Reaching machine learning leverage to advance performance of electrocatalytic CO₂ conversion in non-aqueous deep...

Area of Science:

• Electrochemistry
• Materials Science
• Computational Chemistry

Background:

• Deep eutectic electrolytes (DEEs) offer tunable buffer capacities for advanced electrochemical applications.
• Optimizing the carbon dioxide reduction reaction (CO2RR) is crucial for sustainable energy technologies.
• Minimizing competing hydrogen evolution reactions (HER) is key to enhancing CO2RR selectivity.

Purpose of the Study:

• To apply machine learning algorithms for analyzing CO2RR in DEEs.
• To understand the non-adiabatic nature of the CO2RR process.
• To identify optimal conditions for dominant CO2RR over HER.

Main Methods:

• Utilized machine learning algorithms (ensemble, k-Nearest Neighbors) to analyze CO2RR data.
• Performed microkinetic analysis to determine reaction rates and constants.
• Employed diagnostic and SHAP analyses to evaluate model performance and feature importance.

Main Results:

• Achieved over 99% prediction accuracy for CO2RR using machine learning models.
• Identified the CO2RR process as non-adiabatic with specific time constants.
• Gradient boost ensemble algorithm predicted high asymptotic current densities and turnover frequencies (TOF) on silver electrodes.

Conclusions:

• Machine learning models accurately predict CO2RR performance in DEEs.
• Findings provide insights into optimizing non-aqueous electrolytes for efficient CO2RR.
• The study facilitates convenient TOF measurements at industrially relevant current densities.