Research Information System
Molecular Catalysis, vol.591, 2026 (SCI-Expanded, Scopus)
The increasing demand for sustainable energy has made hydrogen important as a clean energy carrier. Formic acid (FA), a biomass-derived liquid, is a promising approach for hydrogen storage media due to its high hydrogen content. High-efficiency FA dehydrogenation is a challenging goal, particularly due to difficulties in catalyst design, such as the agglomeration of subnanometric metal nanostructures within porous support materials. In this study, hierarchical MFI zeolites were synthesized via seed-assisted crystallization using a multiple quaternary ammonium-based structure-guiding agent (SDA), and different types of metals (Pd, Co, Ni, and Cu) were confined via ethylenediamine-ligand protection. This integrated synthesis approach ensured the homogeneous positioning of metal nanosites (MNS) within the zeolite lattice structure, creating thermally stabilized nanosheets. The resulting catalysts were tested in FA dehydrogenation reactions, proving that this synthesis approach is effective in designing active and stable catalytic systems. Among these catalysts, Pd(0.2)@hMFI achieved the highest catalytic activity under optimized reaction conditions, obtaining a conversion frequency (TOF) of 1801.25 h-1. The presence of secondary mesoporosity within the hierarchical structure improved mass transfer while enhancing the distribution of homogeneous subnanometric metal sites and their reusability. The catalyst retained >90 % of its initial activity after 5 cycles, proving the stability and cost-effectiveness of hierarchical zeolite-based systems with embedded subnanometric active sites for energy applications. To support and improve the experimental data, machine learning models were developed to predict hydrogen production depending on catalyst type, temperature, time, and FA/SF ratio parameters. The XGBoost model achieved the highest accuracy (RMSE = 0.46 and R2 = 0.998) among the tested models, demonstrating the effectiveness of ensemble learning for reliable H2 prediction.