Accelerated identification of high-performance catalysts for low-temperature NH3-SCR by machine learning

Significant efforts have been devoted to the development of catalysts for the removal of environmental pollutants. However, screening catalysts through trial and error consumes a lot of time and resources. Here we present a machine learning approach for selective catalytic reduction (SCR) catalyst discovery based on a custom-build database covering over 2000 related reports. Catalyst characteristics and working conditions were selected as features to predict catalyst activities. The extra tree regression model was identified as the best performer among the eight algorithms examined here for this purpose. Elements with strong oxidizing ability, such as Mn, as the active component were found to be important for developing excellent low temperature SCR catalysts using the calculated feature importance scores. The optimized machine learning model was used to aid the identification of Mn-based SCR catalysts and the NO conversion rate of the Mn–Ce–Co catalyst is greater than 80% in a wide temperature range of 150–300 °C.