Selective Production of Hydrocarbons (C ₅₊ ) from CO ₂ Hydrogenation: Exploring Clusters and Machine Learning

Abstract

Carbon dioxide hydrogenation has emerged as a promising strategy for mitigating climate change by reducing CO2, a major greenhouse gas, into valuable chemicals and fuels. However, advances in catalyst design and process optimization are required for more efficient and selective processes to make CO2 hydrogenation economically viable. In this work, we used a hybrid of cluster analysis and machine learning (ML) approaches to identify conditions for high CO2 conversion (41%) and selectivity toward C2–C4 (32%) and C5+ (40%) hydrocarbons while minimizing CO (10%) and CH4 (15%) formation. The best-performing cluster exhibited a narrow pressure range (20–30 bar), lower reaction temperatures (300–340 °C), high surface area (200–850 m2/g), and specific metal loadings, demonstrating the most desirable catalytic properties. Feature importance analysis for this optimal cluster revealed the following order of influence: H2 reduction > Ba > Ca > mass of catalyst > CNT. These differed from the feature importance of the whole population: Fe > calcination temperature > surface area > pore diameter ≈ reduction temperature > K. These findings underscore the importance of clustering data to uncover latent trends and heterogeneities, which can guide the rational design of catalysts and process parameters for the selective production of higher-value hydrocarbons from CO2 hydrogenation. It further suggests that what drives optimal performance is not the same as what influences the whole data population.