SPE 95402
Revisiting Reservoir Flood-Surveillance Methods Using Streamlines
Rod P. Batycky, Marco R. Thiele, Streamsim Technologies Inc., Richard O. Baker and Shelin H. Chung, Epic Consulting
Services Ltd.
Copyright 2005, Society of Petroleum Engineers
This paper was prepared for presentation at the 2005 SPE Annual Technical Conference and
Exhibition held in Dallas, Texas, U.S.A., 9 – 12 October 2005.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
This paper revisits classic flood surveillance methods applied
to injection-production data, and how such methods can be
improved using modern streamline-based calculations. Classic
surveillance relied on fixed patterns and geometric based wellrate allocation factors (WAF’s).
Here we compare
conclusions about pattern performance from classic
surveillance calculations against surveillance results using
flow-based WAF’s as computed from a streamline
surveillance model. We show that very different conclusions
on pattern performance can be reached. Rather then fixed
patterns, we also introduce injector centered patterns as the
elementary surveillance unit with offset producers being those
that an injector is connected to at any time event. Injector
centered varying patterns give a better measure of an injector’s
true effectiveness since all offset oil it is responsible for is
counted, rather than just production within a pre-defined
pattern.
Here too we show that decisions about how to
manage a flood can be quite different than decisions reached
using a fixed pattern analysis.
In the second part of this paper we illustrate how much
data is required to build a relevant surveillance model. We
compare WAF’s and offset oil production as computed from
full history-matched flow simulation models against
surveillance models for several field cases. Our comparisons
show that as long as offset well rates are a function of
neighboring well rates, as is typical in waterfloods, then only
first order flow effects are required in a surveillance model.
First order flow effects required would include, well locations,
historical rates, gross geological bodies and major flow
barriers. In other words, because well rates are already a
reflection of geology, simply setting up a surveillance model
with the proper geological connectedness will lead to
streamline models that give similar inter-well fluxes, as more
complex history matched simulation models.
Introduction
Fundamental to good reservoir management of waterfloods
and miscible floods is the surveillance of production data. This
type of surveillance is useful to understand flood performance
to date and can highlight good vs poor areas. In particular,
surveillance can identify areas of extreme water cycling,
patterns with poor sweep, or local voidage imbalances,
without having to construct a detailed flow simulation model,
providing more “real-time” monitoring of a flood.
The
specifics of standard surveillance methods, such as a voidage
plot, or pattern recovery plots, are discussed in detail by
Baker. 1,2
The basic element of all surveillance diagnostics is the
association of produced volumes with injected volumes via
well-rate allocation factors (WAF). A WAF defines how
much flow at a producer is due to each offset injector, or visa
versa, how much injection goes to each offset producer.
Unfortunately the WAF’s are also the Achilles’ Heal of these
surveillance methods as results are heavily dependent on the
assumptions made to compute the WAF’s. For large multiwell floods, pattern definition and WAF calculation are a time
consuming process. Traditionally the WAF’s have been based
on geometric arguments (angle open to flow) or were
computed from simple 2D streamline models and usually
assumed fix through time.3-5 Thus routine surveillance on a
pattern-by-pattern basis is accepted as having significant
limitations, or is rarely practiced.
Modern streamline simulators can account for 3D full-field
systems with complex geology, multiple wells, multiple wellrate events, and multi-phase flow physics.6 These types of
simulations routinely show that WAF’s are neither fixed or
solely a function of geometry and that there can be substantial
flow between well pairs outside of predefined patterns.6-9
Although flow-based WAF’s are more realistic, their
calculation for surveillance purposes has never been the main
goal of streamline-based simulation. However, Grinestaff10
recently showed how streamline-based WAF’s can be used
qualitatively for flood management, and Thiele & Batycky11
used WAF’s in a workflow to set well-rate targets to reduce
fluid cycling.
In this paper we will step back from flow simulation and
instead focus on the use of streamline-based WAF’s in
reservoir surveillance of production data. First, we will show
how standard surveillance techniques can now be easily
implemented on a pattern-by-pattern basis, and how these
compare to the results with traditional geometric WAF’s. We
2
SPE 95402
also discuss the limitations of fixed-pattern surveillance,
regardless of how the WAF’s are computed. Additionally, we
propose injector centered patterns that change with time to
reflect the changing wells that an injector supports. Lastly, we
will illustrate the validity of flow-based WAF’s and how much
data is required in a surveillance model to compute relevant
WAF’s for surveillance purposes. Results suggest that flowbased WAF’s are a strong function of first-order flow effects
like historical well rates and gross geological features only implying that relevant surveillance models can be quickly
constructed with minimal data.
production data by relating injection to production. The
disadvantage of this type of model is that there are minimal
predictive capabilities and no visual image of where remaining
reserves may be. In fact, because pore volume is not required
to compute WAF’s, normalized surveillance diagnostics that
highlight pattern recovery factors are meaningless unless
proper pore volumes are defined in the model. Lastly, with a
surveillance model there is no understanding of how
physically realistic the model is, as no history match is
possible. We will illustrate later how much information is
required for a surveillance model to be representative of the
“true” well pairs.
Flow-Simulation Model vs Surveillance Model
Substantial focus in the literature has been devoted to using
and extending streamlines in flow simulation. Development
has centered on the transport of fluids along streamlines and
how the prediction of phase rates and insitu phase distributions
compare with classic simulation methods.
In this paper, we are not interested in the detailed paths
streamlines take between wells or how fluids are transported
along these streamlines. Instead, for surveillance we are only
interested in well connections and the total flux carried along
all streamlines between each well pair – the well allocation
factor. To calculate the WAF at producer p due to injector i,
we sum the total flux of all streamlines at p due to i (Qsp-i),
giving the total flux between the pair at p (Qp-i), and then
normalize by the well rate to give:
Q p −i
∑
nsl
Q sp −i
(1.)
Qp
Qp
To calculate injector centered WAF’s we reverse i & p in the
above equation. Thus while a streamline map may look as
shown in Figure 1, we are interested only in the flux pattern
map derived from the streamlines (Figure 2). The FPmap
collapses all streamlines into single segments where the
thickness can be used to represent the percentage of the flux
from the originating injector (the WAF), and color is used to
identify originating injectors or supported producers.
Once the WAF’s are know we can compute the phase
production rates associated with each injector as:
WAF p −i =
=
Q& op −i = WAF p −i Q& op
s =1
Figure 1: Streamlines colored by originating injector for
Field A at last time event.
(2.)
where Q& op represents the known historical oil rate of producer
p. Thus Equation 2 is simply a pro-rating of historical rates
back to connected injectors as identified by a producer’s
WAFs. Physically this means that if a producer has a given
historical watercut, this watercut is assumed to be the same
from each supporting injector.
The advantage of a surveillance model is that it is easy to
build with the main requirements being historical well rates,
well locations, and a reservoir grid. The grid can range from a
simple 2D homogeneous system to a 3D heterogeneous model
with complex geology. The model requires no history
matching and calculations are fast because there is no
transport problem to solve. However, a pressure solve and
streamline trace are still required so the model can be a good
precursor to a flow simulation model. Chapman & Thompson
note that a surveillance model provides an in-depth look at
flood performance to date, and is another way to interpret
Figure 2: Flux pattern map colored by injector for Field A
at last time event. Thickness represents relative flow rates
from each injector.
Reservoir Surveillance
In this section we will study Field A and compare surveillance
diagnostics based on the classic method of fixed patterns,
versus injector centered changing patterns. For reference,
Field A is medium-weight oil waterflood with 30-years of
SPE 95402
production history, containing approximately 30 injectors &
150 producers. Fluid densities are similar and the end-point
mobility ratio is about 10.
Fixed Patterns – Classic Surveillance Method
To begin, we divided the Field A reservoir into 32 fixed
patterns, some of which are show in Figure 3. A fixed pattern
assumes that the pattern boundaries do not change, and wells
are assigned to a pattern on the basis of either being inside or
outside the pattern boundary. We have selected 8 inner
patterns for closer analysis. We further assumed that each
pattern contains a single injector and that flow from each
producer is only assigned back to patterns with active
injection. These assumptions will aid in comparison with
injector centered changing patterns shown later. Thus, if there
is no active injection in a pattern then production cannot be
allocated to the pattern. Geometric allocation factors were
computed from a perceived angle open to flow between each
producer and the neighboring offset active injection patterns
the producer belongs to. The geometric WAF’s were
recomputed for each time event to account for injector
activations and shut-ins, but are not a function of offset well
rates. The phase rates between each well pair were then
computed using Equation 2 and oil production then assigned
to each pattern. By definition, the sum of the WAF’s at each
producer adds to 100% of the producer’s flow (Figure 4).
Figure 3: Pattern outlines and wells within each pattern
for Field A.
The principal reasons for defining fixed patterns in
reservoir surveillance are to, identify inter-pattern flow,
which patterns have over vs under injection, and pattern
recovery factors. Because recovery factors require knowledge
of pattern pore volumes we focus only on diagnostics that
address inter-pattern flow and offset performance issues.
The key diagnostic plot for the 8 patterns is as shown in
Figure 5, which plots cumulative voidage produced vs
cumulative voidage injected – a cumulative voidage plot.
Patterns below the 45o line have over injection (outflow) while
patterns above the 45o line have influx. These 8 patterns show
a wide range of under to over injection, with some patterns
initially showing net efflux and then crossing over to net
3
influx. Note that even though these are cumulative volume
plots there is still “noise” in the production curves.
25%
25%
19%
33%
11%
50%
geometric WAF’s
37%
flow-based WAF’s
Figure 4: How production is allocated back to patterns a
producer is assigned to for geometric vs flow-based
WAF's. With flow-based WAF's a producer can receive
support from a pattern it has not been assigned to.
Figure 5: Cum reservoir volumes produced vs cum
reservoir volumes injected for 8 patterns of Field A using
geometric allocation factors. Patterns above the unit slope
line are under-balanced whereas patterns below the unit
slope line are overbalanced. PAT1 is in red.
Rather than using geometric allocation factors, flow-based
allocation factors as computed from the streamlines of a
surveillance model can also be used. Presumably flow-based
WAF’s give “better” answers since the inherent limitations
with geometric allocation factors are removed. Flow-based
WAF’s reflect well positions, well rates, and locations and
rates of active neighbor wells – producers or injectors.
Unlike geometric allocation factors, flow-based allocation
factors account for pattern influx/efflux explicitly. For
example, the allocations summed over the patterns for which a
producer is assigned to will in general not add to 100%
(Figure 4). In this example, the remaining production is influx
from another injector and needs to be assigned back to the
4
patterns the producer belongs to. There are many ways to
assign this remaining production, which in turn impacts the
level to which a pattern appears to be over or under balanced.
For work presented here, we have assumed an arithmetic
weighting of the unassigned production to each active injector
pattern a producer belongs to.
Figure 6: Cum oil production vs cum water injection for
11 patterns of Field A (top). Cumulative voidage plot for
the same patterns (bottom). Both plots use flow-based
WAF’s and fixed pattern boundaries. PAT1 is in red.
Again we show the standard cumulative voidage plot for the 8
patterns (Figure 6 - bottom). As a comparison to Figure 5, the
x-axis remains unchanged, but the y-axis (produced volumes)
is quite different for each pattern. Overall these 8 patterns
appear more in balance than when computed via geometric
WAF’s. The key conclusion, which patterns are influxing vs
effluxing, is now quite different. This is expected since any
change to the WAF values (ie: using flow-based WAF’s) will
alter the voidage calculation, illustrating how sensitive a
voidage plot is to the underlying WAF values.
Further note that with flow-based WAF’s the cumulative
curves appear to be smoother (black curve for example). This
is because for geometric WAF’s, a change in production
simply results in a spike change in rate allocated to the
associated injector.
However with flow-based WAF’s,
changes in injection-production also impact the values of the
WAF’s so for example, an increase in injection is offset by an
increase in production, giving smoother curves. Thus other
SPE 95402
surveillance plots, which are derivatives of the cumulative
plots, such as WOR vs cumulative oil production, are
somewhat smoother using flow-based WAF’s.
Injector-Centered Changing Patterns with Flow-Based
Allocation Factors
In reality flow is not confined to fixed patterns. Instead it is
the reservoir geology that dictates well rates and how
producers & injectors communicate. In our experience with
numerous waterfloods, there is often substantial inter-pattern
flow. This aspect is not captured with geometric allocation
factors. Furthermore, operators are limited by how much they
can alter well rates to try and balance defined patterns. Thus,
rather than defining arbitrary fixed patterns, we instead focus
on injector centered changing patterns whose offset producers
are those that the injector supports at each time-event,
regardless of location. The offset producers are identified by
calculating the streamlines connecting to each injector, as
computed from a surveillance model. In addition, the
streamline fluxes are also computed so that flow-based
allocation factors (Equation 1) consistent with the changing
patterns are calculated.
The disadvantage of a variable pattern associated with
each injector over time is that we no longer have a diagnostic
such as the cumulative voidage plot to quantify inflow/outflow
since there is no fixed pattern boundary. Additionally,
because the pore volume is no longer constant for each
pattern, diagnostics that are normalized by pore volume, such
as pattern recovery efficiencies, are more difficult to define.
However, the advantage of defining patterns in this manner is
that we can quantify the true effectiveness of each injector.
Additionally we eliminate the time consuming process of
having to define the patterns in the first place.
Figure 7 illustrates the two key plots of variable pattern
surveillance to quantify injector effectiveness - cumulative
offset oil production vs cumulative injection and current offset
oil rate vs injection rate. Note on the cumulative plot how the
injectors contributing to the most oil production are quite
different than predicted using fixed patterns (Figure 6 – top).
This is because with the fixed pattern method, production due
from one injector can be assigned to another injector counting
as pattern influx.
The bottom plot shown in Figure 7 is the derivative of the
cumulative plot and was presented as the injector efficiency
plot by Thiele & Batycky.11 Each point on the plot represents
a single injector at the last time event, and is a measure of how
effective each injector currently is in the field. Injectors close
to the 45o line would represent the maximum efficiency,
implying that for every barrel injected an offset barrel of oil is
produced. The importance of this plot cannot be overstated.
For example, the cumulative voidage plot from a fixed pattern
analysis (Figure 6 - bottom) for the high cumulative injector
pattern PAT1 (red line) indicates that the pattern is currently at
a cumulative voidage of almost 2, suggesting over injection
and that the injection rate should be cut back. However, this
same injector, highlighted in red in Figure 7 (bottom) is one of
the highest efficiency injectors in the field (25 m3/d oil
produced for 80 m3/d water injected), suggesting that reducing
the injection rate will lead to reduced offset oil production.
SPE 95402
Why these conclusions conflict is highlighted in the
FPmap of Figure 8, which shows this high efficiency injector,
its pattern boundary, and the wells it is currently supporting.
The FPmap clearly shows that out of pattern support is
occurring and where, and it appears because the off-pattern
producers (mainly north-east and south-west) are receiving
minimal support from neighboring injectors. This explains
why this injector has a high injection efficiency – it is
supporting significant oil production outside of its own
pattern. Lastly, this comparison highlights another limitation
of the cumulative voidage plot (geometric or flow-based WAF
derived) – it only tell us that influx/efflux may be occurring
but it does not tell us where. Thus if one is trying to rebalance
patterns, using a cumulative voidage plots is a guessing game
at best. Computing FPmaps for each injector is a better
alternative.
5
PAT 1 p attern
b o und ary
Figure 8: Pattern boundary for PAT1 overlaid on top of
the FPmap showing which wells the central injector (in
red) is supporting.
Accuracy of a Surveillance Model
In order to quickly survey production data we built a model
with only limited realism. We ignored inter-well porosity &
permeability effects, and transport effects such as PVT,
relperm, and gravity. We do not have a sense for how
accurately a surveillance model captures well-pair
connections, and thus how close the allocation factors are to
the “true” reservoir. In other words, what first order effects
are important to ensure we have a relevant surveillance
model? Here we will look at three fields and compare
surveillance diagnostics between a fully history match flow
simulation model, and various simplified surveillance models
to answer this question. For each field we constructed three
surveillance models, the first model contains homogeneous
permeability & porosity arrays but retains the major
permeability barriers and areal extent of the major flow units
(3D w\ barriers), the second model is a 3D homogeneous tanktype model, and the third surveillance model is a 2D
homogeneous tank-type model. For each model, we retained
the same well locations and historical rates, as the history
matched model.
Figure 7: Cum oil production vs cum water injection for
11 injectors of Field A (top). Injection efficiency plot for
the latest time period for 9 of 11 currently active injectors
(bottom). Both plots assume flow-based allocation factors
with injector centered changing patterns. PAT1 is in red.
Field C
Field C and the associated history matching results have
already been presented by Baker et al.7 To summarize, this is
a heavy-oil field with approximately 150 wells, 37 years of
history, 34 of which were waterflood. The reservoir contains
3 main sand bodies, each one in weak communication with the
other where they overlay, but with each sand having very
different areal profiles.
The key metric we are using to compare each model to the
reference history matched model, is the flux between each
well-pair. Figure 9 (top) is a cross-plot, for the last time event,
6
showing the flux for each pair in each surveillance model,
versus the same pair for the history matched model.
Figure 9: Well-pair flux for surveillance models vs wellpair flux for history matched model for the last time event
(top). Correlation coefficient, as computed from the top
plot, but for each time event for each model (bottom).
First, this plot shows that high rate pair connections are
consistently identified in all models vs low rate pair
connections. At rates below about 20 m3/day, pairs are
identified in the surveillance models that are not identified in
the reference model and visa-versa. Second, this figure shows
that the 3D homogeneous model with the major flow units has
a better correlation to the history matched model than either
the 3D or 2D tank-type models. We calculated a standard
correlation coefficient for the time even shown in Figure 9
(top) and all the other time events, and plot in Figure 9
(bottom). Note that for early production the correlation is not
as good as at later times, regardless of which surveillance
model is used. This is because early on there are fewer wells,
more uniform total well rates and no significant waterflood
response. However, from time event 12 and on, injection
support is established and producer rates are a stronger
function of injection rates, meaning well pairs are more easily
identified regardless of model. Thus large rate producers/
injectors have well defined WAF’s whereas WAF’s for small
SPE 95402
rate wells will be more uncertain and more sensitive to the
underlying model type.
Figure 10: Offset oil rate for each surveillance model vs
the history matched model for the last time event (top).
Correlation coefficient for offset oil production for all time
events in Field C (bottom).
What we also what to know however, is how the
correlation in well pair rates translate to the key surveillance
plots illustrated in the previous section. Figure 10 (top),
compares the offset oil rates between the 3 surveillance
models and the history matched model. Similar to the wellpair correlation, we see better agreement with the 3D model
than the tank-type models. More important is that all models
do a good job of identifying high vs low offset oil volumes on
a per-injector basis. Correlation for the offset oil production is
greater than that for well-pair rates and is due to the effect of
summing up WAF values over a number of well pairs for each
injector, to compute the total offset oil rate. For example,
early on when breakthrough has not occurred, even completely
incorrect WAF’s will still sum to the same oil rate on a per
injector basis regardless of which surveillance model is used.
In other words, while the FPmap for a given injector will
differ between models, a surveillance number like offset oil
production does not tell us where offset oil production is
occurring so is less sensitive to correct model connectivity.
Lastly, to save space we show just two model cumulative
production plots in Figure 11. By visual comparison, the 3D
SPE 95402
surveillance model with major flow barriers does a good job
of representing the offset oil production for each injector,
compared with the history matched model.
7
models and the history matched model, for all the time events
(Figure 12). As with Field C, reverting to a tank-type model
reduced the correlation, particularly early on before
waterflood response was significant. However, the averaging
effect of computing offset oil production on a per pattern basis
again resulted in a good correlation regardless of the type of
surveillance model considered.
Figure 11: Comparison of offset cumulative oil vs water
injected for each injector, for the history matched model
and the 3D with barriers surveillance model of Field C.
The obvious question to ask is, why can a surveillance
model agree well with the history matched model? The main
reason, as evident in Figure 9 (top), is that calculated rates
between well-pairs can show a high degree of correlation
among each model and in particular for higher rate
connections. Recall that historical reservoir rates are the same
in all models. Furthermore, well rates are already a reflection
of well communication, which in turn are a reflection of
geology (ie: high rate producers tend to be near high rate
injectors). Thus, barring omission of gross geologic features,
the streamlines will tend to connect well-pairs based on well
rates in a similar fashion, regardless of small scale geologic
features. This is why we see reduced correlation for the two
tank-type models, but a good correlation for the 3D
homogeneous model with the major flow units and major
permeability barriers defined.
Field A
Field A is that used in the previous section for comparing
surveillance diagnostics between fixed pattern vs injector
centered changing patterns. Rather than show the details of
the surveillance comparisons, we just show the summary of
the correlation coefficient between the three surveillance
Figure 12: Correlation coefficient for well-pair flux (top)
and offset oil production (bottom) as predicted by 3
surveillance models compared against a history matched
model, for Field A.
Field B
Field B and the associated history matching results have
already been presented by Baker et al.7 To summarize, this is
a light-oil reservoir with five distinct sand bodies each with
partial continuity areally. There are large ranges in average
permeability between each body and Kv/Kh is about 0.1 within
each sand. Unlike Fields C & A, this field has only a
marginally unfavorable mobility ratio, meaning that producer
injector connections may not be as strong owing to better
areal sweep with less channeling of injected fluids towards
producers.
Again, three surveillance models were built from the
history matched model. Figure 13 shows the correlation
between the various models against the history matched
model, for both well-pair flux, and the computed offset oil
8
production for each injector. Because of the more favorable
mobility ratio, the estimate of well-pair fluxes is not as good,
particularly before significant waterflood response (before
time event 16). The correlation is again worse for the tanktype models but improves if the large scale permeability
barriers are retained.
Figure 13: Correlation coefficient for well-pair flux (top)
and offset oil production (bottom) as predicted by 3
surveillance models compared against a history matched
model, for Field B.
Discussion
We have shown the relevance of a surveillance model for 3
waterfloods with varying geology and fluid properties, but
these fields were all ideally suited to streamline simulation to
begin with. We saw better correlation in all models post
waterflood response, and we saw better correlation in the
models with higher mobility contrasts. Clearly an extension to
the work will be to quantify surveillance models against a
history matched model for additional systems with geology or
underlying displacement physics that are different than what
we have shown here. For example, miscible gas-type
processes, or highly compartmentalized fields.
All models in this work assumed incompressible flow with
open reservoir boundaries. Compressibility raises two issues.
First, does compressibility add any additional considerations
when building a relevant surveillance model? We believe that
SPE 95402
as along as production rates are a function of injection rates,
then the system is behaving similar to an incompressible
system meaning that relevant well connection will still be
identified in a surveillance model. However, the second issue
with compressibility is that total flow between a well-pair may
not be a constant between the pair. Thus, for streamlines that
originate from an injector and arrive at a producer we cannot
quantify how much arriving production is related to the
originating injector versus compressibility effects along the
arriving streamline. Similarly, we cannot quantify how much
injection leads to production, versus losses due to
compressibility. Since these effect will most likely be seen
only in highly compressible systems, which are ill-suited for
streamlines anyway, its unclear how important compressibility
effects really are for surveillance on systems well suited to
streamlines. Further investigation in this area is needed.
Although not addressed in this work as, no models had
aquifers, another advantage of a surveillance model is that
streamline calculations can quantify effects due to fluid lost or
gained from an aquifer. It is possible to identify production
due to aquifers, and thus not allocate this production to a
neighboring injector. Conversely, it is possible to identify
injection losses to an aquifer to quantify overall injection
efficiency with & without these losses.
Lastly, we only focused on diagnostics that indicated
injector efficiency (fluid cycling) or inter-pattern flow. We
did not look at surveillance diagnostics whose primary goal is
to identify areas of low vs high recovery. As we showed, to
identify injector efficiencies the natural pattern element is all
the producers an injector sees coupled with the FPmap.
However, to identify good vs poor pattern recoveries a fixed
pattern element (with flow-based WAF’s) is better, as
cumulative volumes can then be normalized by the fixed pore
volume of each pattern. Similar to our observations for
cumulative voidage plots and their sensitivity to how WAF’s
are computed, our assumption is that pattern recovery factors
too can be quite different depending on whether geometric or
flow-based WAF’s are used, and how influx/efflux volumes
are assigned to each pattern.
Conclusions
1. Classic surveillance using fixed patterns and geometricbased allocation factors is prone to several inaccuracies.
First, geometric allocation factors do not account for
inter-pattern flow, yet we know this flow always occurs.
Second, we showed that fixed-pattern results are heavily
dependent on how the allocation factors are computed. A
simple property like influx/efflux (cumulative voidage
plot) can show a pattern as over producing or over
injecting depending on how the allocation factors are
calculated. Furthermore, while we can compute influx/
efflux we do not know where it is occurring.
2. Flow-based allocation factors used with fixed patterns
give smoother offset production behavior than geometric
WAF’s. As expected, the influx/efflux behavior is also
consistent with the underlying flux-pattern maps, from
which the flow-based WAF’s are derived. Using these
maps it is now obvious to see where influx/efflux is
occurring within each pattern.
SPE 95402
3.
4.
5.
We also proposed defining injector centered patterns
where offset producers associated with each injector can
change with time as the well connections, based on
streamlines, change with time. With this type of pattern,
we want to quantify the true effectiveness of each injector
by computing all oil production it is responsible for
regardless of where the production is occurring. In this
manner, it is possible to reach a different conclusion about
an injector’s effectiveness, than from a fixed pattern
analysis.
We showed that a relevant surveillance model can be
constructed by only providing gross geologic features,
well locations, and historical well rates. We found that
well-pair rates, and thus WAF’s, correlated well between
a history matched model and a surveillance model with
these first order flow effects, especially for high-rate well
pairs. We attribute this behavior to the fact that, well
rates in waterfloods, particularly unfavorable mobility
ratio floods, are already a reflection of the geology and
connectedness of the reservoir. Adding smaller scale
inter-well geology like permeability, porosity, or facies
type, does little to change the overall flow between wellpairs.
When comparing even 2D surveillance models to the
history matched model for each field, we noted better
correlation in offset oil production than well-pair rates.
This is because of the lumping effect of calculating offset
oil over all of an injector’s producers (well-pairs). Thus
the caution is that while surveillance plots may appear to
adequately represent field behavior, the underlying
FPmaps may not be entirely correct.
Acknowledgments
Thank you to Cameron McBurney, Bob Mckishnie and
Frank Kuppe at Epic Consulting for history matching the Field
A, B, C streamline models.
References
1. Baker, R.O.: “Reservoir Management for Waterfloods,”
JCPT (1997) 36(4), 20-24.
2. Baker, R.O.: “Reservoir Management for Waterfloods –
Part II,” JCPT (1998) 37(1), 12-17.
3. Chapman, L.R. and Thompson, R.R.: “Waterflood
Surveillance in the Kuparuk River Unit With
Computerized Pattern Analysis,” JPT (1989), 277-282.
4. Sharma, A.K. and Kumar, A.: “Areal Pattern Distribution
of Remaining Oil Saturations in a Mature West Texas
Waterflood – A Case History,” SPE paper 35202
presented at the Permian Basin Oil & Gas Recovery
Conference, Midland, Texas, 27-29 March 1996.
5. Flanders, W.A. and Bates, G.R.: “Optimizing Reservoir
Surveillance
by
Using
Streamlines
and
the
Microcomputer,” SPE paper 16482 presented at the 1987
Petroleum Industry Applications of Microcomputers held
in Del Lago, Montgomery, Texas, June 23-26, 1987.
6. Batycky, R.P., Blunt, M.J., and Thiele, M.R.: “A 3D Field
Scale Streamline-Based Reservoir Simulator,” SPERE
(November 1997) 246-254.
7. Baker, R.O., Kuppe, F., Chugh, S., Stojanovic, S., and
Batycky, R.: “Full-Field Modeling Using Streamline-
9
Based Simulation: Four Case Studies,”, SPEREE (2002)
5(2), 126-134.
8. Lolomari, T., Bratvedt, K., Crane, M., and Milliken, W.:
“The Use of Streamline Simulation in Reservoir
Management: Methodology and Case Studies,” paper SPE
63157 in proceedings of the 2000 ATCE, Dallas, TX
(October).
9. Grinestaff, G.H., and Caffery D.J.: “Waterflood
Management: A Case Study of the Northwest Fault Block
Area of Prudhoe Bay, Alaska, Using Streamline
Simulation and Traditional Waterflood Analysis,” paper
SPE 63152 in proceedings of the 2000 ATCE, Dallas, Tx
(October).
10. Grinestaff, G.H.: “Waterflood Pattern Allocations:
Quantifying the Injector to Producer Relationship with
Streamline Simulation,” SPE paper 54616 presented at the
SPE Western Regional Meeting, Anchorage, Alaska, 2628 May, 1999.
11. Thiele, M.R. and Batycky, R.P.: “Water Injection
Optimization Using a Streamline-Based Workflow,” SPE
paper 84080 in proceedings of the 2003 SPE ATCE,
Denver, Co, 5-8 October.