Enhancing\nBallistic Transport and C<sub>3</sub>–C<sub>4</sub> Alcohol\nDehydration through Machine Learning-Designed Cationic\nGraphene Oxide Membranes

Machine learning (ML) plays a pivotal role in material\ndesign and\nperformance prediction. However, research in ML related to fabricating\ntwo-dimensional (2D) graphene oxide (GO) membranes remains limited,\nfacing challenges due to inherent structural variations and the need\nfor precise modifications. Inspired by biological cells, this study\nhighlights the importance of incorporating cations into GO membranes\nto enhance ballistic transport and alcohol dehydration performance.\nThrough the exploration of different cations, it is identified that\nthe Ca²⁺-GO membrane not only stabilizes the membrane structure\nby hydrogen bonding interactions, but also maximizes the water-capture\nability of GO membranes by electrostatic attractions. For the first\ntime, the CatBoost algorithm is employed in conjunction with Monte\nCarlo-molecular dynamics simulations to quantitatively assess the\ncorrelation and feature importance of operating temperature, chemical\ngroup, cationic loadings, cationic size, and its charges with membrane\nperformance. A backpropagation ML algorithm is then developed to generate\nthe post-training response for performance prediction with an accuracy\nabove 0.96. Optimal Ca²⁺-GO performance is predicted at\n32.1 mg·g^–1 cationic loading, with water separation\nfactors of 5922 and 46,369 for alcohol (C₃–C₄) dehydration, respectively, and water permeance ranging from\n48.5 to 123.6 GPU, nearly 10 times higher than commercial membranes.\nThis theoretical study pioneers an accurate ML algorithm to fabricate\nthe cationic GO membranes, serving as a blueprint for developing high-performance\n2D membranes for alcohol dehydration.