GDF's gas storage in aquifer at Soings-en-Sologne comprises two unconnected domes (285 Mm3(n) at north top, 370 Mm3(n) at south top) and is located at 1000 m/ss. Many difficulties have arisen since the first gas filling in 1981: at least 55 Mm3(n) lost at north top, lack of knowledge about the whereabouts of gas bubble boundaries, water production being too large (water content 5.3 g/cm3), and neutron logs showing gas traces in the caprock.
The 25m thick reservoir includes a 15m thick intermediate caprock and is located in a Keuper's faulted fluvial brained depositional system. The geological model takes into account the logging data at 117 wells of the Chemery-Soings gas storage complex, as well as the results of a 3D seismic survey operated on the north top. This model, though deterministic, takes advantage of a geostatistical study. GDF's in house diphasic flow simulator is used for this purpose. The reservoir model defines 5 Local Grid Refinements (LGR). The maximum grid extension is 153.6x153.6 km2 (LGR1). The whole diphasic volume is included in the 48616 cells of LGR5 (88x38 over 6.6x2.85 km2 and 14 layers). Only 33253 of these cells are active. Traditionally, faults modeling is realised either by applying correcting coefficients to horizontal links (Fig. 1A) or by vertically shifting horizontal cell links through the fault plane (Fig. 1B) or by using a comer point method. All these methods, except the last one, were tested within the framework of this study, without success. Looking at a finite volume geometry conveys the wrong idea that cells belonging to a same layer are disconnected. To visualize correctly the reality, it is necessary to imagine flow corridors, flow lines and transfer surfaces. The schematic view reflects the reality only at the layer contacts along the fault. Our method aims to take into account the geometrical discontinuity without replacing a folded structure by another. Transfer surfaces which are perpendicular to flow lines by definition are always vertical at the fault. The number of horizontal links starting from a cell is equal to the number of opposite cells through the fault (Fig. 1C). As a consequence, the amount of fluid transfered between a cell and its opposite neighbours is directly proportional to their common transfer surfaces. Note also that a 'multi-link' option is required in the simulator to implement this method. Firstly, the reservoir simulator deletes initial horizontal links at cells located on the fault segments described by the user. At each of these cells, several neighbouring cells are found and connected, and therefore new links are generated. This method is equivalent to an automatic local vertical regridding and guarantees a geometrical term validity in the transmissibility calculation. Therefore, flow matching depends on permeability only.
Sedimentation gaps are the second main cause of reservoir heterogeneity. Modeling this type of discontinuity implies two kinds of dead cells: non-zero thickness for clay cells and zero thickness for sedimentation gaps. An automatic calculation procedure allows the connection of cells belonging to layers to be separated by sedimentation gaps. Accordingly, this device avoids local artificial over-pressure and excessive spread of gas bubble. The feature is therefore completely transparent for the user. A pressure and saturation history-matching over 15 years of operation was performed using both classical and new features. Fig. 2 shows a saturation map generated by a classical modeling method. The gas bubbles are abnormally connected. Fig. 3 shows a saturation map corresponding at the same date and generated by these particular methods: gas bubbles are disconnected and the pressure fit is correct. Lost gas in the north top was localized in a minor top and its volume computed (80 Mm3(n)). P. 315^