8.2 Can we take production data into account ?
Some engineering data are easy to use in building a geomodel. SCAL measurements are the basis for defining transition zones and relationships between the distance to the contact, porosity, water saturation and facies (Adams, 2016) while well tests and kh estimates help guiding the 3D permeability modeling.
Two types of data are much more difficult to integrate: production data and pressure data. Both are essential for the team to be able to make decisions about the development of the field. But they are very challenging because each measurement is the consequence of the complex relationship of rock and fluid properties. On the static side, captured by the geomodel, elements like fault compartiments, the units stratigraphy, facies geobodies, facies proportions, porosity and permeability ranges can all have an impact. On the dynamic side, the initial water saturation in the reservoir (also captured in the geomodel), the water saturation before and at the time of the measurement, the relative permeability models, the different fluid viscosity among other parameters will influence production and pressure data.
While a measurement a neutron porosity at a certain depth along a well can be easily related to the total porosity of the rock at that location, production and pressure data are a complex reverse engineering problem.
Assume some input parameters will generate a certain response from the system due to the physical processes at play. The goal of reverse engineering is to validate values for the input parameters that would explain a response from the system that have been measured. In the context of this chapter, pressure and production data are such responses.
The problem of reverse engineering is that the solution is rarely unique and multiple combination of values for the input parameters can reproduce the measured response.
Figure 1 gives a simple example of reverse engineering. We use arithmetic average to compute an average porosity value in the large cell coming from merging three cells A, B and C together. Computing the average porosity out of the known input porosity gives one unique results: (18%+22%+20%)/3=20%. But what about the reverse problem? Knowing that a known average porosity is equal to 20% and knowing it was obtained by arithmetic average, what is the porosity in each of the input cells A, B and C? As illustrated by Figure 1, there is an infinite number of solutions. Porosity could be equal to 20% in all cells, or to (15%/20%/25%) in the cells A/B/C. But it could be the same three values but distributed differently among the three cells. At last, it could even be some extreme set of values like (60%/0%/0%). There is no way to know. The only thing that can be done is to at least restrict the field of possible solutions to what is physically acceptable for the dataset. While the values (60%/0%/0%) give an average of 20% porosity, how often do we find rocks with 60% of porosity? If the cells are all known to represent sands with porosity between 10 and 30 for example, then an additional rule could be imposed: all three values must be between 10 and 30.
That example not only illustrate the challenge to find the input values, but it also shows that geologists have a role to play. The additional rule on porosity to be between 10% and 30% come from the understanding of the facies. Someone with less knowledge about the geology could have considered the solution (60%/0%/0%) valid as it’s mathematically valid. This is but a simple illustration of a problem often encountered in history matching: engineers can be tempted to edit the geomodel in ways that do improve the match, but sometimes without realizing that the edited geomodel is not representative anymore of the geology of the reservoir. In history matching, it’s important for the geomodeler and the geologist to be kept in the loop.
Figure 2 illustrates the challenge of connecting production data to the size and location of a sand geobody. In this example, the evolution of the cumulated oil production through time is known. The production climbed steadily until it plateaued. This well can’t produce anymore through primary production. The well crosses a sand geobody and we know the oil is coming from that geobody. We also assume that only the size of the geobody influences how much oil was produced so far. Of course, in a real situation, such cumulative oil production would be the result of the multiple rock and fluid characteristics. Figure 2 illustrates the challenge of connecting production data to the size and location of a sand geobody. In this example, the evolution of the cumulated oil production through time is known. The production climbed steadily until it plateaued. This well can’t produce anymore through primary production. The well crosses a sand geobody and we know the oil is coming from that geobody. We also assume that only the size of the geobody influences how much oil was produced so far. Of course, in a real situation, such cumulative oil production would be the result of the multiple rock and fluid characteristics.
What can be said about the size of that sand geobody based on the cumulative oil production? If we are building the geomodel at the beginning of the production of the well, we would only have a very small cumulative production. That total could come from any size of geobody. The geobody could be a very small sand bar, or an extended sand geobody. We would have no way to narrow it down. But if we are building the geomodel around the end of the life of that producer, then we can narrow down a little. For example, one might say a geobody as small as seen on Figure 3A is not possible. At minimum, we need to have a geobody big enough to allow for that production to have occurred. But apart from that, everything is still possible. Figure 3B could be a geobody just big enough for that total production and the production stopped simply because that geobody was completely drained. Figure 3C is a much larger geobody. In that case, we could make the hypothesis that this well as finished draining the volume of the geobody that it can drained. If we want to get more oil out of that geobody, we need to drill a second vertical well. This proposal could be part of our development plan for this reservoir. But where shall we drill that new well? Indeed, even if Figure 3C is a valid geobody, are we sure of its spatial distribution? Figure 3D and E shows the same geobody than on Figure 3C but shifted to the left or the right. If the reservoir is like Figure 3D, then we need to drill the new well to the right of the existing well… but if the reservoir is like Figure 3E, we need to drill to the left of the existing well. And we still can’t tell if the reservoir has even that large shape. It might be just of a size like in Figure 3B and in that case, any new well would only penetrate the surrounding shale. Again, you have here illustrated the complexity to tie a measure of production to a specific, unique rock characteristic. At best, such data can sometimes help us discarding extreme un-realistic scenarios.
Figure 4 shows another example of extreme unacceptable geomodel which can be discarded. In this example, the reservoir is undergoing waterflooding (Figure 4A). Since water started to be injected on well A, the oil production on well B has increased. Both wells penetrate some sand which are filled with oil (oil and water in well A). The fact the waterflooding is a success implies the two wells penetrate the same sand geobody. From there, a geomodel in which we see two geobodies (Figure 4B) instead of a single geobody crossed by both wells (Figure 4C) is unacceptable and should be improved by the geomodeler.
Pressure data can also help excluding extreme non-realistic geomodeling results. If multiple wells have similar pressure data, then either they belong to the same compartiment or the same geobodies or they belong to separate block sbut which happen to share the same initial pressures. If wells have different non-correlated pressure data, then they are in different blocks/geobodies.
In the schematic reservoir of Figure 5, two faults have been interpreted on seismic but the branching is such that it doesn’t isolate a block per se. Let’s assume that all the wells A, B, C and D have initial pressure data. And let’s assume that wells C and D have similar pressures which are significantly different from the pressures in A and in B respectively. This leads to three conclusions. Firstly, it validates that well A belongs to a block isolated from the other wells by the Main Fault. Secondly, it shows that the current extension of the Branching Fault is problematic. If well B was part of the same block than wells C and D, then well B would have a initial pressure similar to the one in the two other wells. But it’s not the case. The most logical, geological solution is to extend the Branching Fault North until it completely isolates the well B into a small block delimited by the Main Fault and the Branching Fault (Figure 6A). Thirdly, the faults are sealed, meaning the simulation engineer will need to set their transmissibility to zero.
Even after the decision has been made to extend the Branching Fault to the North, the geomodeler must still be careful on how he takes it into account while building the faulted 3D-grid. The geomodeler must make a sealed fault network, meaning that the 3D-grid must indeed be cut all the way at the point of branching between the faults. While easy to check for models with only a couple of faults, mistakes can be more difficult to spot in large reservoirs with multiple faults. It’s not uncommon for geomodels to have mistakes such as found in Figure 6B. The team decided to branch the faults, but due to mistake in the geomodelling process, some small opening remains. This would be enough for fluids to leak from one block to the other during flow simulation (especially for gas reservoirs). Geomodelers must make sure there is no leak in their fault network (Figure 6C).
Similar situations can be found in reservoirs like Figure 7 in which wells seem to belong to multiple clearly identified geobodies. In this example, it’s initially assumed that there are three fluvial channels in this reservoir. Their initial geological interpretation (Figure 7A) places the wells B and C in the same channel. But it contradicts the initial pressure data: wells A and C have similar pressure while the pressure in B is significantly different. It forces the geologist to change the interpretation to place the wells A and C in the same channel (Figure 7B).