Multi-objective optimization of cushion gas design in underground gas storage using machine learning

• A multi-objective optimization workflow for replacing cushion gas with inert gases was presented. • The simultaneous increase in carbon sequestration and gas recovery efficiency was examined as the process optimization goals. • The role of machine learning methods in significantly improving the optimization speed was notable. • Effective integration of underground natural gas storage operations with carbon dioxide sequestration can achieve environmental objectives. Underground natural gas storage (UGS) has played a vital role in ensuring the energy supply chain stability in various regions worldwide over the past decades. The increasing trend in energy and natural gas consumption globally necessitates further development of UGS projects. Depleted oil reservoirs, due to their limited suitable storage environments in some areas and proximity to major consumption centers, can be utilized to develop natural gas storage further. In this study, a UGS project in an oil reservoir is optimized by designing it in a manner that facilitates the sequestration of CO₂ gas through the replacement of a portion of the cushion gas with inert gases. To address the computational cost of dynamic simulation, a proxy model was developed using three regression algorithms: SVR, XGBoost, and MLNN. Among them, the MLNN model, with three hidden layers, achieved the best performance (R² = 0.9886 for training and 0.9562 for testing), and reduced MAE by over 40 % compared to SVR. This model enabled rapid multi-objective optimization using the MOPSO algorithm. Several Pareto fronts were generated, and the best front provided 500 optimal solutions after 300 iterations. The proposed framework significantly reduces optimization time—by over 90 % compared to conventional simulation—while capturing the complex trade-off between maximizing gas recovery and CO₂ sequestration. This research is among the first to integrate intelligent proxy modeling with multi-objective optimization for cushion gas design in UGS systems, offering a novel path toward environmentally responsible storage strategies.