文献类型：期刊文献

中文题名：基于机器学习的酸烃界面预测与离子液体设计

英文题名：Machine Learning-Based Acid-Hydrocarbon Interface Prediction and Ionic Liquids Design

作者：田一凡[1];高维群[2,3];卢静宜[1];郑伟中[2,3];孙伟振[2,3]

机构：[1]华东理工大学信息科学与工程学院,上海200237;[2]华东理工大学化学工程与低碳技术全国重点实验室,上海200237;[3]华东理工大学化工学院,上海200237

年份：2025

卷号：51

期号：6

起止页码：793

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

基金：科技部科技创新2030人工智能重大专项(2023ZD0121001);中央高校基础科研业务费支持。

语种：中文

中文关键词：离子液体;机器学习;酸烃界面性质;合理设计;进化算法

外文关键词：ionic liquid;machine learning;acid-hydrocarbon interface property;rational design;evolutionary algorithm

摘要：为了精准预测离子液体对酸烃界面性质的影响并设计出新颖的离子液体组合,本文采用随机森林等机器学习方法,分析离子液体与酸烃界面性质之间的内在联系,并基于不同的描述符构建了酸烃界面性质的预测模型。为了合理设计新型的离子液体组合,将离子液体的SMILES编码转换为连续且数据驱动的分子描述符(Continuous and Data-Driven Molecular Descriptors,CDDD),并利用基于成功历史的自适应差分进化(Success-History Based Adaptive Differential Evolution,SHADE)算法在潜在空间中进行精准搜索和解码,同时,结合子结构约束以确保生成结构的合理性。以烷基化反应中的界面厚度(δ_(w))和界面张力(γ)为例,构建的预测模型在测试集上对界面厚度的决定系数为0.952,界面张力的决定系数为0.901,显示出较高的预测精度。此外,通过SHADE算法优化设计,成功生成了328对满足界面厚度和界面张力要求的离子液体组合,显著扩展了符合标准的离子液体组合的可行域。本文研究不仅提高了对酸烃界面性质的预测能力,还为离子液体的合理设计提供了新的方法和思路。
Ionic liquids,with their green chemistry attributes and tunable properties,have garnered significant interest as potential additives in sulfuric acid-catalyzed C4 alkylation.Given the vast combinatorial possibilities of anion-cation pairs,traditional experimental screening methods are inefficient in exploring extensive chemical spaces,thus falling short.To address this challenge,we employed machine learning techniques—most notably the random forest(RF)algorithm—to establish correlations between the structural features of ionic liquids and their enhanced acidhydrocarbon interfacial properties.This approach enabled the development of predictive models based on diverse descriptors,advancing our understanding and facilitating the selection of ionic liquids for specific applications.Furthermore,to streamline the rational design of novel ionic liquid combinations,we utilized continuous and data-driven molecular descriptors(CDDD)derived from the SMILES codes of the compounds.These descriptors were fed into the Success-History based Adaptive Differential Evolution(SHADE)algorithm,which efficiently navigates and decodes the potential space to identify promising candidates.Substructure constraints were also integrated to ensure the rationality and feasibility of the generated structures.Focusing on key parameters such as interfacial thickness(δ_(w))and tension(γ)in C4 alkylation,the constructed predictive models achieved a determination coefficient(R2)of 0.952 for interfacial thickness and 0.901 for interfacial tension on the test set,indicating high predictive accuracy.Additionally,through optimization via the SHADE algorithm,328 ionic liquid combinations meeting the requirements for interfacial thickness and tension were successfully generated,significantly expanding the feasible range of qualifying ionic liquid combinations.This work not only enhances the capability to predict acidhydrocarbon interfacial properties but also provides novel methodologies and insights for the rational design of ionic liquids.