Accurate Resolution of Near-Well Effects in Upscaled Models Using Flow-Based Unstructured Local Grid Refinement

We present a new approach for representing wells in coarse-scale reservoir simulation models. The technique is based on an expanded well model concept which provides a systematic procedure for the construction of the near-well grid. The method proceeds by first defining an underlying fine-scale model, in which the well and any key near-well features such as hydraulic fractures are fully resolved using an unstructured grid. In the (coarse) simulation model, the geometry of the grid in the expanded well region, and the associated radial transmissibilities, are determined from the solution of a fine-scale, single-phase, well-driven flow problem. The coarse-scale transmissibilities outside of the well region are computed using existing local upscaling techniques or by applying a new global upscaling procedure. Thus, through use of near-well flow-based gridding and generalized local grid refinement, this methodology efficiently incorporates the advantages of highly-resolved unstructured grid representations of wells into coarse models. The overall model provided by this technique is compatible with any reservoir simulator that allows general unstructured cell-to-cell connections (model capabilities, in terms of flow physics, are defined by the simulator). The expanded well modeling approach is applied to challenging 3D problems involving injection and production in a low-permeability heterogeneous reservoir, tight-gas production by a hydraulically-fractured well, and production in a gas-condensate reservoir. In the first two cases, where it is possible to simulate the fine-grid unstructured model, results using the expanded well model closely match the reference solutions, while standard approaches lead to significant error. In the gas-condensate example, which involves a nine-component compositional model, the reference solution is not computed, but the solution using the expanded well model is shown to be physically reasonable while standard coarse-grid solutions show large variation under grid refinement.