Dynamic Refinement - An Alternative to Upscaling

David Hicks (david.hicks@cmgl.ca), Account Manager, European Region for Computer Modelling Group Ltd (http://www.cmgl.ca) discusses the benefits of dynamic grid refinement and its implementation in CMG's compositional simulators.  For a fuller account the reader is referred to SPE 79683 by Peter Sammon of CMG presented at the SPE Reservoir Simulation Symposium in Houston, Texas, U.S.A., 3–5 February 2003. Peter is development coordinator for the GEM simulator.  His main research interests are simulation grids and solution methods.

Introduction
Many IOR processes are currently simulated only as small generic sections, or single well models, due to limitations in current hardware and simulation technology. Larger scale models cannot be run with the required accuracy as upscaling is not always straightforward, or even possible. This tends to make IOR simulation models more of a theoretical backup to the relevance of the process rather than an assistance in overall field implementation and optimisation planning. In this article I would like to introduce an old concept, which has now been implemented successfully and efficiently in both EoS and a K value based simulators.

The Problem
Many IOR processes have interface regions that are relatively thin when compared to the typical cell sizes used at a field or pilot scale. For example, a combustion front may exist over a front thickness of only a few centimetres, whereas the block size required for the field may be tens, or hundreds, of metres. Up-scaling of this and similar processes can prove difficult, if not impossible, to do accurately. Thus, there are obvious problems in properly representing interfaces in such models. Unfortunately, when the physical interactions at the interface are not well represented, the production estimates generated by the simulation can easily over or under estimate true recovery.

Another difficulty associated with modelling interface-driven processes is how to up-scale rock properties. It is possible that the presence of an interface could affect the choice of the up-scaling procedure that should be used. The properties of the interface, such as whether it is predominantly horizontal, or vertical, could introduce an important variable into up-scaling. This process can be complicated enough without introducing further dimensions to the problem.

So we would like to avoid up-scaling, but to do so we would have to use centimetre to metre scale grid blocks. This would result in impossibly large models and impractical simulation runtimes. A method which removed the need for up-scaling, but could be implemented at a field scale, would allow engineers to both accurately represent their process and apply it to overall field development planning.

The Solution
CMG, with funding and assistance from the TotalFinaElf research centres in Pau and London, has developed a 3D dynamic gridding capability for use in its advanced process modelling simulators GEM and STARS. These simulators allow a high resolution grid to track any interface area as it moves around the model. Gridblocks are amalgamated and refined based on a set of defined triggering mechanisms allowing high resolution grids where they are needed and coarser grids where they are not. Figure 1 shows an interface being tracked by the fine grid structure.

Figure 1: Snapshot of moving front (Click image for larger view)

The simulation speed gains that are observed in applying this technique have been seen to scale almost exactly with the number of gridcells being simulated. This is due to the sparse solver technology that is used throughout the simulator, and is the main reason why the dynamic gridding implementation was successful. This allows the formation and removal of connections in the reservoir to create virtually no overhead to the simulator, making the process incredibly efficient.

Thus, dynamic grid refinement offers a possible solution to many accuracy and model size issues in IOR simulation.

Figure 2 is a series of images showing the process operating dynamically in the simulator.

Figure 2: Dynamic grid (images taken from movie sequence) (Click on an image for a larger view)