Machine learning-assisted optimization of element ratios in high-entropy alloys for hydrogen evolution reaction

Abstract

As high-entropy alloys (HEAs) are widely applied in energy conversion and catalysis, efficiently and accurately designing HEAs with excellent catalytic performance has become a key challenge in research. Traditional HEAs design methods rely mainly on experience and extensive experiments, which are low in efficiency and high in cost. To overcome these challenges, density functional theory (DFT) calculations and machine learning (ML) methods have gradually been applied to the performance prediction and design of HEAs. We propose the “Differentiated Feature” method to train high-precision Light Gradient Boosting Machine (LGBM) models to predict the catalytic performance of Fe_aCo_bNicCu_dMo_e HEAs (0.18 ​< ​a, b, c, d, e ​< ​0.23, a+b ​+ ​c ​+ ​d ​+ ​e ​= ​1) in the hydrogen evolution reaction (HER). By combining DFT calculation results with machine learning models, we successfully identify Fe_0.222Co_0.185Ni_0.185Cu_0.203Mo_0.203 HEAs with the best HER performance. Comparing the prediction results with experimental and DFT calculation data further validates the effectiveness of this method in predicting HER performance. This study provides new insights into the design of HEAs and accelerates the development of high-performance HEAs.