Machine Learning Exploration of Experimental Conditions for Optimized Electrochemical CO2 Reduction

Electrochemical CO2 reduction has attracted significant attention as a potential method to close the carbon cycle. In this study, we investigated the impact of the electrode fabrication and electrolysis conditions on the product selectivity of Ag electrocatalysts using a machine learning (ML) approach. Specifically, we explored the experimental conditions for obtaining the desired H2/CO mixture ratio with high CO efficiency. Notably, unlike previous ML‐based studies, we used experimental results as training data. This ML‐based approach allowed us to quantitatively assess the effect of experimental parameters on these targets with a reduced number of experimental trials (only 56 experiments). An inverse analysis based on the ML model suggested the optimal experimental conditions for achieving the desired characteristics of the electrolysis system, with the proposed conditions experimentally validated. This study constitutes the first demonstration of optimal experimental conditions for electrochemical CO2 reduction with desired characteristics using the experimental results as training data. Developed a method to find the promising conditions for achieving the desired CO2RR product (high FECO and ξ) assisted by machine learning (ML). The predictive model was trained using multiple experimental parameters of the catalyst ink composition and electrolysis conditions. Experimental validations demonstrated the effectiveness of the proposed method in accelerating product optimization, even when using a small dataset.