Data‐driven ionic liquid design for methane dehydration: Multi‐feature fusion modeling and molecular insights

Abstract Rapid and reliable prediction of ionic liquids (ILs) thermodynamic properties is critical for their high‐throughput screening and rational molecular design. This study constructs a multi‐feature fusion machine learning model GNN‐COSMOg by integrating molecular graph structure, group counting, and molecular descriptors, realizing accurate prediction of cationic σ ‐profile and COSMO volume, with the relative mean deviations of 2.869% and 1.606%, respectively—markedly outperforming the traditional GC‐COSMO model. Integrated into the computer‐aided molecular design framework, the model was employed for task‐specific ILs discovery. For natural gas dehydration, the ionic liquid 1‐methoxymethyl‐3‐methylimidazolium thiocyanate ([EOMIM][SCN]) was screened out, synthesized, and validated, followed by a detailed mechanistic analysis of its performance. Specifically, at 4.227 bar and 303.15 K, the solubility of methane in [EOMIM][SCN] is merely 0.004 mol/mol, and it exhibits superior dehydration capacity compared to the benchmark solvent—a performance further enhanced by the hydrogen bonding interactions between its cations and water molecules.