Machine Learning-Accelerated Disentanglement of Activity-Selectivity Trade-off of Multielement Oxide Denitration Catalysts

Developing rapid screening methods for high-performance catalysts is one of the research priorities in environmental science and engineering. However, the activity-selectivity trade-off is generally observed in developing catalysts. Herein, a machine learning workflow to accelerate disentangling the activity-selectivity trade-off of multielement oxide selective catalytic reduction (SCR) denitration catalysts was proposed. An acidity descriptor and a redox descriptor were defined through the systematic investigation of 12 experimentally synthesized MnCe-based catalysts. Assisted with the above-constructed acidity and redox descriptors, a clustering analysis method was successfully applied for rapid screening of the high-performance multielement oxide catalysts. Combining the results of clustering analysis and the performance of the 12 measured catalysts, a group of catalysts with potentially high activity and high N-2 selectivity were identified. Three candidate catalysts in this group were evaluated experimentally and exhibited far superior activity and selectivity to those of MnCeOx, indicating the high efficiency of the machine learning strategy. This work provides new ideas for future green catalyst design and leads to significant progress in environmental protection and sustainable resource utilization.