Multi-Objective Bayesian Optimization Framework for CO₂ Injection Strategy Design: A Sleipner-Inspired Study

This study develops a multi-objective Bayesian optimization framework for the design of geological CO₂ injection strategies. The problem is formulated over a mixed decision space that includes well configuration, injection rate, and staged allocation of injection across reservoir layers. Evaluating candidate strategies requires repeated simulation of subsurface migration processes, which motivates the use of data-efficient optimization techniques. The optimization is posed with three competing objectives: maximizing retained CO₂ storage, minimizing upward migration through a leakage proxy, and minimizing total project cost. To handle the mixed categorical-continuous decision variables and non-smooth response behavior, a multinomial logit surrogate model is employed. Within this framework, several acquisition strategies are investigated, including expected hypervolume improvement, scalarized upper confidence bounds, Thompson sampling, and expected preference improvement. Optimization performance is assessed using hypervolume-based metrics, convergence behavior, and characteristics of the resulting Pareto sets. Testing on the Sleipner field in the North Sea, the proposed framework demonstrates its usefulness for analyzing trade-offs in CO₂ storage design. The study relies on a simulator for fluid migration in a multi-layer storage system, implemented through an integrated R-Julia workflow. Results show that the proposed method identifies a variety of high-quality trade-off solutions within a small evaluation budget. The scalarized upper confidence bounds method performs best among the tested strategies, producing the largest and most evenly spaced Pareto front. The analysis shows that vertical injection allocation is the most important factor in controlling storage performance and behavior under full-capacity conditions.