Predicted C, O, H, and S adsorption energies on bimetallic surfaces (with a M1:M2 ratio of 3)

The best-performing ML model was then applied to a list of bimetallic alloys, the adsorption energies of which were not readily available. A total of 24 metal elements were considered and permuted with one another, which generated a set of over 500 bimetallic alloys. One of the input features used for the ML model is the ratio of the two individual components within the binary system. By changing the numerical value of the “ratio” feature, the ML model is able to deal with a given binary alloy with any M1 or M2 concentration. In this work, we focused on bimetallic materials with a M1:M2 ratio of 3 (i.e. 75 mol.% of M1 and 25 mol.% of M2).

性能最优的机器学习（Machine Learning）模型随后被应用于一系列吸附能暂未公开可得的双金属合金。本次研究共纳入24种金属元素并进行两两排列组合，由此生成了超过500种双金属合金集合。该机器学习模型的输入特征之一为该二元体系中两种金属组分的占比。通过调整“占比”特征的数值，该模型可处理任意M1或M2组分占比的给定二元合金体系。本研究聚焦于M1与M2组分占比为3:1的双金属材料（即M1占比75 mol.%，M2占比25 mol.%）。