Various geostatistical models have been proposed to generate possible descriptions of the internal structure of heterogeneous reservoirs. For actual reservoir engineering studies, there are practical problems in dealing with large numbers of possible images. This paper proposes an efficient way of sorting the images, accounting for dynamic criteria, allowing a few images to be selected for detailed study.
The method consists in performing simplified reservoir simulations in which the pressure field is held constant in time, rather than being updated at each time step. The pressure field used is calculated by assuming a steady state flow within the reservoir but using the actual boundary conditions (well rates or pressures). The saturation evolution takes into account the multiphase parameters. In this paper, it is shown that these simplified simulations are sufficient to sort the images for a model problem. Here, the criteria used are the breakthrough times and the oil productions.
The advantage of this approach is that significantly less computer time is needed to perform the simplified simulations than the full simulations. This makes it possible to run simplified simulations for many equiprobable images in order to find the extreme behaviors. Some images are then selected for complete simulations. In addition, a working methodology is proposed to determine the number of geostatistical descriptions to be considered.
Using the information collected at the wells and the geological knowledge of the depositional environment, geostatistical simulation models generate representations of the internal structure of oil reservoirs. Owing to the sparse information available on the horizontal extension of the geologic bodies, a geostatistical model provides a wide variety of possible interpolations of well-to-well lithofacies. The making of production forecasts based on reservoir images raises two essential problems:
To go from a geological representation to adiscretization of the reservoir in cells for the fluid flow simulator.
To quantify accurately the uncertainty in the production forecasts based on the differentpossible reservoir geometries.
The first problem is usually a problem of information reduction. A practical method is proposed here to deal with the second problem, which is essential for the practical use of geostatistical models.
Production forecasting demands large-scale calculations to simulate the hydrodynamic behavior of reservoirs.