The vast chemical compositional space presents challenges in catalyst development using traditional methods.Machine learning(ML)offers new opportunities,but current ML models are typically limited to screening a singl...The vast chemical compositional space presents challenges in catalyst development using traditional methods.Machine learning(ML)offers new opportunities,but current ML models are typically limited to screening a single catalyst type.In this work,we developed an efficient ML model to predict hydrogen evolution reaction(HER)activity across diverse catalysts.By minimizing features,we introduced a key energy-related featureφ=Nd0^(2)=ψ0,which correlates with HER free energy.Using just ten features,the Extremely Randomized Trees model achieved R^(2)=0.922.We predicted 132 new catalysts from the Material Project database,among which several exhibited promising HER performance.The time consumed by theML model for predictions is one 200,000th of that required by traditional density functional theory(DFT)methods.The model provides an efficient approach for discovering high-performance HER catalysts using a small number of key features and offers insights for the development of other catalysts.展开更多
基金supported by the National Key R&D Program of China(Grant No.2021YFB3500403)the Youth Fund of the National Natural Science Foundation of China(Grant no.52305443).We gratefully acknowledge HZWTECH for providing computational facilities.
文摘The vast chemical compositional space presents challenges in catalyst development using traditional methods.Machine learning(ML)offers new opportunities,but current ML models are typically limited to screening a single catalyst type.In this work,we developed an efficient ML model to predict hydrogen evolution reaction(HER)activity across diverse catalysts.By minimizing features,we introduced a key energy-related featureφ=Nd0^(2)=ψ0,which correlates with HER free energy.Using just ten features,the Extremely Randomized Trees model achieved R^(2)=0.922.We predicted 132 new catalysts from the Material Project database,among which several exhibited promising HER performance.The time consumed by theML model for predictions is one 200,000th of that required by traditional density functional theory(DFT)methods.The model provides an efficient approach for discovering high-performance HER catalysts using a small number of key features and offers insights for the development of other catalysts.
