Multi-objective Bayesian Optimization of CO₂ Storage Inspired by Sleipner Project

We present a multi-objective Bayesian optimization framework to maximize CO₂ storage and minimize leakage during migration. The framework considers two conflicting objectives: maximizing CO₂ storage and minimizing leakage during migration via invasion percolation. Due to the presence of discrete inputs, we employ a multinomial logit model. Using the Pareto front, we compute the expected hypervolume improvement (EHVI) and evaluate other acquisition functions, including scalarized confidence bound (SCAL), Thompson sampling (TS), and expected preference improvement (EPI). The approaches are tested and compared on a 5-layer example and the 2019 Sleipner CO₂ storage project in the North Sea.