High-throughput computation and machine learning studies are conducted for the rational design of ensembles of alloy catalysts for selective catalytic oxidation of NH3 (NH3-SCO). High-throughput calculation is first performed for the binding energy of intermediates on all possible sites of low-index facets. The general features from intrinsic information of the surface, adsorbate, and binding site are proposed for binding energy prediction on alloy catalysts, and highly accurate tree-based ensemble machine learning models with low MAE of <0.15 eV and high R2 of ≥0.97 are obtained for binding energy prediction of key intermediates involved in NH3-SCO. Furthermore, binding energies of stable configurations of pivotal intermediates in NH3-SCO are extracted to construct the new data set without known labels about catalytic performance, and unsupervised clustering analyses are conducted twice to quickly categorize ensemble systems into different sets for searching promising candidate catalysts. The different clusters are identified, and primary elementary processes of NH3-SCO on representative ensemble systems of the clusters are studied to further reveal the properties of clusters and the clusters with promising catalysts are found ultimately. Our theoretical study could provide a novel strategy for the rational design of catalysts and accelerate the discovery of new alloy catalysts for NH3-SCO.
