Development of Machine Learning Models to Discover and Optimize CO2RR Catalysts

Abstract

The increasing demand for environmentally sustainable solutions to carbon dioxide (CO2) emissions has led to a concerted effort to identify efficient CO2 reduction reaction (CO2RR) catalysts. Unfortunately, the experimental investigation of such catalysts can be challenging, complex, and costly. By leveraging the power of machine learning, we aim to provide a more efficient and cost-effective approach to the search for CO2RR catalysts that can contribute to the reduction of CO2 emissions.

In this study, machine learning models with K-Fold Cross validation were used on a dataset containing CO2 reduction reaction (CO2RR) catalysts to analyze their properties, and to investigate the importance of each input parameter with the predicted output (*CO Adsorption Energy). Out of 13 features, Filing, Skewness, Center of a d-band and Atomic Radius were found to be most significant and correlated with the predicted output. Multiple ML models were used from the simplest Multiple Linear Regression model to the complex Artificial Neural Network model. The lowest performance was the Linear Regression model with an accuracy of 0.894 and an RMSE of 0.17 eV on the test set, this indicates that there could be a linear correlation between the provided input features and the chosen output. The highest performance was made by the Random Forest Regression model which yielded an accuracy of 0.969 and an RMSE of 0.093 eV on the test set. The ML models have successfully predicted the *CO Adsorption Energy with high accuracy that is computationally cheap compared to DFT. This would allow cheaper and faster screening for efficient CO2RR catalysts. Based on the RFR model, A grid search was conducted to screen for possible combinations of input variables for the output *CO Adsorption Energy in the desired region.

Date of Award	14 Dec 2023
Original language	American English
Supervisor	Ali Alhammadi (Supervisor)