determining step for solvent-involved reaction of COS was firstly recognized and the reaction rate constants of determining step were obtained via quantum chemistry computations in order to establish the initial training dataset. Furthermore, two molecular descriptors, alphaV and betaV, representing the steric hindrance around the amine groups were defined to build the ML algorithm and the SHapley Additive exPlanations (SHAP) approach was used to interpret the model. By introducing four specific descriptors, our ML model exhibits the largely reduced mean absolute error (MAE) of 0.80 for predicting the rate constants (logk) for COS reaction with potential solvents. These reaction rates are primarily determined by the charge distribution and steric hindrance of solvent molecules. Documented data confirm that our ML model involving custom-defined molecular descriptors successfully predict the reaction kinetics. This study enables a general strategy for machine-learning-guided exploration and identification of solvent candidates for compound capture and separation through chemical absorption process.