olwijn
commented
3 years ago The figure shows the first result of a scheme that maps the ECLIPSE model cell volumes to Flownet cells based on specified depth levels (which could represent fluid contacts). Pore volume mapping is done such that no volumes are transferred across these depth levels. The total volume accumulated in a flow tube section between 2 depth levels is then evenly distributed over the cells in that tube section, again preventing transfer across the depths levels.
The figure shows that the volume distribution is not fully symmetric. This is most likley related to the KDTree algorithm which makes an arbitrary choice for identifying a nearest point if multiple points are at equal distance.
The default value for EQLOPT3 is -5, which results in an average grid block saturation based on a subdivision of 10 sub-blocks. Presumably a varying number of sub-blocks will lie below or above the contact (this must be related to the thickness of the FlowNet grid block and therefore also of the sub-blocks), leading to a vertical gradient of average saturations.
The equilibration with EQLOPT3 set to 0 is based on gridck block midpoints only and results in a clear OWC in FlowNet, consistent with the Eclipse model, resulting the following initial oil saturation:
The FOIIP still does match the Eclipse value however:
The initial saturations (So, Sw) above the contact are (0.9, 0.1) in both models. Below the contact they are (0.1, 0.9) in Eclipse and (0,1) in FlowNet. The only difference in SWOF is that in FlowNet the table extends to an Sw value of 1 (Kro is 0 for Sw larger than 0.9). If this is adjusted (manually) in the FlowNet deck the FWIIP and FOIIP match better:
A different result is obtained when EQLOPT3 is set to -1 instead of 0:
In summary, to achieve a clear OWC in FlowNet EQUIL parameter EQLOPT3 (9th entry) should be set to 0 instaed of 1*. Extending the SWOF table to Sw=1 results in the absence of oil below the contact, i.e. So=0, also when Sorw > 0.
Pore volume from the original reservoir structure is assigned ('mapped') to the nearest flow tube cells, and then distributed equally over all cells in the tube. We do not make a distinction between vertical and lateral distance, resulting in attribution of water volumes to cells in the oil zone. Also the redistribution of cell volumes over the tube may effectively result in the transfer of volume across the oil-water contact (OWC). The outcome will generally be an overestimation of oil volume (and, probably less importantly, to an underestimation of water volume).
In the figure below, the rectangular grid represents the grid of an Eclipse model, while the yellow grid represents the flow network. (The figure is not to scale; the lateral width of Eclipse grid cells (DX, DY) is typically a factor 10 or more larger than the cell thickness DZ). The mostly water-containing pore volumes located below the OWC (white dashed line) represented by the blue grid cells will all be assigned to the flow tube cells (yellow) that are mostly, or in this case entirely, located above the OWC. After equilibration the flownet cells will therefore contain mostly oil such that water volume has effectively been converted into oil volume.
A first possible treatment would be to adopt a layer-based approach for design of the network that would ensure the placement of nodes in all layers (e.g. the additional orange nodes in the figure below, which would lead to creation of more flow tubes (light orange grid cells)). While this is currently supported, no vertical connection will be created between nodes in different layers (this functionality is used for disconnected reservoir zones). This could be modified to support connections between internal layers representing (possibly weakly) connected reservoir zones with different properties.
A second, conceptually still simple, approach for addressing the volume distribution could be to use OWC information (provided in the config file or extracted from the data) to limit or prevent transfer of reservoir volumes located below the OWC to flow tube cells above the OWC and vice versa.
A combination of the 2 approaches may be necessary, i.e. ensure an appropriate distribution of nodes below the contact, and also prevent the mapping of volumes across the contact.
A question is how to practically deal with flow tubes that connect nodes on either side of the OWC (see figure below).
We could assign volumes below the OWC only those cells of each flow tube that lie below the contact (and same for volumes/cells above the contact). In the figure, the only flownet cells below the OWC are the 3 yellow cells furthest to the bottom left, which would then get larger volumes than the cells located above the OWC. In this particular case their combined volume would correspond to the entire volume below the OWC in the original model. To avoid this, a cutoff in lateral distance (the green dashed line) could be applied (i.e. localization) such that only the nearest volumes will be assigned to any flow tube cells, while the volume located further away (the water zone on the right of the green vertical line), would not be part of the Flownet.
However, a complicating factor is the redistribution of cell volumes over a flow tube. Redistribution of cell volumes is necessary because not all flow tube cells may directly receive any volume. Strong contrasts in neighbouring cell volumes may also lead to instable numerics (not checked). For flow tubes that connect nodes on either side of the contact this redistrubution may also be a cause for transfer of pore volume across the contact.