Integrated design and optimization for a unified carbon capture and storage system using a machine-learning-assisted multi-objective optimization framework

Abstract

Efficient deployment of carbon capture and storage (CCS) systems requires the integrated design of surface facilities and subsurface storage reservoirs. Herein, we propose a unified CCS system design framework in which dynamic nodal analysis, multi-objective optimization, and machine-learning are combined to support efficient and scalable decision-making. Optimization is focused on four key design variables: the CO(2 )discharge pressure at the hub terminal, the inner diameters of the pipeline and injection tubing, and the injection temperature, with storage capacity, injection safety, and economic feasibility as the performance objectives. Dynamic nodal analysis couples surface and subsurface flow physics-specifically, compressible pipe flow dynamics for surface transport and Darcy-based porous media flow for subsurface storage-to simulate transient system behavior. A multi-objective particle swarm optimization algorithm linked to a full-physics facility-reservoir model was used to derive an initial Pareto-optimal front covering approximately 3.8% of the feasible design space. This was refined and expanded by using machine-learning-based proxies, including neural network, random forest, and boosting techniques trained on the optimization results. The framework was applied to a potential CCS assessment project in the Republic of Korea, an onshore CO(2 )terminal, a 175-km subsea pipeline, and a 1-km vertical injection well. The machine-learning models demonstrated high predictive performance (R2 > 0.96; Mean Absolute Percent Error <= 5%) and significantly reduced computational time compared to a full-physics simulation. This hybrid framework offers a practical and accurate approach for early-stage CCS system design and lays the foundation for real-world deployment under complex multi-objective constraints.