Design and Analysis of Metal Oxides for CO₂ Reduction Using Machine Learning, Transfer Learning, and Bayesian...

Area of Science:

• Chemical Engineering
• Materials Science
• Computational Chemistry

Background:

• Resource recycling is crucial for sustainable chemical production and disposal.
• The reverse water gas shift-chemical looping (RWGS-CL) reaction offers a pathway for CO2 utilization.
• Controlling thermochemical redox cycling in RWGS-CL is challenging, impacting CO2 and H2 conversion.

Purpose of the Study:

• To develop optimized metal oxides and experimental conditions for CO2 conversion to CO.
• To achieve specific CO2 and H2 conversion extents using advanced computational methods.
• To enhance the predictive accuracy of mathematical models for RWGS-CL reactions.

Main Methods:

• Machine learning and Bayesian optimization were employed to screen metal oxides and optimize reaction parameters.
• Transfer learning was utilized to improve model prediction by integrating oxygen vacancy formation energy data.
• Random forest regression analysis was performed to understand variable importance and prediction accuracy.

Main Results:

• Successfully developed metal oxides and optimized conditions for targeted CO2 and H2 conversion.
• Enhanced mathematical model accuracy through transfer learning and incorporation of material properties.
• Identified key variables influencing RWGS-CL reaction performance via feature importance analysis.

Conclusions:

• Machine learning and Bayesian optimization are effective tools for optimizing complex chemical processes like RWGS-CL.
• Transfer learning significantly improves predictive capabilities for material-driven reactions.
• The study provides a framework for designing efficient CO2 conversion systems for resource recycling.