You are on page 1of 13

SPE 101880

History Matching Considerations of an Analogue Reservoir Model (ARM)


E.T. Montague, SPE, Curtin U. of Technology*; D.H. Sherlock, SPE, CSIRO Petroleum; and E. Santoso, SPE, Curtin U. of
Technology**

Copyright 2006, Society of Petroleum Engineers


Introduction
This paper was prepared for presentation at the 2006 SPE Asia Pacific Oil & Gas Conference and The ARM project is a unique experimental program between
Exhibition held in Adelaide, Australia, 1113 September 2006.
CSIRO Petroleum and Curtin University of Technology that has
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
been devised to investigate issues relating to uncertainty in
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to reservoir simulations of channelised fields and their seismic
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 expression1. The ARM program is designed to integrate aspects of
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
seismic and reservoir engineering, through the construction of a
without the written consent of the Society of Petroleum Engineers is prohibited. Permission to laboratory-scale model of sand bodies with flow characteristics
reproduce in print is restricted to an abstract of not more than
300 words; illustrations may not be copied. The abstract must contain conspicuous that can be monitored and modelled, and that can be scaled to
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.
reflect fluid flow behaviour on a field-scale.

Abstract The increase in costs associated with the push towards deepwater
This paper describes a research program at the Australian reservoir production means that risks in exploration and
Resources Research Centre (ARRC) to establish and use an development remain significant, despite continual technological
analogue model to gain insight into issues of uncertainty in advances. The understanding of risk and uncertainty in these
numerical reservoir simulation. Reported in this paper are the frontier environments presents a major challenge to the industry.
initial findings from history matching the production response of In these environments permeability architecture and reservoir
an Analogue Reservoir Model (ARM) and its numerical connectivity are key uncertainties which lead to differences in
representation in a finite difference simulation model. The ARM dynamic reservoir performance and estimates of ultimate
is a large-scale physical model comprising two intersecting recovery. This is particularly important in deep-water reservoirs
synthetic sandstone channels encased within an impermeable where the understanding of complex, channelised geological
acrylic matrix. The initially oil-saturated model was waterflooded systems is restricted by the limited resolution of seismic data and
via the upper channel and the injected water was dyed blue to by fewer well penetrations because of the high cost of drilling.
allow the displacement process to be recorded on video. Accordingly, the project sponsors (Chevron and Woodside
Production rates, pressures, water cuts, etc were recorded and Energy Ltd.) requested that the ARM be designed to mimic a
used to history match the reservoir simulation model. turbidite channel system, i.e intersecting sinuous channels with
varying degrees of connectivity.
The simulation model was built using Roxars Nextwell software,
and populated with the rock and fluid properties of the ARM. This paper presents the reservoir simulation modeling component
The initial history matching attempt did not yield a good match. of the ARM project, and discusses the various methods and
The model was unable to match the preferential flow of water results achieved during the process of history matching the ARM
along the lower channel. It was discovered later that the lower to the measured production data. This reservoir simulation work
channel had become inadvertently fractured during fabrication of follows up on the geophysical modeling, which has been
the model. Although incorporating the fracture system into the published previously1.
simulation model significantly improved the quality of the history
match, the character of the water cut development was dependant Analogue Reservoir Modelling (ARM)
upon the relative permeability curve used. It has been well ARM is based around a synthetic cementation technique that
documented that different relative permeability curves are used to allows scaled analogue representations of reservoir systems to be
model fluid flow through porous media as opposed to fracture and conducted in the laboratory. The cementation technique, known
fracture-matrix flow This ARM provides a unique opportunity to as Calcite In-situ Precipitation System (CIPS), allows sandstones
test the applicability of published fracture relative permeability to be fabricated with predetermined physical properties such as
models by using these in history matching the production porosity, permeability and impedance2. Laboratory tests have
response between the reservoir simulation model and the physical shown that CIPS sandstones closely reproduce the acoustic and
model. mechanical properties of natural sandstones3.

A two channel model is presented here as a proof of concept.


The 1:1000 scale model is 1m x 1m x 0.11m, and consists of two
* Now at Brunei Shell Petroleum Sdn. Bhd
** Now at Roxar Pty Ltd.
2 SPE 101880

intersecting sandstone channels within an impermeable and maintained at standard conditions (Pressure = 1 atmosphere,
transparent acrylic matrix. A 3D geological model of the channels Temperature = 25oC).
was initially designed using ROXARs RMS and NextWell
software. The design was then drafted into a CAD model for
computerised machining of the channels into the acrylic
background that represents surrounding low-permeability shales
or mudstones. This ensured exact correspondence between the
numerical and physical model architecture to allow valid
comparisons between the observed and simulated production
response.

Vertical flow communication between the channels exists in two


separate intersecting areas with contrasting vertical
permeabilities. Baseline reservoir-saturation conditions (oil-
saturated with irreducible water) were established by flushing
approximately 10 pore volumes of oil through the initially water-
saturated model. Production from both channels was then Fig. 1 Simulation model of the ARM.
performed via a water flood from a single injection point at the
inlet of the upper channel. The injected water was dyed blue to
allow the displacement process to be recorded on video. Frequent
well tests were conducted to measure production rates. Data such
as time to water breakthrough, produced water-oil ratio, and
cumulative recovery were recorded and used to history-match the
reservoir simulation model. Differential pressures between the
two channels and between the injection and production wells
were also recorded during the experiment for the history match.

Fluid-Flow Scaling
Analogue models cannot reproduce all of the complexities of a
reservoir, but by adhering to established fluid-flow scaling
criteria, many of the same phenomena can be simulated in the
laboratory. Fig. 2 (modified from Lenormand et al., 1988), shows
the different styles of two-phase fluid displacement in terms of Fig. 2 Schematic representation of two-phase fluid displacement
the dimensionless capillary number (Ca) and mobility ratio (M) of styles in porous media, illustrating that the initial ARM flow rates are
the fluids. The capillary number represents the ratio of viscous to scaled to represent conditions in the field.
capillary forces. The permeability of the channels, viscosity of the
fluids, and flow rates used in the model were chosen so that the During this period injected water swept the oil in the upper layer
relative balance of viscous, capillary, and gravity forces acting on until it reached the first area of connection between the upper and
the fluids were representative of the field scale environment being the lower layers. The water then began to migrate into the lower
simulated. channel, developing a preferential flow path at the edge of this
lower channel and thereby leaving large amounts of unswept oil
Experimental Procedure in its central regions. Negligible water encroachment was
The ARM flooding experiment comprised four distinct stages. observed in the upper layer beyond the connection area, but some
fluid movement into this layer must have been occurring since,
To assist in visualizing the experimental arrangement the from simple material balance considerations, 10% of the Stage 1
simulation grid (which will be described in detail later) is shown liquid production from was emanating from this top layer.
in Fig. 1. With reference to this figure, blue dyed water was However, this movement of fluid was limited and decreasing with
injected into the upper layer (IU) and fluid production was time because by the end of the Stage 1, the pressure and therefore
monitored continuously at both ends of the upper (PU) and lower the flow rate in the upper channel was approaching zero. Dry oil
(PL) layers. An overview of the fluid production over the four was produced from the upper layer (PU) throughout the period.
stages is shown in Figs. 3 & 4, where the scales have been
deliberately kept identical to assist in making direct comparisons.Flow was dominated by the lower layer, with 90% of the total
production being collected from this channel. Undyed, clear water
Fig.5 shows the ARM in its initial state, prior to the injection of
the dyed water. Each stage of the experimentation will now be breakthrough in this channel took place after approximately 20
described in more detail. minutes and for the remainder of this period the water cut steadily
increased as the oil rate declined. The initial production of undyed
Stage 1 - The first stage lasted for 125 minutes during which time water shows that some of the water originally present in the ARM
the water injection rate into the upper layer was held constant at had drained into the bottom of the lower channel. Approximately
700mL/hr. Both producers were open for production and were 100 min after water breakthrough the oil rate had declined to a
SPE 101880 3

very low value and the model was approaching the situation of
merely circulating water through the lower channel. During the
latter half of Stage 1 the concentration of dyed water in the
production effluent was increasing with time.

Stage 2 The lower channel producer (PL) was closed in order to


force water into the upper channel and sweep the undrained
central regions of the upper channel. The upper channel producer
(PU) remained open, at standard conditions, and the injection rate
into this top layer was maintained at its original constant value of
700mL/hr. This second stage lasted for 60 minutes.

An interesting observation was the water continued to flow


through the lower channel via the first connection area and bypass
the oil in the central section of the upper channel. The water was
now forced from the lower channel into the upper channel Fig. 4 ARM lower channel production history
through the second connection area, which contained a lower
permeability (approx. 1 Darcy) baffle between the layers and Fig. 7 shows a digital photograph of the water distribution at the
water breakthrough occurred very quickly. Also, some 45 min end of Stage 4. The abrupt end to the passage of water in the
after Stage 2 commenced the coloured dye broke through at the upper channel, and the subsequent poor sweep of the oil in that
upper producer. By this time the water cut from the upper channel channel, is clearly visible.
had reached a value of 90% and so a situation of circulating water
was developing again. Now, however, the options of diverting the
flow path were limited to altering the injection rate.

Stage 3 - The water injection rate into the upper channel was
increased nearly fourfold to 2500mL/hr. Again, only the upper
channel producer was open for production. It was hoped that
increasing the rate might encourage at least some incremental
recovery from the top layer by increasing the relative balance of
viscous to gravity forces. There was some uplift in the oil rate but
the watercut continued to rise continuously, reaching the 97%
level, making accurate measurements of the relatively small
amounts of oil problematic.

Stage 4 - The water injection rate into the upper channel was
further increased to 4400mL/hr for a period of 10 minutes, and
despite the significant increase in viscous forces, the water could
not be forced into the upper channel and continued to circulate
around the model. There was some evidence of a slight uplift in
the oil rate but since the watercut remained at the 97% level the Fig.5 Plan View of the initial State of the ARM
uncertainty in the measurements made it difficult to quantify the
incremental recovery.

Fig. 3 ARM upper channel production history Fig.6 Plan View of the ARM at the end of Stage 2
4 SPE 101880

Fig.9 X-ray CT scan sections depicting fracture network in lower


channel in downstream area of the ARM.
Fig.7 Plan View of the ARM at the end the experiment
Despite these limitations, these simple models are a logical first
Model Limitations step on the path to constructing realistic analogue models. The
It needs to be appreciated that the rather simple analogue models methodology developed will be applicable, in due course, to more
built to date, although far from simple in the making, will not geologically representative models to be built in the proposed
perfectly reflect realistic reservoir architectures even when follow-up work.
suitably scaled. These initial models, built and examined at
ambient temperatures and pressures, are of such an ideal design Simulation Model Construction
that it is unrealistic to suppose they might be found in nature. The previously discussed experimental overview defines the
Also, any imperfections that appear during the model model and the boundary conditions (pressures, injection and
construction, such as cracks, faults and unplanned boundary production rates) that need to be matched using a suitable
effects, will in no way add the desired realism since their simulation model
occurrence has almost certainly not been generated in an
analogous way to subsurface features found on the reservoir scale All dynamic simulation work was carried out using ROXARs
in nature. In the case of the ARM, CT (Computer Tomography) Nextwell Reservoir Simulation code. The output data was viewed
scans were taken of the physical model and these revealed that a using Geovisuals Review visualisation software. A fine-scaled
degree of fracturing exists in the lower channel. This had, model was initially built, but the simulator had convergence
presumably, been inadvertently introduced during the problems due to the small size of the grid cells. In order for the
manufacturing and /or transportation of the model. A number of model to run, the grid needed to be upscaled by a factor of two.
CT scan cross sections were taken (Figs. 8 & 9), and these were This was achieved through the application of GOCAD and the
used to generate the ARM fracture network in the simulation generous assistance of Chevron Australia staff members.
model.
Input Parameters. The final, upscaled 3D gridblock model
comprised more than 120,000 cells, but due to the topology of the
model only about a quarter of the cells were active. Details of the
grid parameters are given in Table 1.

The upper channel is modeled within the first ten layers (z (0,10))
whereas the lower channel is modelled from the sixth to the
twentieth layer (z (6,20)). This means that an intersection may
occur between the sixth and the tenth layers (z (6,10)). Fig. 10
illustrates an example of the intersection.

Fig.8 X-ray CT scan sections depicting fracture network in lower


channel in upstream area of the ARM. Fig. 10 - Cross section of channel Intersection in simulation model.
SPE 101880 5

TABLE 1 - Simulation Grid Parameters OWC level was placed near the bottom of the model, then there
was insufficient volume of water available for production during
Model Dimensions
history matching. Since the model was unable to be used to
Direction: x y z determine the OWC level, an intermediate OWC of 0.035m was
Number of Gridblocks 101 60 20
assumed, which provides sufficient distance from the production
outlet and volume of water in the model.
Dimensions of Gridblocks (mm) 10 10 2
Length (m) 1.01 SCAL data was not available at the time of the modeling, so end
0.6 0.04
point saturation of 10% have been assumed, i.e. Swc=Sor=10%.
Total Number of Gridblocks 121200
The CIPS rock is strongly water-wet, and so a krw of 0.1 was
Active Cells 31635 assumed, and an end-point mobility ratio of 0.5 was calculated
Inactive Cells 89565 (Appendix1-Equation 1) The relative permeability curves were
generated based on the fractional flow theory of Buckley and
Channel Properties
Leverett and scaled to match the favourable end point mobility
Region: Upper Lower Common ratio (M) of 0.5. Therefore, the relative permeability curves are
Porosity 0.318 0.325 0.250
required to generate a fractional flow curve (Appendix1-Equation
2) with the inflection point near Sor. The relative permeability
kx (D) 38.9 30.0 1 curves were generated using the general Corey Exponent
ky (D) 38.9 30.0 1 equations (Appendix1-Equations 3 & 4). The Corey exponents
used were No=1.6 and Nw=6, and the relative permability and
kz (D) 38.9 30.0 1
fractional flow curves are depicted in Figs. 11 & 12 respectively.
The oil / water capillary pressure was assumed to be zero, which
The operating conditions, the number of injectors and producers, is reasonable considering the very high permeability of the rock.
and the fluid properties are given in Table 2.

A configuration of one vertical injector per column across the


face of the upper layer was considered to be the best way of
replicating the distributed injection across the face. Each
gridblock in the face was assigned an initial water saturation of
100% to remove any unwanted relative permeability effects.

TABLE 2 - Simulation Model Parameters


Initial Conditions

Pressure (bar) 1
o
Temperature ( C) 25
Initial OWC (m) 0.035
"Wells"

Number of injectors 1 (upper channel)


Number of producers 2 (1 per channel) Fig. 11 ARM porous media relative permeability curve.
Internal Diameter of Tubing (mm) 6
Fluid Properties

Fluid: Water Oil


3
Density (kg/m ) 1100.040 800.026
Formation Volume Factor (rb/STB) 1.0 1.0
Viscosity (cp) 1.0 5.0
-5 -1
Compressibility (10 bar ) 4.225 1.000

An initial OWC had to be placed into the bottom of the lower


channel in order to model the early breakthrough of undyed water
as mentioned previously. Some early sensitivities were run to try
to determine the contact level, however in all cases the simulation
model predicted rapid onset of water production. This is due to a
combination of the high permeabilities, high kv/kh ratio, and the
small vertical distance to the production outlet. However, if the
Fig. 12 - ARM porous media fractional flow curve.
6 SPE 101880

0.020 100% 0.020 100%

ARM_Oil Rate
0.018 90% 0.018 90%
Initial Run Oil rate

0.016 ARM_Water Cut 80% 0.016 80%


Initial Run Water Cut
0.014 70% 0.014 70%

ARM_Oil Rate
Oil Rate (m 3 /d)

0.012 60% 0.012 60%

Oil Rate (m3/d)


Initial Run Oil Rate

Water Cut

Water Cut
ARM_Water Cut
0.010 50% 0.010 Initial Run Water Cut 50%
Lower channel shut in

0.008 40% 0.008 40%

0.006 30% 0.006 30%

0.004 20% 0.004 20%

0.002 0.002 10%


10%

0.000 0%
0.000 0%
0 50 100 150 200 250 300
0 20 40 60 80 100 120 140
Time (Minutes)
Time (Minutes)

Fig. 13 Lower channel history match; initial model. Fig. 14 Upper channel history match; initial model.

3.5

3
ARM dp End Stage 3
Simulated dp
2.5
Pressure (bara)

2
End Stage 2

1.5

End Stage 1
0.5

0
0 50 100 150 200 250 300
Time (minutes)

Fig. 15 Pressure history match; initial model. Fig. 16 Simulation model at end Stage 1; initial model.

Fig. 17 Simulation model at end Stage 2; initial model. Fig. 18 Simulation model at end Stage 4; initial model.

Results Fig. 16 shows the simulation model at the end of Stage 1, where
Initial Model Figs. 13 to 18 depict the results of the initial water in the upper channel has almost migrated to the production
simulation run, with no history matching applied, and the quality outlet, and the lower channel is almost fully swept. This does not
of the match was poor. During Stage 1, the water cut has not built match with the physically observed response (see Fig. 6) where
up in the lower channel until near the end when the injected water water migration along the central section of the upper channel
reaches the production outlet, and the drop in fluid rates and does not occur even at the end of the experiment, and the lower
pressure observed in the upper channel of the ARM are unable to channel is only partially swept. The simulation model predicts a
be matched. steep increase in water cut at the beginning of Stage 2, because
the upper channel is mostly filled with injected water.
SPE 101880 7

The pressure differential (dp) across the model is also poorly reservoirs. Several of these were selected to cover a range in
matched, especially after Stage 1 where in the ARM the fluid possible relative permeability curves and to use the history
continues to flow via the lower channel, but in the simulation matching of the ARM as an indicator as to which of the fracture
model nearly all the fluid flow occurs through the more tortuous models are most appropriate for the ARM system.
upper channel. This pressure mismatch is exacerbated during the
later stages where the fluid injection rates are increased. Table 3 The first fracture relative permeability model used was proposed
shows that the oil recovery factor of the initial simulation model by Romm4, where he suggests straight line relative permeability
was too high compared to the ARM, because the uniform sweep curves are the most appropriate for modelling fluid flow through
predicted by the simulation model did not occur in reality. fractures (see Fig.18). Rossen and Kumar5 proposed another
model that accounts for the impact that gravity segregation and
Fracture Model The results of the initial model clearly show that asperites have on wetting-phase flow along the fracture walls (see
the preferential flow of fluid along the lower channel was unable Fig.19). Another model considered was the one proposed by
to be captured, and via the pressure differential that the average Babadagli & Ershagi6, where effective fracture relative
permeability of the system was consistently too low. Based on permeability (EFRP) curves are calculated based on an imbibition
this information and the fluid flow patterns observed in the lower index. Unfortunately, their relative permeability model was not
channel it was realised that the fractures measured in the CT able to be extended to deal with the very high permeabilities and
scans were dominating the production performance of the ARM. low flow rates present in the ARM. However, they found that
Therefore, to achieve a history match the measured fracture with an increase in velocity and viscous forces there was an
system needs to be incorporated into the simulation model. increase in the effective relative permeability of water. These
finding were further extended by the pore-scale modeling work of
Modelling the fracture system was a laborious task which Hughes and Blunt7, which demonstrated increases in the
involved mapping the fracture system by interpolating in between maximum wetting phase saturation (krwmax) and reductions in non-
the CT scan sections and identifying the equivalent grid blocks in wetting phase residual saturations (Snr) with increases in capillary
the simulation model. Most of the fractures were vertical, parallel number. Fig. 20 verifies their findings showing an increase in
with the lower channel orientation, and concentrated mostly effective permeability at the higher injection rates during the
around the first, deeper connection area. The final fracture system ARM experiment. However, a complication in using these models
adopted in the simulation model is depicted in Fig. 17. The is that the relative permeability curves would theoretically need to
fracture cells are assigned an effective porosity of 1, and their be adjusted in time depending on the injection rates.
own saturation region (SATNUM) in order to model various
fracture relative permeability curves. The pore volumes of the
fractures were modified to model the aperture of the fractures.
Although there is not a precise measurement of this parameter, it
is estimated to be between 1 to 2 mm. In order to preserve the
hydrocarbon pore volume of the system, the pore volume
removed from the fracture cells are placed back into the matrix
cells of the CIPS porous media adjacent to the fractures.

Fig. 20 Romm straight line fracture relative permeability curves.

Fig. 19 Fracture pattern used in simulation models.

Future history matching runs are all conducted with the fracture
network in place in order to try to model the preferential flow of
fluid through the lower channel. The permeability of the fracture
is a key history matching parameter, but another important one is
the relative permeability model to be applied to the fracture
network. A literature search yielded a number of relative
permeability models for modelling multi-phase flow for fractured Fig. 21 Rossen & Kumar fracture relative permeability curves.
8 SPE 101880

A number of initial sensitivities were run and it was found that the depletes more rapidly than in reality. It then takes some time for
injected fluid continued to flow across the first intersection point the injected water to pressurise the water in the base of the
and into the central area of the upper channel despite the channel and to get the water cut to increase.
incorporation of the fracture models and regardless of the
permeabilities used. However, it was found that there was a limit 100%
on the maximum fracture permeability that could be used before 90%
ARM Water Cut
Rossen & Kumar_no OWC
the simulator encountered convergence problems. This was Romm_no OWC
80% Extreme krw_no OWC
observed to be related to the effective aperture of the fracture. The Romm_with OWC
Extreme krw_with OWC
smaller the fracture aperture (pore volume) used, the lower the 70%

maximum permeability. For example: 60%

W ater Cut
Maximum permeability of 1000D, assuming a fracture 50%
aperture of 2mm. 40%
Maximum permeability of 100D, assuming a fracture 30%
aperture of 1mm.
20%
This is interpreted to be resulting from the smaller pore volume
more aggressively channeling the injected water through the 10%

model, generating larger saturation discontinuities and causing 0%


0 20 40 60 80 100 120
numerical instability at the default tolerances of the simulator.
Time (Minutes)

A number of initial sensitivity runs of the fracture relative


Fig. 23 Comparison of water cut matches of lower channel for
permeability curves were conducted on the ARM without an fracture model runs.
OWC in order to determine the relative permeability model that
was most appropriate for modelling fracture flow. The Rossen &
Kumar relative permeability curve had to be modified in order to
run in the simulator. Fig.21 summarises the results and found that
the Rossen & Kumar model was the too conservative and did not
model any water production in the lower channel. The straight
line Romm relative permeability model performed significantly
better. An additional sensitivity was run with an extreme fracture
krw curve as depicted in Fig. 24 and it gave the best results.

Fig. 24 Extreme fracture relative permeability curves.

The decrease in pressures and liquid rates observed in the upper


channel is still not matched, but as can be seen by comparing
Figs. 16 and 28, the migration of water into the upper channel has
been significantly reduced with the introduction of the fractures.
The pressure match has improved considerably from the non-
fracture case, but is still much higher than the observed values.

The history matching results are explained by the fact that the
Fig. 22 Relationship between permeability and water injection rate. simulator continues to model significant volumes of fluid flow
across the first connection area and into the upper channel. The
The fracture relative permeability curves were then applied to the fracture model does not provide enough conductivity in the
model with an initial OWC. The Rossen & Kumar model resulted fractures to preferentially flow water into the lower channel,
in convergence problems, and because it yielded the worst match resulting in slow water cut development in the lower channel and
was then discarded from future analyses. The extreme krw model consistently high pressure drops across the model. Comparing
still yielded the best match with regard to the water cut Figs. 29 and 6 (end Stage 2) reveals that the sweep predicted by
development in the lower channel and its results are presented in the simulation model is too uniform, whereas the video footage
the analysis of the fracture modelling run. clearly demonstrates the fingering of water through the fractures.
The CIPS porous media remains initially oil saturated and
Figs. 25 to 30 depict the results of the facture model simulation gradually water imbibes into this strongly water-wet rock. Once
run, and although the history match has improved, overall the the lower channel of the ARM is shut-in, water continues to flow
quality of the match is still not good. During Stage 1, the water through the fractures in the lower channel. However, Fig. 29
breakthrough in the lower channel is too early and then drops off shows that the bulk of the simulated flow is in the upper channel.
because the volume of water in the base of the lower channel
SPE 101880 9

0.020 100% 0.020 100%

0.018 ARM_Oil Rate 90% 0.018 90%


Simulated Oil rate
0.016 ARM_Water Cut 80% 0.016 80%
Simulated Water Cut ARM_Oil Rate
0.014 70% 0.014 70%
Simulated Oil Rate

Oil Rate (m 3 /d)


ARM_Water Cut
Oil Rate (m3/d)

0.012 60% 0.012 60%

W ater Cut
Simulated Water Cut

Water Cut
0.010 50% 0.010 50%
Lower channel shut in
0.008 40%
0.008 40%

0.006 30%
0.006 30%
0.004 20%
0.004 20%
0.002 10%
0.002 10%
0.000 0%
0.000 0%
0 50 100 150 200 250 300
0 20 40 60 80 100 120 140
Time (Minutes)
Time (Minutes)

Fig. 25 Lower channel history match; fracture model. Fig. 26 Upper channel history match; fracture model.

2.5

ARM dp
Simulated dp
2
Pressure (bara)

End Stage 3

1.5

End Stage 2
End Stage 1
0.5

0
0 50 100 150 200 250 300
Time (minutes)

Fig. 27 Pressure history match; fracture model. Fig. 28 Simulation model at end Stage 1; fracture model.

Fig. 29 Simulation model at end Stage 2; fracture model. Fig. 30 Simulation model at end Stage 4; fracture model.

Table 3 also shows that the recovery factor from this fracture Since explicity modeling the fractures was unable to yield a
relative permeability model is too high, and is consistent with the satisfactory history match it was decided to use the fracture
previous observations that the simulated sweep efficiency is not modelling option in Nextwell and simulate the ARM as a dual
representative of the physical observations of the ARM. permeability system.
10 SPE 101880

Dual Permeability Model TABLE 3 Recovery Factor Summary


In this model, matrix-matrix flow is possible and it is possible Model Initial Cum Oil Recovery
to control the fracture-matrix interaction through the sigma co- OIP Production Factor
effiecient (). The dual permeability fracture model is based on (L) (L) (%)
Total at end of experiment
the work of Kazemi and is described in the Nextwell Technical ARM 1.65* 0.85 51
Manual8. In order to model the preferential flow of water through Initial Run 1.65 1.41 85
the fractures it was decided to model it with minimal matrix- Fracture Model 1.65 1.37 83
fracture interaction , i.e.(=0). Dual permeability Model 1.65 1.32 80

Lower Channel at end Stage 1


Fig. 31 shows the extreme krw curve has a slight advantage over ARM 1.07* 0.50 47
the Romm curve so the extreme krw curves were also used for the Initial Run 1.07 0.84 78
dual permeability model. There was the same issue with the Fracture Model 1.07 0.74 69
Dual permeability Model 1.07 0.62 58
previous fracture model in that there was a maximum
permeability that could be used in the simulator without the Upper Channel at end Stage 1
computer encountering convergence problems. In this case it was ARM 0.58* 0.15 26
100,000 D. The maximum fracture permeability was also Initial Run 0.58 0.47 81
Fracture Model 0.58 0.35 60
dependant on the effective pore volume of the fracture, i.e. the Dual permeability Model 0.58 0.25 43
smaller the pore volume, the lower the maximum fracture * Estimated
permeability.
Seismic Modelling
Integral to the ARM program was the recording of scaled time-
lapse 3D seismic surveys that were designed to provide insight
into the seismic expression of turbidite channels, which are
typically at the limit of, or below, seismic resolution in the deep
water environment. Details of the seismic program have been
published elsewhere1, but a brief description is included here.

Two sets of scaled time-lapse 3-D seismic data were acquired


before and after production at frequencies of 50kHz and 1MHz.
Careful attention was paid to scaling the dimensions and physical
properties of the model so that the high-frequency response
observed in the laboratory would represent the field scale
environment. 1MHz data was recorded to provide a high-
resolution image of the channels and accurate monitoring of the
oil and water distribution. However, data was also recorded at
50kHz, which is an order of magnitude lower frequency than is
Fig. 31 Comparison of water cut matches of lower channel for dual traditionally used in the laboratory. This frequency was chosen to
permeability model runs.
represent field resolution seismic data and honours the correct
scaling of the relative wavelength to channel size for a typical
Figs. 32 to 37 depict the results from the dual permeability model
turbidite reservoir. At this frequency, the individual channels are
and they show that the simulated water cut development in the
within tuning thickness and this allows for valid comparisons
lower channel is the most aggressive to date and the distribution
with seismic attributes in field data from similar reservoirs.
of water at the end of Stage 1 is the closest match to the ARM
Analysis of frequency dependent velocity dispersion mechanisms
response. The previous models show a relatively even sweep of
shows that the response to changes in saturation during
the lower channel, but in the dual permeability model the
production observed at 50kHz in the laboratory is representative
distribution of water saturation is more controlled by the fracture
of the expected seismic response at 50Hz in the field.
network. Table 3 shows the significant improvement in the
recovery factor match of the dual permeability model at the end
The distribution of the oil and water within the individual
of Stage 1. However, after the lower channel is shut-in (Stage 2
channels is not resolvable in simple maps of the changes in
onwards), the simulated behaviour of the dual permeability model
seismic amplitude from pre- to post-production. However, a
becomes similar to that of the previously described fracture
more comprehensive analysis of a range of seismic attributes is
model. This is because the dual permeability model is also unable
able to reveal some insight into the relative distribution of oil and
to model the preferential flow of water through the fractures in
water zones in the ARM. Fig. 38 is a map of the change in slope
the lower channel once production from the lower channel ceases.
of reflection strength from a time window that spans both
This also explains the results in Table 3 which show that by the
channels. The water-flooded zones within the ARM are
end of the experiment the cumulative production and recovery
highlighted as a result of the increased impedance contrast. This
factors are similar.
attribute map is consistent with the known water swept zones
within the model. In the absence of visual confirmation of the
SPE 101880 11

0.020 100% 0.020 100%

0.018 ARM Oil Rate 90% 0.018 90%

Simulated Oil Rate


0.016 80% 0.016 80%
ARM Water Cut
0.014 Simulated Water Cut 70% 0.014 70%
ARM_Oil Rate

O il R a te (m 3 /d)
0.012 60% 0.012 Simulated Oil Rate 60%
Oil Rate (m3/d)

W ater C u t
Water Cut
ARM_Water Cut
0.010 50% 0.010 Simulated Water Cut 50%

0.008 40% 0.008 40%


Lower channel shut in
0.006 30% 0.006 30%

0.004 20% 0.004 20%

0.002 10%
0.002 10%

0.000 0%
0.000 0%
0 20 40 60 80 100 120 140 0 50 100 150 200 250 300
Time (Minutes) Time (Minutes)

Fig. 32 Lower channel history match, dual permeability model. Fig. 33 Upper channel history match, dual permeability model.

2.5

ARM dp
Extreme krw
2
Pressure (bara)

1.5

0.5

0
0 50 100 150 200 250 300
Time (minutes)

Fig. 34 Pressure history match, dual permeability model. Fig. 35 Simulation model at end Stage 2; dual permeability model.
.

Fig. 36 Simulation model at end Stage 2; dual permeability model. Fig. 36 Simulation model at end Stage 4; dual permeability model.

water distribution, which is a luxury of ARM that is not available Way Forward
in the field, this kind of analysis of time lapse seismic data can be Results from this proof of concept ARM are limited in this early
a powerful aid to refining and reducing the non-uniqueness that is stage of development and are clearly compromised by some
inherent in history matching. In the case of this ARM it would problems that were encountered during the fabrication process.
prove invaluable because the ambiguity of the history matching However, through lessons learned from this process and
results could mask the fact that the central part of the upper subsequent testing of different ways to improve the geological
channel was unswept and the site for a potential infill well realism of channel fill lithologies, we are now in a position to
location. reduce many of the limitations of the model reported here.
12 SPE 101880

It is worth discussing the impact of the shortcomings of the finite


difference simulation to fully model the conductive nature of the
fractures and the impact the behaviour of the ARM. In particular,
is the inability of the simulator to model the initial 20 minutes of
water free production, despite the proximity of water in the base
of the channel. The current theory is that because there is oil
initially flowing from the upper channel into the fractures of the
lower channel that this is supporting the initial oil production in
the lower channel. This hypothesis suggests that due to a
combination of gravity and viscous forces generated by flow from
the production outlet that a segregated flow of oil is occurring
along the top of the lower channel. Therefore, in this scenario oil
is not flowing along the water filled base of the lower channel,
and it is possible that the water may be acting as a dead volume.

It is proposed that the water in the base of the lower channel does
Fig. 38-Post-production seismic attribute map (slope of reflection
strength). not flow until the injected water from the upper channel flows
into the fractures of the lower channel. This then provides flow
It is proposed to developed this technology further through and pressure support to the basal water volume in the lower
improvements in the model construction process, and with more channel. From the video footage taken from the top of the model
precise ground truthing of the physical properties and the nature it is uncertain as to when the injected water enters the lower
of the fluid distribution. Thee improvements are expected to yield channel. However, it is possible that it enters the lower channel
valuable insights into issues of uncertainty regarding the reservoir around the 20 minute mark, which is when the ARM starts to
simulation and seismic expression of complex reservoirs. produce water. The inability of the finite difference simulator to
model the highly conductive nature of the fractures results in the
The extension of the reservoir simulation aspect of the project is hump observed in the simulated water production profile. It is
to research alternatives on how to improve the history match of possible that the observed differences are because nature is
the ARM. It is planned to investigate other simulation modelling smoother and more conductive than can be captured in the more
techniques such as streamline and finite element methods. discontinuous world of numerical modelling. From these
Logistics permitting, the history matching performance of other observations we can ask; is there is a technique or tool currently
simulators and algorithms will be analysed. Also to be available that can numerically model the behaviour of the ARM,
investigated is the impact of re-gridding the ARM such as and what are the potential impacts on predicting production
upscaling, grid orientation effects, and the use of PEBI (Voronoi) performance at the field scale?
grids and local grid refinements (LGRs).
Acknowledgements
Conclusions and Discussion The authors would like to thank the Curtin Department of
This paper summarises the initial findings from the reservoir Petroleum Engineering and CSIRO Petroleum for supporting this
simulation work conducted for the ARM project. It was found project, in particular Geoff Weir for assistance in the preparation
that in order to best model the highly conductive fractures in the of this manuscript. Also thanks to Roxar Australia for providing
ARM, the following was required: the simulation software and to Robert Frost for valuable technical
1. Using the highest fracture permeability as possible before support. The funding and support provided by Chevron and
encountering convergence problems in the simulator. Woodside for the ARM project is greatly appreciated. Also thanks
2. Modelling high water mobility in the fracture, using an extreme to Geovisual and Brennan Williams for providing the 3-D
krw fracture relative permeability curve. visualisation software, and to Corelab Australia for providing the
3. Using the dual permeability fracture model, with minimal core analysis data and fluids.
interaction between the matrix and the fracture.
References
Despite incorporating all these recommendations into the 1. Sherlock, D., Scoby-Smith, L. and Montague, E.: Time-lapse
simulation model, a reasonable history match against the ARM Analogue Reservoir Modelling of Turbidite Channel Sands,
still could not be achieved. A consistent observation in all the Exploration Geophysics (2005) Vol.36, No.2.
models was their inability to simulate the preferential flow of 2. Sherlock, D.H., and Siggins, A.F., The development of
water through the fractures in the lower channel once production synthetic CIPS sandstones for geophysical research: 15 th t

from the lower channel had ceased (Stage 2 onwards). Clearly the Geophysical Conference and Exhibition, Australian Society of
fractures are highly conductive and provide a path of least Exploration Geophysicists, Expanded Abstracts. 2003
resistance for the flow of fluid; however this behaviour is unable 3. Kucharski, E., Price, G., Li, H., and Joer, H.A., Laboratory
to be modelled numerically. To understand the differences evaluation of CIPS cemented calcareous and silica sands: 7th
between the physical and numerical responses will be a focus area Australia New Zealand Conference on Geomechanics, 1996, 102
for future study. 107.
4. Romm, E.S.: Fluid Flow in Fracture Rocks, Nedra
Publishing House, Moscow, 1966.
SPE 101880 13

5. Rossen, W.R. and Kumar, A.T.A., Effect of Fracture Relative


Permeabilities on Performance of Naturally Fractured
Reservoirs, paper SPE 28700 presented at the 1994 SPE
International Petroleum Conference and Exhibition, Veracruz,
Mexico, October 10-13.
6. Babadagli, T. and Ershagi, I., Improved Modelling of
Oil/Water Flow in Naturally Fractured Reservoirs Using Effective
Fracture Relative Permeabilities, paper SPE 26076 presented at
the 1993 Western Regional Meeting, Alaska, May 26-28.
7. Hughes, R.G. & Blunt, M.J., Pore-Scale Modelling of
Multiphase Flow in Fractures and Matrix/Fracture Transfer,
paper SPE 56411 presented at the 1999 SPE Annual Technical
Conference and Exhibition, Houston, Texas, October 3-6.
8. Roxar Pty. Ltd, Nextwell User Guide, Version 5.0, 1991, p.
443-448

Appendix 1 Equations and Nomenclature

'
k rw w
M = Equation 1
k ro' o
1
fw = Equation 2
k
1 + w . ro
k rw o
Nw
S S wc
k rw = k w
'
rw
Equation 3
1 S wc S or
No
1 S w S or
k row = k
'
ro
Equation 4
1 S wc S or
M - End point mobility ratio.
krw - Water end-point relative permeability.
w- Water viscosity.
kro - Oil end-point relative permeability.
o - Oil viscosity.
fw - Fractional flow of water.
krw - Water relative permeability.
kro - Oil relative permeability in the presence of water.
Sw - Water saturation.
Swc - Connate water saturation.
Soc - Residual oil saturation.
Nw - Corey exponent for water.
No - Corey exponent for oil.

You might also like