Although eutectic solvents (ESs) have garnered significant attention as promising solvents for carbon dioxide (CO2) capture, systematic studies on discovering novel ESs linking machine learning (ML) and experimental validation are scarce. For the reliable prediction of CO2‐in‐ES solubility, ensemble ML modeling based on random forest and extreme gradient boosting with inputs of COSMO‐RS derived molecular descriptors is rigorously performed, for which an extensive experimental CO2‐in‐ES solubility database of 2438 data points in 162 ESs involving 106 ES systems are collected. With the best‐performing model obtained, the CO2 solubilities of 4735 novel combinations of ES components are first predicted for estimating their potential in CO2 capture. The top‐ranked candidate combinations are subsequently evaluated by examining the environmental health and safety properties of individual components and assessing the potential operating window based on solid–liquid equilibrium (SLE) prediction. Three most promising ES systems are finally retained, which are thoroughly studied by SLE and CO2 absorption experiments.

中文翻译：

---

机器学习通过实验验证促进了用于 CO2 捕获的共晶溶剂设计


尽管共熔溶剂 （ES） 作为有前途的二氧化碳 （CO2） 捕获溶剂而受到广泛关注，但关于发现将机器学习 （ML） 和实验验证联系起来的新型 ES 的系统研究很少。为了可靠地预测 CO2-in-ES 溶解度，严格执行了基于随机森林和极端梯度提升的集成 ML 建模，并使用 COSMO-RS 衍生分子描述符的输入，为此收集了一个广泛的实验 CO2-in-ES 溶解度数据库，其中包含 162 个 ES 中的 2438 个数据点，涉及 106 个 ES 系统。通过获得性能最佳的模型，首先预测了 4735 种 ES 成分的新组合的 CO2 溶解度，以估计它们在 CO2 捕获中的潜力。随后，通过检查单个组件的环境健康和安全特性，并根据固液平衡 （SLE） 预测评估潜在的操作窗口，对排名靠前的候选组合进行评估。最终保留了三个最有前途的 ES 系统，并通过 SLE 和 CO2 吸收实验对其进行了深入研究。