Machine Learning-Aided Catalyst Modification in Oxidative Coupling of Methane via Manganese Promoter

Abstract

To enhance CH4 conversion values in oxidative coupling of methane (OCM) reactions under an O2-lean condition (CH4/O2 = 6.0), a support vector regression (SVR) and one-hot encoding manner implemented machine learning (ML) is examined. From an open-source high-throughput screening (HTS) database of 300 OCM catalysts made by random sampling from a materials space, the top 10 three-element-supported catalysts with C2 yield higher than 11.0% and C2 selectivity higher than 80.0% at CH4/O2 = 6.0 were selected as targets for modification. Then, ML-aided investigation of an additive fourth element as a promoter was performed at the SVR field based on the HTS database among 350 catalysts (40,330 data points). Application of one-hot encoding to ascertain positive elements for CH4 conversion revealed that manganese (Mn) frequently appears at CH4 conversion higher than 44.0%. After the 10 selected catalysts were prepared with the Mn additive, their OCM performance was compared with those of pristine three-element-supported catalysts. Results show that four catalysts represent positive features on C2 yield in the presence of additive Mn working as a promoter. Consequently, 5 wt % Mn-loaded LiFeBa/La2O3 and LiBaLa/La2O3, respectively, show attractive OCM performance of 16.3% C2 yield with 88.4% selectivity and 13.8% C2 yield with 71.9% selectivity, even under an O2-lean condition (CH4/O2 = 6.0).