Abstract: The quest for efficient and stable solid-state hydrogen storage materials at room temperature remains a formidable challenge. This work presents a machine learning (ML)-guided framework to systematically evaluate the hydrogen storage performance of orthorhombic MBene (ort-MBene) materials. A comprehensive set of 18 ort-MBenes was investigated using density functional theory (DFT), revealing a strong correlation between their geometric characteristics and the charge transfer of the transition metals. Through this integrated ML-DFT approach, five promising MBene structures—Ti₂B₂, Cr₂B₂, Mn₂B₂, Zr₂B₂, and Hf₂B₂—were identified, exhibiting exceptional hydrogen adsorption capacities with preferential adsorption occurring at two bridge sites. The analysis of feature importance and Shapley additive explanations revealed that structural parameters such as the ratio of lattice constants (b/a) and the length of bonds (ΔL) were critical in determining adsorption stability, accounting for 71% of the variance in adsorption energy (E_av). Crucially, Ti₂B₂ demonstrates room-temperature viability: AIMD simulations confirm rapid hydrogen release (3.27 wt% within 0.5 ps) at 300-400 K. The results show that this ML-DFT framework can effectively accelerate the discovery of novel hydrogen storage materials, advance the fundamental design principles of MBene-based hydrogen adsorption materials, and serve as a valuable tool for future developments in the field of solid-state hydrogen storage.

Key words: Machine learning, Mbene, Density functional theory, Hydrogen storage, Geometric features