Impacts of catalyst and process parameters on Ni-catalyzed methane dry reforming via interpretable machine learning

Dry reforming of methane (DRM) is a promising technology for valorizing two potent greenhouse gases, namely CO2 and CH4. Although design principles for active and stable catalysts are well-established, predicting selectivity of DRM over competing side reactions that alter product H2/CO ratio remains challenging. Here, we curate a set of 1638 data points from published literature and train tree regressor models to predict H2/CO ratio as a function of 11 catalyst and process parameters. The CatBoost regressors achieved best prediction performance with R2 = 0.91. Feature importance analysis reveals, in addition to well-known effects of process parameters on DRM performance, the potential roles of Ni particle size and loading in tuning H2/CO ratio independently of reactant conversions. Our exploratory study highlights ability of data-driven ML models to unearth structure-property relationships in heterogeneous catalysis by isolating effects of individual design parameters in a manner that would be difficult to achieve experimentally.