Abstract: |
The limitations of variogram-based simulation programs to model complex, yet fairly common, geological elements, e.g. sinuous channels, have been recognized for a long time. In applications involving flow simulation, such as hydrocarbon reservoir modeling, variogram-based methods typically generate high entropy models that very often misrepresent the heterogeneity and connectivity of the actual field under study, and thus may provide incorrect flow performance predictions. Several solutions have been proposed to go beyond two-point statistics to be able to reproduce geologically realistic patterns. Because limited sample data makes the inference of multiple-point statistics extremely challenging, the idea of using an external source, namely a training image that describes the expected subsurface geology, has been a very attractive solution. However, important issues need be addressed to make that idea practical.
The first fundamental question is: where to find training images, or how to generate them? This paper explains why unconditional object-based modeling is a simple but comprehensive approach to build training images. Because training images are conceptual, objects do not need to be conditioned to actual data, thus complex facies geobodies as well as complex interactions among facies can be implemented and simulated without dealing with traditional data conditioning limitations of object-based modeling. The selection of appropriate training images and their consistency with actual data are also discussed in this paper. The next question is about handling non-stationarity: how to impose spatial variations of facies proportions or facies geobody geometries? Considering the training image as a (stationary) collection of patterns, and then imposing external constraints such as variable azimuth fields and facies probability cubes, has appeared to be a very effective workflow. Finally, an overview of tools, lessons learned and best practices to make multiple-point statistics simulation time-efficient is provided, and illustrated with various reservoir modeling case studies. |
