文献类型：期刊文献

英文题名：High-throughput screening of efficient crosslinking agent used in the carbon materials preparation from aromatics-enriched oil via a synergistic strategy of DFT and machine learning

作者：Huang, Jian[1];Liu, Cao[1];Zhang, Yuanqin[1];Zhou, Zhenyu[1];Ren, Mannian[3];Cui, Lingrui[1,2];Cao, Fahai[1,2];Xu, Jun[1]

机构：[1]East China Univ Sci & Technol, Sch Chem Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Large Ind Reactor Engn Res Ctr, Minist Educ, Shanghai 200237, Peoples R China;[3]Sinopec Henan Refinery & Petrochem Co, Luoyang 471003, Peoples R China

年份：2026

卷号：423

外文期刊名：FUEL

收录：;EI(收录号：20261520463551);WOS:【SCI-EXPANDED(收录号:WOS:001733892800001)】；

基金：This research is financially supported by National Natural Science Foundation of P.R.China (22308104) .

语种：英文

外文关键词：Crosslinking agent; Machine learning; High-throughput screening; Carbon materials

摘要：As the by-products of crude oil processing, the aromatics-enriched oils, including fluid catalytic cracking slurry oil and ethylene tar, are superior raw materials for the preparation of high value-added carbon materials. The core of this preparation process lies in the conversion of polycyclic aromatics into macromolecules with twodimensional linear structures through crosslinking reaction, whose industrial application is limited by the high cost and low reaction yields of conventional crosslinking agents. Herein, we propose a synergistic strategy combining density functional theory (DFT) and machine learning (ML) to screen novel crosslinking agents. Using alpha-methylnaphthalene (alpha-MN) as a model aromatic, the reaction pathways and Gibbs free energy barriers (Delta G) of 50 candidate crosslinking agents were systematically calculated. By combining quantum-chemical descriptors (Type-I) with topology-derived descriptors (Type-II) and employing a random forest (RF) algorithm, a model capable of accurately predicting reaction energy barriers was developed. Feature importance and SHAP analysis revealed that mu, LUMO, LEAE_Var, BCUTare-1 l, and ATSC1i are key parameters influencing reaction activity. Guided by this model, a high-throughput predictions on 8,923 alcohols were conducted, from which eight molecules were selected for DFT calculations. Combined with experiments on the synthesis of B-COPNA resins, the accuracy of constructed model was validated. Both computational and experimental results demonstrated that 4-tert-butyl-2,6-bis(hydroxymethyl) phenol (TBBP), as a predicted high-activity molecule, exhibits excellent reaction activity in experimental validation. The DFT-ML collaborative screening framework constructed in this study provides a universal and generalizable novel approach for the molecular design and efficient preparation of aromatics-based carbon materials.