Machine Learning to Optimize Electrocatalysis in the Oxygen Evolution Reaction

Water electrolysis to produce green hydrogen is a sustainable approach for energy storage. Water electrolysis is limited by electrocatalytic performance in the anodic oxygen evolution reaction (OER). Electrocatalytic activity is determined by factors such as porosity, conductivity, and especially defect sites. A novel synthesis procedure developed by the Folkman lab uses a one-step solid-gas reaction that forms transition metal nitrides (TMNs) based on cobalt, iron, and nickel. The synthesis parameters of the TMNs including chemical composition (metal stoichiometry), time, and temperature dictate material properties like porosity, conductivity, and defect sites in correlated/unpredictable ways, making rational catalyst optimization difficult. Machine learning (ML) algorithms, especially Support Vector Regression (SVR) and Gaussian Process Regression (GPR) are useful for modeling complex systems in high dimensional spaces. Using GPR coupled with SVR and iteratively adding experimental data enables optimization of catalyst performance and may provide insights with respect to the experimental factors governing it. Ultimately, these studies will validate the use of ML in catalyst discovery and optimization will yield next-generation TMN catalysts for water electrolysis based on earth-abundant metals.