The design and discovery of catalysts for simultaneous catalysis of chlorobenzene and nitrogen oxides via domain knowledge guided machine learning

Abstract

A domain knowledge guided machine learning (ML) strategy is proposed to design and discover catalysts for the simultaneous catalysis of chlorobenzene (CB) and nitrogen oxides (NOx). Symbolic regression (SR) is used to find an analytic formula for the acidity of metallic ions and a few ML regression models are developed for the redox ability in terms of standard reduction potential (SRP) of metallic ions, resulting in the recommendations of bi-functions catalysts. The domain knowledge considers that metallic ions with acidity values in the range of 0.5–1.2 are suitable for both catalytic support and active components, while catalytic active component requires metallic ions with high redox ability (SRP>0.4). Symbolic regression gives an analytic formula of ion acidity and the support vector regression model finds the redox ability from the available features. At last, experimental and characterization results verify catalytic performances of the screened-out catalyst formulas.