Machine Learning Accelerated Bilayer MN₄-X-MN₄ Structure Construction And ORR/OER Catalytic Performance Study

The high cost and low efficiency of density functional theory（DFT）calculations are one of the obstacles to theoretical design in energy conversion materials design and research.Machine learning（ML）as an efficient learning method can accelerate the material selection process,improve the computational efficiency and reduce the computational cost.Monolayer metal-nitrogen-carbon catalysts for cathode oxygen reduction reaction（ORR）in fuel cells and anode oxygen evolution reaction（OER）in electrolytic water hydrogen generators show potential to replace noble metals,but their catalytic activity and stability still face great challenges.To address the low efficiency of fuel cell ORR and electrolytic water hydrogen generator OER,this paper uses ML and DFT combination methods to design bilayer bridge-bonded X-ligand MN₄-X-MN₄（X is more electronegative O and N）catalyst structures to rapidly screen high performance ORR and OER catalysts and explore the mechanism of different transition metals and ligand environment modulation on the metal-electron structure of the active site.The details are as follows:（1）Stability,catalytic performance and training set of bilayer bridge-bonded X-ligand MN₄-X-MN₄catalysts.The bridge-bonded ligands MN₄-O-MN₄and MN₄-N-MN₄each contain 225 combinatorial structures（upper and lower transition metal M combinations）,24combinatorial structures are randomly selected as the initial training set for machine learning,and the DFT method is used to design and optimize the geometric structure and calculate the formation energy to analyze the stability;explore the relationship between the free energy and the overpotential of the MN₄-X-MN₄structure,and clarify the active site metal the changes of metal adsorption strength on reaction intermediates.（2）Machine learning algorithm training and catalytic performance prediction.Four machine learning algorithms,GBR,RFR,SVR and KNR,were used for training,and the GBR-1,GBR-2,GBR-3 and RFR-1 algorithms with the best fit were selected to predict the catalyst overpotential and analyze the ML model feature importance to predict the catalysts with excellent ORR and OER from 225 MN₄-O-MN₄and 225 MN₄-N-MN₄structures performance of catalysts.（3）DFT reverse validation of ML prediction results and their catalytic performance.The reverse validation of ML prediction results using DFT showed that the average absolute error of ML prediction results was only 0.03 V～0.09 V.From 225 MN₄-O-MN₄structures,Co N₄-O-Rh N₄(?_ORR=0.34 V)and Rh N₄-O-Ag N₄(?_OER=0.29 V)as the best ORR and OER catalysts,respectively.From 225 MN₄-N-MN₄structures,Fe N₄-N-Mn N₄(?_ORR=0.31 V)and Fe N₄-N-Ni N₄(?_OER=0.32 V)were screened as the best ORR and OER catalysts,respectively;the electronic properties indicated that the charge transfer of the active center metal,the d-band center are regulated by the transition metal and ligand environment;the chemical bond strength of the reaction intermediate and the active site metal are modified under the regulation of the transition metal and ligand environment,and the catalysts have higher ORR and OER performance,which can be applied to fuel cell cathode and anode catalyst of electrolytic water to hydrogen device.