Artificial neural network aided virtual screening of additives to a Co/SrCO3 catalyst for preferential oxidation of CO in excess hydrogen

Preferential oxidation (PROX) of 0.7–1 vol% CO was investigated using the stoichiometric amount of O2 in excess hydrogen. Cobalt supported on SrCO3 showed high selectivity to PROX of CO, and the new additive to the Co/SrCO3 catalyst was investigated for the high tolerance towards CO2 and H2O. Representative 10 elements (B, K, Sc, Mn, Zn, Nb, Ag, Nd, Re, and Tl) were selected to represent the physicochemical properties of all elements suitable for additives of solid catalyst. A supported cobalt catalyst with one kind of the above additive was prepared for CO PROX reaction. The activities at 240 °C and the physicochemical properties of the 10 elements were used as training data of a radial basis function network (RBFN), a kind of artificial neural network. After the training, the RBFN predicted the catalytic performance of the supported catalyst containing various element X as Co–X/SrCO3. The elements such as Bi, Ga, and In were predicted to be promising additives. Finally, the catalytic performance of these additives was experimentally verified. Sixty four percent of CO conversion and 70% selectivity for PROX at 240 °C was achieved in the presence of excess carbon dioxide and steam by Co 3.2–Bi 0.3 mol%/SrCO3 pretreated at 345 °C.