Hybrid machine learning and computational fluid dynamics framework for optimizing supercritical CO₂ pipeline networks in large-scale carbon capture and storage

This study presents an integrated framework combining computational simulations and machine learning (ML) to design and optimize supercritical carbon dioxide (CO₂) pipeline networks for large-scale carbon capture and storage (CCS) applications. Using ANSYS fluent, the fluid dynamics of supercritical CO₂ transport were analyzed under various operational conditions to generate synthetic insights and complement real-world datasets collected from CO₂ sequestration activities and industrial sources. Key variables such as pressure, temperature, and pipeline diameter were extracted, engineered into time-lagged and rolling features, and used to train a random forest regressor model. Feature importance analysis revealed that short-term flow trends, volatility, and geographic segmentation are dominant predictors of CO₂ flow behavior. The ML model achieved high performance, with an R² score of 0.9897 and a low RMSE of 0.0335, indicating strong predictive reliability. Hyperparameter tuning further optimized model accuracy, while visual analyses including actual vs. predicted comparisons, residual distributions, and feature interaction heatmaps validated the model’s robustness. The results demonstrate that data-driven approaches can complement simulation-based methods to improve the design, monitoring, and operational efficiency of CO₂ transport systems. This work provides a scalable, intelligent solution for CCS pipeline optimization, supporting the global transition toward low-carbon infrastructure.