@

초록

Electrochemical CO2 reduction (CO2RR) presents a sustainable approach to converting greenhouse gases into valuable chemicals. However, achieving precise product selectivity remains challenging. This study investigates the influence of applied potential and catalyst morphology on CO2RR selectivity using a machine learning (ML) framework. Unlike most existing ML applications which primarily rely on open databases, we collect 378 experimental data points from literature focusing on Cu and Cu-derived catalysts in H-type cells. Various ML models are trained using both numerical and categorical parameters, with the artificial neural network model demonstrating the highest accuracy in predicting Faradaic efficiencies of key CO2RR products. The SHapley Additive exPlanations (SHAP) analysis identifies applied potential as the primary determinant of product distribution. Additionally, it reveals synergistic effects between catalyst morphology and size, influencing pathways toward CO, C-1, and C2+ products. These insights offer guidance for the rational design of Cu-based CO2RR systems by optimizing operational parameters to enhance multi-carbon product selectivity. The study introduces a novel dataset and a data-driven methodology that integrates experimental findings with ML techniques, advancing the mechanistic understanding of CO2RR and providing a scalable approach for catalyst discovery and optimization.

키워드