Deep eutectic solvents (DESs) are recently proposed as green materials to remove nitric oxide (NO) from released streams into the atmosphere. The mathematical aspect of this process attracted less attention than it deserved. A straightforward approach in this field will help engineer DES chemistry and optimize the equilibrium conditions to maximize the amount of removed NO. This study covers this gap by constructing a reliable artificial neural network (ANN) to correlate the NO removal capacity of DES with equilibrium pressure/temperature and solvent chemistry. So, firstly, the physical meaningful features are selected to make the DES chemistry quantitative. It was found that the density is the best representative for the hydrogen-bound acceptor and hydrogen-bound donor. Also, the density and viscosity of the DESs exhibit the highest correlation with the NO solubility. Then, the hyperparameters of three famous ANN types (feedforward, recurrent, and cascade) are determined by combining trial-and-error and sensitivity analyzes. Finally, the ranking test distinguishes the ANN type with the lowest uncertainty toward estimating NO dissolution in DESs. The cascade neural network (CNN) with twelve and one neurons in the hidden and output layers equipped with the tangent hyperbolic and radial basis transfer functions is identified as the best ANN type for the given purpose. This model predicts 292 DES-NO equilibrium records collected from the literature with mean absolute errors = 0.033, relative absolute errors = 1.49 %, mean squared errors = 0.002, and coefficient of determination = 0.9998. Also, the present study helps understand the role of DES chemistry and operating conditions on the amount of removable NO by DESs. 1,3-dimethylthioureaP4444Cl (3:1) is recognized as the best DES to separate NO molecules from gaseous streams, respectively. The simulation results show that the unit mass of the best DES is capable of absorbing up to ∼27 mol of NO.

中文翻译：

---

机器学习辅助预测一氧化氮 （NO） 在各种共熔溶剂中的溶解度


最近有人提议将深共熔溶剂 （DES） 作为绿色材料，用于去除释放到大气中的液流中的一氧化氮 （NO）。这个过程的数学方面受到的关注不如它应有的关注。该领域的简单方法将有助于设计 DES 化学并优化平衡条件，以最大限度地提高去除的 NO 量。本研究通过构建可靠的人工神经网络 （ANN） 来填补这一空白，以将 DES 的 NO 去除能力与平衡压力/温度和溶剂化学相关联。因此，首先，选择具有物理意义的特征以使 DES 化学具有定量性。研究发现，密度是氢结合受体和氢结合供体的最佳代表。此外，DES 的密度和粘度与 NO 溶解度的相关性最高。然后，通过结合试错法和灵敏度分析来确定三种著名的 ANN 类型 （前馈、递归和级联） 的超参数。最后，排序检验区分了在估计 DES 中 NO 溶出时具有最低不确定性的 ANN 类型。在隐藏层和输出层中具有 12 个和 1 个神经元的级联神经网络 （CNN） 配备了切线双曲和径向基传递函数，被确定为适合给定目的的最佳 ANN 类型。该模型预测从文献中收集的 292 条 DES-NO 平衡记录，平均绝对误差 = 0.033，相对绝对误差 = 1.49%，均方误差 = 0.002，决定系数 = 0.9998。此外，本研究有助于了解 DES 化学和操作条件对 DES 可去除 NO 量的作用。 1,3-二甲基硫脲P4444Cl （3：1） 被认为是分别从气态流中分离 NO 分子的最佳 DES。模拟结果表明，最佳 DES 的单位质量能够吸收高达 ∼27 mol 的 NO。