Machine learning–guided CO2 methanation: From catalyst design optimization to techno-economic and life cycle assessment analyses

The increasing share of variable renewable energy sources has intensified the need for scalable, cost-effective, and environmentally benign energy carriers. In this study, a comprehensive data-driven framework is developed to guide CO2 hydrogenation toward synthetic methane production. The developed framework integrates machine learning (ML)–based catalyst informatics, multi-objective optimization, process simulation, techno-economic analysis (TEA), and life cycle assessment (LCA). A gradient boosting regressor accurately predicts CO2 conversion and product selectivity with R2 ≥ 0.85. Feature importance analysis shows that reaction temperature is the dominant factor, followed by active metal loading for CO2 conversion and pore volume for CH4 selectivity. Multi-objective optimization identifies catalyst–condition domains capable of achieving 90–99 % CO2 conversion, 93–99 % CH4 selectivity, and ≤6 % CO selectivity, while minimizing metal usage. These predictions are externally validated against independent experimental datasets, demonstrating the strong generalizability of the selected model to previously untested catalyst systems. When ML-optimized conditions are embedded into Aspen Plus® simulations, single-stage methanation at 315 °C, 2 MPa, and GHSV ≈6000 mL/gcat·h yields net-negative carbon footprints (−0.67 kg CO2 eq/kg CH4) using hydrogen and either pure CO2 or flue gas. Although the estimated CH4 break-even prices (4.39–4.93 USD/kg) exceed current market levels, an endpoint LCA reveals up to 52 % lower total environmental damage than fossil-derived CH4. A comparative analysis of alternative CO2 utilization pathways underscores critical trade-offs in hydrogen efficiency, process complexity, and end-use compatibility. This study provides a robust, integrated framework for advancing carbon-negative CO2 valorization by combining informatics, simulation, experimental validation, and sustainability assessment in power-to-X systems.