文献类型：期刊文献

英文题名：Constructing machine learning framework through defining absorption-desorption index for high-throughput screening of VOC absorbents

作者：Jiang, Zhenwu[1];Wen, Haiyang[1];Liu, Chuanlei[1];Zhao, Qiyue[1];Cui, Yupeng[1];Xu, Mengna[1];Wang, Lan[1];Gao, Wenhao[1];Shen, Benxian[1,2];Sun, Hui[1,2,3]

机构：[1]East China Univ Sci & Technol, 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]Xinjiang Univ, Sch Chem Engn & Technol, Minist Key Lab Oil & Gas Fine Chem, Urumqi 830046, Peoples R China

年份：2026

卷号：403

外文期刊名：SEPARATION AND PURIFICATION TECHNOLOGY

收录：;EI(收录号：20262420892847);WOS:【SCI-EXPANDED(收录号:WOS:001796837200001)】；

基金：This work is financially supported by the National Natural Science Foundation of China (Grant 21878097 and 22178109) .

语种：英文

外文关键词：VOCs; Solvent; COSMO-RS; Machine learning; Absorption-desorption

摘要：Volatile organic compounds (VOCs) are key precursors of ozone (O-3) and fine particulate matter (PM2.5), making their abatement essential for air-quality improvement. Solvent absorption is attractive due to high efficiency and operational flexibility, yet absorbent screening often emphasizes absorption capacity alone and lacks a unified metric that simultaneously accounts for regenerability under cyclic operation. Here, we report a machine learning framework through introducing an absorption-desorption index (ADI) for high-throughput screening of VOC absorbents. ADI was defined through using the summed solute and solvent partial pressures during absorption (P-A) and the relative volatility between solvent and solute during desorption (R-D), enabling integrated evaluation of absorption and regeneration. Using n-hexane as a model VOC, three solvents, ethyl dodecanoate (EDD), dioctyl ether (DOE), and isopropyl myristate (IPM) were identified as candidates and experimentally validated against literature, COSMOtherm library-optimal, and a commercial solvent. On the independent test set, the P-A and R-D models achieved R-2/RMSE/MAE values of 0.904/0.0117/0.0085 and 0.854/0.9584/0.6285, respectively. Experimental validation on six solvents shows that the predicted ADI correlates positively with the measured n-hexane removal efficiencies (Pearson's r = 0.907, p = 0.0125; Spearman's rho = 0.829, p = 0.0416). If only the post-model-development screening stage is considered, screening 100,000 candidates required only similar to 1.53 h, including descriptor generation and model inference. Overall, this work provides a practical prescreening strategy for rapidly prioritizing experimentally testable organic solvents for cyclic VOC capture.