The capture and storage of toxic industrial chemicals such as H₂S using porous polymer networks (PPNs) has shown promising application because of their high porosity, high surface area, high stability, low-cost and lightweight. In this work, 17,846 PPNs with the diamond-like topology were computationally screened to identify the optimal adsorbents for the removal of H₂S and CO₂ from humid natural gas based on the combination of molecular simulation and machine learning algorithms. The top-performing PPNs such as hPAFs-0201 with the highest adsorption performance scores (APS) were evaluated and identified based on their adsorption capacities and selectivity for H₂S and CO₂. The strong affinity between water molecules and the framework atoms in a few PPNs has a significant impact on the adsorption selectivity of acid gases. Based on decision tree analysis, we found two main design paths of the optimal PPNs for natural gas sweetening, which are the PPNs with LCD ≤ 4.648 Å, V_f ≤ 0.035, and PLD ≤ 3.889 Å, and those with 4.648 Å ≤ LCD ≤ 5.959 Å, ρ ≤ 837 kg m⁻³. In addition, we constructed different machine learning models, particularly artificial neural network, available to accurately predict the APS of PPNs. 2D projection map of geometrical properties of PPNs using the t-distributed stochastic neighbor embedding (t-SNE) method shows that the screened 390 samples exhibit the similar structures. Among the top-23 PPNs with the highest APS, hPAFs-0201 has enhanced natural gas sweetening performance due to its strong affinity between the N-rich organic linkers and acid gases. hPAFs-0752 shows the highest isosteric adsorption heat of H₂S and CO₂ (Q°_st = 49.84 kJ mol⁻¹), resulting in its second-highest APS as well as high hydrophilicity. Based on the combination of molecular simulation and machine learning, comprehensive insights into the high-throughput screening of PPNs in this work will provide new ideas for the design of high-performance PPNs for gas separation.

中文翻译：

---

基于神经网络的天然气脱硫多孔聚合物网络高通量计算筛选

由于多孔聚合物网络（PPNs）具有高孔隙率、高表面积、高稳定性、低成本和轻质等特点，在捕获和储存有毒工业化学品（如 H ₂ S）方面具有广阔的应用前景。在这项工作中，基于分子模拟和机器学习算法的结合，计算筛选了 17,846 个具有类金刚石拓扑结构的 PPN，以确定用于从潮湿天然气中去除 H ₂ S 和 CO ₂的最佳吸附剂。根据其吸附容量和对 H ₂ S 和 CO _{2 的}选择性，评估和鉴定了性能最佳的 PPN，例如具有最高吸附性能评分 (APS) 的 hPAFs-0201. 少数PPNs中水分子与骨架原子之间的强亲和力对酸性气体的吸附选择性有显着影响。基于决策树分析，我们发现4.648埃，对天然气脱硫最佳的PPN，这是LCD屏的PPN≤两个主要的设计路径V _˚F  ≤0.035，PLD≤3.889埃，那些4.648 A≤LCD≤ 5.959 Å, ρ  ≤ 837 kg m ^-3. 此外，我们构建了不同的机器学习模型，特别是人工神经网络，可用于准确预测 PPN 的 APS。使用 t 分布随机邻域嵌入 (t-SNE) 方法的 PPN 几何特性的二维投影图表明，筛选出的 390 个样本表现出相似的结构。在具有最高 APS 的前 23 个 PPN 中，hPAFs-0201 由于其富含 N 的有机连接体和酸性气体之间的强亲和力而增强了天然气脱硫性能。hPAFs-0752 显示出 H ₂ S 和 CO ₂的最高等量吸附热（Q ° _st  = 49.84 kJ mol ^-1)，导致其第二高的 APS 以及高亲水性。基于分子模拟和机器学习相结合，对本工作中PPNs高通量筛选的综合洞察，将为设计用于气体分离的高性能PPNs提供新思路。