Artificial-intelligence-assisted design principle for developing high-performance single-atom catalysts
Liangliang Xu, Xingkun Wang, Xiaojuan Hu, Yue Wang, Canhui Zhang, Wenwu Xu, Wenhui Zhao, Ning Xu, Dongyoon Woo, Hanxu Yao, Xiaofan Li, Heqing Jiang, Minghua Huang, Jinwoo Lee, Xiao Cheng Zeng, Zhongkang Han
Abstract
Artificial intelligence (AI)-assisted approaches are powerful means for advancing catalyst design, as they can significantly accelerate the development of novel catalysts. However, the underlying mechanisms of these approaches often remain elusive, which may lead to unreliable results due to a lack of clear understanding of the involved processes. Herein, we present an AI strategy that combines machine learning (ML) and data mining (DM) to identify high-performance catalysts while elucidating the key factors that govern catalytic performance in complex reactions. Applying this AI strategy to evaluate the electrocatalytic oxygen reduction performance of 10,179 single-atom catalysts (SACs), we identified several high-performance SACs and determined the critical influencers of their activity. Experimental validations further confirm the effectiveness of the AI strategy, with the optimal target Co-S 2 N 2 /g-SAC achieving a high half-wave potential of 0.92 V. This AI-assisted approach significantly enhances the transparency and reliability of data-driven discoveries, providing new insights that benefit the rational design of materials.