Вы находитесь на странице: 1из 10

V INGEPET 2005 (EXPL-4-MF-137)

THE APPLICATION OF STREAMLINE SIMULATION IN FLUVIAL MATURE FIELDS −


COLOMBIA AND ECUADOR

Mikael Frorup, O. Beltran, J. Hernandez, L. Marquez, A. Baker, W. Gaviria, Y. Del Castillo


Schlumberger

Abstract
Streamline simulation is gaining use in the region for better and for worse. The use of this approach can
be divided into two categories. It is either used for screening of geostatistical property models or for water
flood optimization. The objective of this paper is to present how streamline simulation can be used
successfully to accelerate study time and when the approach starts to fail. Examples and lessons learned
will be illustrated from studies carried out in Colombia and Ecuador.
In geostatistical modeling the properties are distributed according to well and seismic observations only.
The dynamic information (pressure and production) is not used to condition the static property model.
Streamline simulation is applied after the static model realizations have been created to run a dynamic
simulation using the historical production information. The simulation results are then checked against the
historical data and realizations that cannot support the production history are ruled out. Streamline
simulation is often applied in these cases because it can deal with large-scale grid models and requires
little or no upscaling. The observation made, is that sometimes the streamline simulation is not required to
carry out the screening. In many cases simple material balance calculations can give the same results but
in other cases it is essential. This will be illustrated with examples.
Streamlines illustrate flow paths and can consequently be used to illustrate displacements in a field scale.
Areas that are not being swept under the current injection pattern can be identified. A diagnostic
approach will be presented based on streamline simulation that allows for identification of potentially by-
passed areas. The model can as well be used to assess the impact of a better “draw-down” management
of the wells in the field, that will allow less well to well interaction. The use of streamline simulation for
water management will primarily be illustrated with generic examples for clarity.
Common to both applications of streamlines are a number of limitations. First of all gravity segregation is
difficult to represent in streamline simulation as it requires a reduction of the time steps. As a result the
streamline simulation may not be faster than the traditional finite difference solver. Secondly, to obtain
fast calculation times, the reservoirs have to be two phase oil/water systems. Introducing a gas phase
slows the simulation down as the front tracking (Buckley-Leverett) approach no longer can be used.
Recent developments allow for compositional streamline simulation, which is faster than the traditional
compositional simulation, but is slower than the front tracking algorithm.
Streamline simulation has been proven as a tool for screening of geostatistical property models and water
flood optimization. It can be used in diagnostic studies focused on problem identification and solution. In
these studies the streamline simulation has been able to reduce study time significantly.

Introduction
Streamline simulation is an alternative to the traditional finite difference solver algorithms. The approach
is based on using the pressure solution to calculate the pressure gradients and the streamlines and then
to calculate the saturations along the streamlines either by a Buckley Leverett approach or by a series of
one-dimensional simulation models. The governing equations are discretized but solved on separate
structures, pressure by grids and saturations by streamlines. As long as the flow paths do not change
dramatically in time, longer time steps can be taken so that models containing a finer resolution and more
cells can be solved in a shorter timeframe. The concept of streamline simulation was introduced in the
early 1990s [1].

Figure 1 illustrates why we are interested in streamline simulation. For the smaller models using a
streamline simulation approach may be more time consuming than solving all the equations using finite
difference. This is due to the time is takes to define the streamlines for every timestep when new
pressures are calculated. Typically, when models are in the range of hundreds of thousands of grid
blocks, then streamline simulation represents a reduction of simulation time compared to finite difference
solvers. In general in Colombia, Ecuador and Peru many reservoirs are found in fluvial or delta
depositional environments. As a result the sand distribution is often very complex with numerous reservoir
levels and numerous rock types. Often a three dimensional static reservoir model is constructed to
capture the heterogeneity in space. A sand thickness can be very small (3 to 5 ft), a high vertical
resolution is required to capture the problems. In area the model is typically larger with block lengths
somewhere in the range of 50 to 100 m. The result is often a static property model with a few million of
V INGEPET 2005 (EXPL-4-MF-137) 2
blocks. Reducing the number of blocks for the dynamic model is always an issue. However, it is limited
how much models can be scaled up. In Figure 2, the issue of upscaling is illustrated. This comes from a
study of a fluvial reservoir. The reservoir units were identified in the property model and ranked according
to size. The plot shows the cumulative oil in place in the largest, the two largest reservoirs and so on. The
model was then upscaled by merging grid blocks and average properties to reduce the number of cells.
With this upscaling the reservoir size distribution changed. The upscaling is associated with a smearing of
the model, a loss of resolution. There is therefore, a desire to use the fine grids for numerical simulation.

nce
CPU time

re
iffe
ine

d
aml

ite
Stre

Number of cells F in
Figure 1: CPU time Consumption (from Schlumberger Sofware Development, Abingdon UK).

1 0.9

0.9
0.8

0.8
0.7

0.7
0.6
C u m Freq

0.6
C u mFreq
F re q

fre q

0.5
0.5
Cum

Cum

0.4
0.4

0.3
0.3

Upscaled Model Upscaled Model


0.2
0.2

Geol Model Geol Model


0.1 0.1

0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Body Rank Geol_Model Body Rank


Body Rank 1 - 10 Upscaled Model Body Rank 1 - 10 Geol Model
Upscaled Model

Figure 2: Problem Preserving Pore Volume in Upscaling.


The calculation of the streamlines in a reservoir model allows for a stronger visualization of flow pattern
and unswept areas, which can help to optimize the sweep [2]. First of all the streamlines can be used to
identify which injectors and aquifers the water in a specific well comes from and the fraction of support
from the different sources. Secondly, the volume of oil in place between each source and a producer can
be quantified. This allows for an identification of the areas of the slowest recovery and areas of too high
injection levels. The illustration on the left in Figure 3 shows the typical saturation distribution plot that can
be obtained from any finite difference solver. The illustration to the right shows how injector producer
pairs can be identified from streamlines for the same case, same timestep. From a typical finite difference
model only saturations or pressures can be plotted. These values do not show as clearly the sweep
pattern.
V INGEPET 2005 (EXPL-4-MF-137) 3

Figure 3: Saturation distribution and identifying producer to injector connections.

In recent studies, the application of streamline simulation to support production allocation calculations has
been very useful. Production allocation has traditionally been based on monitoring the productivities of
the different sand levels and then allocating the well production to the individual sand levels according to
the conductivities. When production logs are available these have been used to redefine the allocation
factors. This technique has been used in many studies. The problem with the approach is that it does not
take into account that the pressure support at the different sand levels may vary and the opening and
closing of neighboring wells may impact the allocation. Creating a simple match of liquid production and
injection in a streamline simulator has been a fast way to calculate the production allocation factors to
sands in a number of studies.

Workflow
The objective of streamline simulation can either be identical to the standard numerical simulation or
focused on sweep optimization (area and vertical sweep efficiency increase). It is assumed, prior to
running the simulation model that a diagnostic analysis has been carried out of the well performance to
identify the wells that may have mechanical problems. There is no validity in trying to match the
performance of these wells. The general steps are as follows:

1. Upscaling.
2. Export 3D property model (Geostatistical model).
3. Import property model and production events and history into the streamline application.
4. Run the model to get a match on liquid rates (injection and production).
5. If a match is not reached, check if sands in wells are continuous and/or it is a permeability issue.
Dismiss the realization, if it is a conductivity issue. Adjust the permeability model in the 3D
property model application, if it is a conductivity issue. Repeat points 3 and 4 until the model
injects and produces the historically observed levels.
6. Subdivide model.
7. When a match to the total volumes has been obtained, check the static pressure information and
adjust the pore volume to get a match on pressures.
8. Address the fractional flow match. Adjust the relative permeability curves and check the relative
conductivities of the horizons being produced simultaneously.
9. Streamline paths.
10. Streamline volumes/rates.
11. Uncertainty evaluation.
12. Allocation Factors.
13. Forecasting.
14. Risk Assessment.

Most of the steps mentioned are similar to any numerical simulation job. But there are a few differences.
First of all, with respect to upscaling, the streamline simulation approach can usually run at a lower level
of upscaling. Once a liquid (all phases) production and injection match has been obtained, it is possible to
segment the model into smaller units as the no flow boundaries can be identified. These boundaries are
V INGEPET 2005 (EXPL-4-MF-137) 4
not crossed by a streamline. This segmentation is in general independent of the relative permeability
function (fractional flow behavior).

The upscaling or reduction of number of grid blocks in the model is a very sensitive area. Not only can the
resolution of the fine model be lost, the relative permeability function is sensitive to the grid block size.
The relative permeability curves may consequently have to be adjusted to maintain the same flow profile
for larger size blocks. This subject is fairly complicated so it is only mentioned in this paper.

Checking that the historical volume rate for injection and production can be obtained is the first check of
the validity of the static model. In some cases where the rates cannot be met at all, the sand continuity is
often the issue. Likewise areas where the rates can be met but the model pressures are fairly constant
compared to a declining historical. In cases where there are only problems with the model rate level, the
reason is often related to the permeability transformation. With a focus on liquid production and injection
rather than the individual phases, it is possible to dismiss some geostatistical realizations without much
effort.

Validation of Sand continuity


Initial Property Definition of Productivity Error Finite
Model (Reservoir Definition of model segments Diffe.
Characterization)
Solver

Streamline
Solver
3D
Property
Model Stream
Adjustment of
Model
Permeability Model
With respect to
Streamline results
Fractional Flow
Match

Figure 4: Typical Workflow.


Since the model runs using a very low level of upscaling or no upscaling, the permeability adjustments
are made in the static model and then exported to the streamline simulator. This loop (Figure 4) may be
iterated a couple of times. By adjusting the permeability correlation in the static model more of the
realizations will match on conductivity and the realizations to dismiss become mainly those where sand
continuity is incorrect. Once a match of the total liquid rates is obtained, the model can be segmented
according to the flow lines. This may lead to a simplification of the global model and allow for the
construction of a number of smaller models. These smaller models may run in a typical finite difference
application at reasonable calculation times. The alternative to doing the work in finite difference solver is
to continue to work with the streamlines and to match the fractional flow and static pressures in the
streamline simulator.

In a traditional finite difference solver, the work goes straight into running the model(s) in a forward mode
after the history match has been reached. Streamline solvers allow for a significant amount of diagnostic
work based on the match model. First of all for every producer on line it is possible to identify the source
of the water and the corresponding contribution to the total flow. In Figure 3, the streamline colors
illustrate which producer the lines are connected to. In Figure 5, the lines are filtered with respect to a
single producer and the colors represent which source (injector/aquifer) is providing the flow. For every
injector/aquifer to producer connection it is possible to calculate the remaining oil in place and the
displacement rate. From this, the production time (oil in place over rate) can be calculated. This number
illustrates where insufficient and excess injection is taking place. This is key information in optimizing
water injection especially in mature fields. The tight agreement that can be obtained between the fine
streamline model and the 3D property model strengthens these conclusions.
V INGEPET 2005 (EXPL-4-MF-137) 5

Figure 5: Source of Water connected to one specific producer.


Many of the fields in Colombia, Ecuador and Peru are of a fluvial or delta type. More sand levels are
encountered of limited thickness. To get a commercial rate out of a well more sand levels are often
produced co-mingled. Production logging is seldom carried out and therefore the production allocation to
sand is uncertain. As the field matures it becomes increasingly important to understand where the past
production came from in order to understand where the remaining oil is located.

Production allocation in a well producing co-mingled from more sands is traditionally based on the
productivity indices of the individual sand intervals. This information could have been obtained from
production logging and if no production logging is available the allocation if often based on the
permeability times net height conductivity of the individual zones. The problem with this approach is that
the productivity near the wellbore does not represent inter-well reservoir conductivities nor does it take
into account any imbalance of injection to production. In addition water injection may not occur at all sand
levels. In the example section this is illustrated through a simplified generic case. A simulation model is
never a 100% match to the conditions taking place in the reservoir and is mainly used to represent
general flow behavior and to give answers to observations that cannot be measured directly. When using
production allocation from modeling, it is a combination of model and measurement. The production
allocated to the wells is multiplied by the allocation factors to the individual sands derived from the
simulation model. This results in a production allocation corrected for the error in the model match at well
level.

Study Examples

Production Allocation
The use of streamline simulation for production allocation is described through an example in this section.
In principle the same work flow can be carried out using a finite difference solver. The benefit of using a
streamline simulation application is to get results more closely aligned to the underlying static property
model. The example is created to represent a generalization of a problem often encountered in the region
in water flooded reservoirs. The model consists of three reservoir levels of constant but different
permeability and porosity. Two producers and 4 injectors are defined. The producers produce from all the
reservoir levels but injectors only inject water into 2 of the 3 zones. Injection and production is controlled
by pressure constraints. The geometry of the model is illustrated in Figure 6.
V INGEPET 2005 (EXPL-4-MF-137) 6

Figure 6: Model Structure.


The model was executed and ran for a period representing the reservoir behavior. It is assumed that in
general in simulation studies the productivities can be matched more easily than the fractional flow
behavior. From the original case the production history is used as the model production constraints and a
straight-line relative permeability function is used instead of the original Corey functions. The model now
matches the original model in terms of liquid rate, but not in terms of fractional flow, which is often the
case in many field studies (see Figure 7).

The workflow described is in simplified terms that which is obtained when a match of total liquid rates are
obtained as described in the section,

Workflow points 4 and 5. The issue is, can the model be used for production allocation prior to obtaining
a good match on fractional flow? The reason being that the workflow to the point of having a match on
liquid production and injection is significantly shorter than the time it takes to obtain an agreement on
fractional flow behavior. To verify this, the allocation factors for oil and water for the top and bottom sands
were calculated for every timestep for the reference case and the liquid match case. In the addition the
allocation factor using a traditional permeability height approach was calculated. This is constant in time
as the properties do not vary. The results are illustrated in Figure 8 for one of the producer wells.

Figure 7: Model and Closest match using straight line relative permeabilities.
V INGEPET 2005 (EXPL-4-MF-137) 7

PRD1 Production Allocation

0.8

0.7

0.6

Allocation Factor (Oil)


0.5 Middle Top - Liq Match
Middle - Liq Match
Bottom - Liq Match
Top - Original
0.4 Middle - Original
Bottmom - Original
Top - Kh Allocation

0.3
Top Middle - Kh Allocation
Bottom - Kh Allocation

Bottom
0.2

0.1

0
1

11

13

15

17

19

21

23

25

27

29

31

33

35
Figure 8: Calculated Allocaton Factors.
The results show that in this case the allocation factors calculated from formation productivities are
completely out compared to the allocation of the reference model. Even though the case using straight
line permeability functions gives a poor fractional flow match it gives a close agreement to the allocation
factor for the different phases by wells. All the reference cases studied gave a match in the allocation
factor of less than or about 10 percent even though the fractional flow match was out of scale in some
cases. It validated that the model could be used reliably to calculate the production allocation factors even
though no match to the fractional had been reached.

Sweep optimization
To illustrate the use of streamlines for sweep optimizations, reference is taken to the model illustrated in
Figure 3. This is a two-phase model with oil and water present. For the initial timestep the oil in place
connected by streamlines are calculated injector by injector. The ratio of these volumes divided by the
associated well injection is calculated. This illustrates where additional water injection is required. The
injection rates are then proportioned with respect to remaining oil connected to the injector. In principle
this optimization can be calculated for each timestep. In this case it was only carried out for the initial
timestep and the results are illustrated below in Figure 9. In this simple example the production gain is
more than 10%. Allowing the wells to run under pressure constraints in general causes a water short cut
to be created between close producer injectors that have a high conductivity connection. Water needs to
be allocated to wells that are connected to larger volumes of oil.
V INGEPET 2005 (EXPL-4-MF-137) 8
Figure 9: Injection Optimization.

Field Example
The processes described in the previous examples have been carried out on field scale. But for
simplification purposes, the approaches have been associated with some simplified models. On a field
scale the work needs to be subdivided into horizons to understand if any zones require to be shut-in. The
calculation of allocation factors from the streamline model has allowed for analytical diagnosis processes
to be carried out on a zone by zone level. As a result some people have been able to work on the
production diagnostics level by level whilst others have been working on improving the fractional flow
match. By the time a fractional flow match has been achieved, the diagnostic/analytical approaches have
identified potential infill locations and workover potential is ready to be verified by the dynamic model.

A very strong feature in streamline simulation is the fact that for any streamline it is possible to calculate
the time to producer or from injector at any point along the streamline. The streamlines can then be
truncated with respect to the remaining license period and plotted. In areas where no streamlines can be
observed, no production will take place during the remaining license period. This is illustrated in
Figure 10, using an example from Colombia. In the northern part of this field, flow is taking place
throughout. But in the central part and to some extent in the southern part areas can be found (black)
where no sweep will take place within the license period.

Figure 10: Areas that will not be swept within the license period.

Problem Area
In the illustrated examples, it appears that streamline simulation has a great advantage. But the approach
is associated with a warning as well. Whereas most finite difference solvers define their time steps for the
solver in accordance with the numerical solver errors, this has not yet been implemented in all streamline
solvers. In the majority of cases the user defines the time steps. In cases where gravity segregation is a
significant issue this can represent a problem.

In
Figure 11, a comparison of saturation distribution at a given timestep is illustrated for a streamline model
(left) and a finite difference model (right). The two models represent the same 2D sweep at the same
timestep. The displacement is of an unfavorable mobility ratio. The water is moving fast due to this and
due to the density difference in the high permeable rock (~2 Darcy), water segregates and moves along
the bottom (water under-run). To capture the segregation, the finite difference solver shortens the
timesteps to the range of days. In the streamline simulator, time steps of months have been used.
V INGEPET 2005 (EXPL-4-MF-137) 9
Clearly, in this example, the streamline approach fails to represent the gravity segregation using large
time steps. To get identical solutions and water breakthrough times, the time steps in the streamline
model have to be reduced to under one week. At these time steps, the finite difference solver is faster.

Streamline Finite Difference

Figure 11: Stremline and Finite Difference - Large Steps.


In general if the model can be run in a two phase mode and there is a limited gravity segregation, the
streamline approach gives reliable results compared to the finite difference solver. In a 3 phase model,
there is no general conclusion and so far little work has been done on understanding the compositional
mode.

Conclusions
Streamline simulation has now been carried out as an alternative to the traditional finite difference
simulation in a number of cases and the use is increasing. Currently, it is being used in numerous studies
both by service companies and operators. Where the use of 3D property modeling is increasing, a
change to streamline simulation has been observed. The benefits have been observed in a reduction of
the study time and less time has been spent on upscaling. But the full benefit of the streamline simulation
compared to finite difference simulation has only been used in a few studies.

Before starting any work, it is important to identify which time steps are required in the streamline
simulators to give accurate results and to check that at this scale of time steps, the solver is faster than
the finite difference solver. Once this has been established, the streamline approach can be used for a
number of tasks.

• Validation of the property model: A check can be carried out to see if the static interpretation can
justify the production levels observed
• Segmenting the model: Using the streamlines it is possible to identify how to segment a model
into smaller areas
• Production allocation factors: the allocation factors to co-mingled sands can be estimated from a
model matched to liquid rates. Combined with an analytical application this allows for a rapid
evaluation of upswept areas.
• Injection optimization: by accounting for the volumes between injector and producer as well as
the production rates, it is possible to direct water to areas where the depletion rate is lower,
allowing for a better distribution of the injected water.
• Identifying unswept areas in the field: truncating the streamlines with respect to field
development timeframe, makes it possible to identify areas where oil cannot be
recovered within the timeframe and thereby identify potential infill well locations.
V INGEPET 2005 (EXPL-4-MF-137) 10
References
[1]Bratvedt, F, Bratvedt, K., “A New Front-Tracking Method for Reservoir Smulation”, SPE19805, 1992

[2]Lolomari, T., Bratvedt, K. Crane, M., “The Use of Streamline Simulation in Reservoir Management:
Methology and Casee Studies”, SPE63157, 2000

Вам также может понравиться