Abstract
Current reservoir characterization technologies allow very detailed description of the reservoir. Unfortunately, most of the time, these detailed descriptions are lost in the numerical flow simulation process because the models are required to be rescaled. This is due to the limitation of current flow simulation algorithms which could not handle very large models.
This paper proposes a multiscale numerical method that will efficiently solve flow problems of very large, multi-segmented reservoirs. This method contains three-levels of coincidental computational meshes that decouple the diffusive and convective flow problem in both temporal and spatial domains. We implicitly solve for the diffusive-pressure field in the block-scale mesh and then interpolate the solution to the coarse-scale mesh, and then finally to the fine-scale mesh. The convective-saturation solution is solved explicitly on the fine-scale mesh. The saturations are then rescaled back to the coarse-scale meshes and the block-scale mesh pressure field are re-solved.
This algorithm will preserve the flow character of the geocellular models as it minimizes the rescaling requirements. The algorithm is amenable to parallel computing and because of the non-iterative nature of the de-coupled systems of equation, will also allow better scalability for parallel computing environments.
