Machine learning-derived stage-specific design rules for metal-organic framework selection in seasonal hydrogen storage

Economic viability of seasonal hydrogen storage depends on minimizing recovery costs as reservoir pressure and hydrogen purity decline during withdrawal, necessitating pressure-swing adsorption purification where adsorbent selectivity directly impacts separation efficiency. Existing metal-organic framework (MOF) screening evaluates performance under idealized conditions such as equimolar mixtures and single pressures, providing no guidance for matching adsorbent geometry to evolving pressure and composition profiles during withdrawal. We establish quantitative, stage-specific design rules using machine learning and explainable AI on 712 experimentally synthesized MOF structures across four withdrawal stages spanning 60 to 25 bar and 98% to 65% hydrogen purity. Optimal geometric parameters controlling selectivity shift systematically across pressure and composition conditions: accessible volume dominates at high pressure, void fraction controls intermediate stages, and pore aperture governs low-pressure performance. Multiple viable pathways at each stage provide synthesis flexibility. These stage-resolved criteria enable MOF selection based on realistic operating conditions rather than idealized scenarios.