Development of Catalysts and Machine Learning Applications for Direct Conversion of CO2 to Hydrocarbons

Abstract

As global efforts to mitigate climate change and achieve carbon neutrality accelerate, increasing attention is being directed toward technologies that convert carbon dioxide (CO₂) into valuable fuels and chemicals. In particular, the aviation industry has shown significant interest in the production of sustainable aviation fuel (SAF) as a strategy to reduce dependence on fossil fuels. Among the various CO₂ conversion pathways, catalytic hydrogenation of CO₂ into liquid fuels or heavy hydrocarbons is theoretically promising yet still faces substantial technical challenges for practical implementation. Conventional catalyst-based CO₂ conversion technologies often suffer from high reaction temperatures, low selectivity, and limited catalyst stability, making it difficult to achieve consistent production of target products. Moreover, identifying optimal metal compositions is time-consuming and resource-intensive due to the complex and nonlinear interactions among multiple catalyst components. These challenges highlight the need for a data-driven approach that can efficiently explore compositional space and systematically analyze relationships between catalyst features and performance. In this study, Na-promoted zinc ferrite catalysts were synthesized using co-precipitation and incipient wetness impregnation. These catalysts had variations in Na, Zn, and Fe contents and calcination temperatures. The catalysts were characterized by experimental analyses. Their catalytic performance was evaluated in a 20 bar, 340 °C syngas reaction using a custom-designed 8-channel reactor system. Machine learning (ML) models were constructed by adopting the categorical boosting (CatBoost) algorithm. Hyperparameters were optimized through grid search and performance was validated using leave-one- out cross-validation (LOOCV). The models reasonably predicted several target properties such as Fe Time Yield for C₅–C₁₆ hydrocarbons (FTYC5-C16), naphtha selectivity, and SAF selectivity predictions. To identify optimal catalyst compositions, three optimization algorithms, artificial bee colony (ABC), equilibrium optimizer (EO), and Bayesian optimization (BO), were implemented. Overall, this ML-guided approach efficiently explored compositional space, identified high-performing catalysts, and demonstrated potential for accelerated design of CO₂ hydrogenation catalysts for SAF and naphtha production.