Upscaling for improved oil recovery

Upscaling for the simulation of IOR processes is a big challenge, for several reasons. Firstly, the geological model may be complex, in order to resolve important structures, such as channels or shales, and to target remaining oil. Secondly, the recovery processes also tend to be complex, involving two-phase or three-phase flow. Careful flow simulation is therefore required to estimate oil recovery. Many upscaling methods have been developed to reduce the number of cells in a geological model, so that the models can be used for flow simulation. Inevitably, coarse-scale simulations are less accurate, especially in highly heterogeneous models. In the early stages of development of a field, this may not be important, because the reservoir model is highly uncertain. Later in the life of a reservoir, however, it is important to maintain the accuracy of flow simulation, especially when trying to predict the additional oil produced by IOR techniques. We present a method for upscaling which is suitable for use at the IOR stage of development of a reservoir. This method maintains a high accuracy compared with the fine-scale simulation, and is feasible for highly heterogeneous, multi-million cell models. We perform a single-phase pressure solution over the fine scale geological model, using well pressures and natural reservoir boundary conditions (e.g. high aquifer pressures, sealing faults), and calculate upscaled transmissibilities and well connection factors. This upscaling approach takes longer to run than conventional upscaling methods, which apply the boundary conditions locally to each coarse- scale cell. However, it has the advantage of avoiding errors due to inappropriate boundary conditions, and the effects of the main geological structures can thus be preserved. In addition, this method is quicker, more robust and more practical than dynamic two-phase upscaling. The method is demonstrated to have the capability of reproducing two-phase, and even three-phase (WAG) behaviour in the models tested.