Dual-atom catalysts for hydrogen evolution reaction have received widespread attention, but precise screening and prediction of high-performance catalysts through simple methods remains a challenge. In this study, we perform high-throughput density functional theory (DFT) calculation and machine learning (ML) to screen and predict the transition metal dual-atom catalysts with N-doped graphene support (TMDACs) for acidic hydrogen evolution reaction (HER). The Fe_Zn and V_Fe DACs were proposed to be the most promising candidates for Pt-based catalyst toward acidic HER from 406 TMDACs, based on the characteristics of HER activity, formation, thermodynamic stability, abundance, environmental friendliness. The Fe_Zn and V_Fe DACs with excellent HER performance is due to the synergistic effect deriving from the interaction between H and dual metal atoms in TMDACs. By determining 6 different ML models with four kind of input features, we find the artificial neural networks (ANN) model can predict the HER performance of TMDACs most accurately only using simple input features, including one-hot-encoding of atomic number and Gibbs free energies of transition metal single-atom catalysts. This work not only proposed the potential TMDACs with high HER performance, but also verified that the ANN model can accurately predict the HER activity of diatomic catalysts with simple input features.

中文翻译：

---

基于高通量计算和简单特征搜索替代预测模型的析氢反应过渡金属双原子催化剂筛选

用于析氢反应的双原子催化剂受到了广泛关注，但通过简单的方法精确筛选和预测高性能催化剂仍然是一个挑战。在本研究中，我们进行高通量密度泛函理论（DFT）计算和机器学习（ML）来筛选和预测用于酸性析氢反应（HER）的氮掺杂石墨烯载体过渡金属双原子催化剂（TMDAC） 。基于 HER 活性、形成、热力学稳定性、丰度和环境友好性等特点，Fe_Zn 和 V_Fe DAC 被认为是 406 TMDAC 中最有前途的酸性 HER 铂基催化剂候选者。 Fe_Zn和V_Fe DAC具有优异的HER性能，这是由于TMDAC中H和双金属原子之间的相互作用产生的协同效应。通过确定具有四种输入特征的 6 种不同的 ML 模型，我们发现人工神经网络（ANN）模型仅使用简单的输入特征（包括原子序数的 one-hot 编码和吉布斯自由）就可以最准确地预测 TMDAC 的 HER 性能过渡金属单原子催化剂的能量。这项工作不仅提出了具有高 HER 性能的潜在 TMDAC，而且验证了 ANN 模型可以通过简单的输入特征准确预测双原子催化剂的 HER 活性。