Local-Global Upscaling for Compositional Subsurface Flow Simulation

Compositional flow simulation is required for the modeling of many important subsurface processes including enhanced oil recovery and geological carbon storage. These simulations can be very demanding computationally, which motivates the need for efficient alternatives to detailed fine-scale flow modeling. In this work, a local-global upscaling procedure for oil-gas compositional problems is developed. With this approach, the key upscaling computations are performed over local (rather than global) fine-scale regions. The pressure boundary conditions required for these computations are determined through an approximate global coarse-scale simulation. Gas saturation and component molar fraction boundary conditions are prescribed based on the injected fluid conditions. An 'attribute-based' approach, which enables the efficient assignment of coarse-scale relative permeabilities and compositional functions based on quickly computed (static and single-phase flow) features, is introduced. Numerical results are presented for two- and three-dimensional models (containing eight and four components, respectively) with flow driven by wells. These results demonstrate that the local-global upscaling method, with or without the attribute-based property assignment, provides significantly more accurate predictions than standard (single-phase-parameter) upscaling methods, though accuracy is not as high as that achieved using more expensive global compositional upscaling. Speedups of about an order of magnitude, relative to fine-scale simulations, are observed using the attribute-based local-global procedure. The robustness of the upscaled models to variation in well bottom-hole pressures is also assessed. These computations demonstrate that the upscaled models retain accuracy over a range of well settings, which suggests that they will be useful in computationally demanding applications such as well control optimization.