Development and surrogate-based calibration of a CO2 reservoir model

Simulation-optimization framework is a widely used approach for numerical model calibrations, though its primary difficulty is its high-demand of computational efforts. In this study, Bagging MARS (BMARS) adapted from Multivariate Adaptive Regression Splines (MARS) algorithm, is used to construct the surrogate of a three-dimensional CO2 reservoir model, which is developed to simulate CO2 injection and migration in a fault-compartmentalized underground reservoir. The BMARS surrogate model is then used in model calibration to estimate specified reservoir model input parameters efficiently. The results demonstrate that the BMARS model can improve fitting stability and predictive accuracy against the ordinary MARS model. Parameter sensitivity analysis, which is efficiently conducted using the BMARS model, suggest that permeability of Fault#10 and caprock dominate the pressure buildup in this fault-compartmentalized reservoir. Hence priority should be given to investment in estimating these two reservoir properties. Overpressure propagation and CO2 migration in the reservoir responding to three years of CO2 injection are also analyzed using the calibrated model. The calibrated water-CO2 flow model could be a useful tool to evaluate the future operation and risk assessment of the reservoir. The results comparison and sensitivity analysis demonstrated the proposed BMARS-based simulation-optimization framework is an efficient and accurate model calibration approach.