SPE 106679

Permeability Upscaling Techniques for Reservoir Simulation
J. R. Villa, SPE, PDVSA Intevep; M. O. Salazar, SPE, Universidad Central de Venezuela
Copyright 2007, Society of Petroleum Engineers, Inc.
This paper was prepared for presentation at the 2007 SPE Latin American and Caribbean
Petroleum Engineering Conference held in Buenos Aires, Argentina, 15-18 April 2007.
This paper was selected for presentation by an SPE Program Committee following review
of information contained in an abstract submitted by the authors(s). Contents of the paper,
as presented, have not been reviewed by the Society of Petroleum Engineers and are sub-
ject 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 pre-
sented at SPE meetings are subject to publication review by Editorial Committees of the
Society of Petroleum Engineers. Permission to copy is restricted to an abstract of not more
than 300 words. Illustrations may not be copied. The abstract should contain conspicuous
acknowledgement of where and by whom the paper was presented. Write Librarian, SPE,
P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., Fax 1-214-952-9435.
Abstract
Upscaling reservoir properties for reservoir simulation is one of
the most important steps in the workflow for building reservoir
models. Upscaling allows taking high-resolution geostatistical
models (10
7
-10
8
grid blocks) to coarse scale models (10
4
-10
5
grid blocks), manageable for reservoir simulation, while re-
taining the geological realism and thus effectively representing
fluid transport in the reservoir
1,2
. This work presents a study of
the effectiveness of different available techniques for permea-
bility upscaling and the implementation of a new technique for
upscaling of relative permeability curves based on the nume-
rical solution of a two-phase system and the Kyte and Berry
method
3
.
The reference fine scale model considered in this study is a con-
ceptual fluvial reservoir based on the Stanford V model
4
. The
reference fine scale isotropic and locally heterogeneous per-
meability distribution was upscaled to different upscaling ratios
by means of analytical (static) and numerical single-phase (pre-
ssure solver, dynamic) techniques. Two-phase flow simulations
were performed on the reference fine grid and upscaled models
using a comercial black-oil simulator. Arithmetic, harmonic,
and geometric averages were defined for static upscaling of
the permeability distribution. The dynamic upscaling process
considered one-phase and two-phase upscaling. One-phase
upscaling considered upscaling of the permeability distribution
and two-phase upscaling considered upscaling of the permea-
bility distribution and relative permeability curves.
Flow simulation results for waterflooding in the coarse scale
model indicated relevant discrepancies with the fine grid re-
sults. Compared to fine-scale, flow results of the single-phase
upscaling process indicated that the coarsest upscaled models
did not match the water breakthrough times, water cut values,
or well pressures from the reference model. The finer upsca-
led models reproduced the reference results more accurately
than the coarser models. The two-phase dynamic upscaling
technique implemented in this work resulted in the best match
with the flow simulation results of the fine grid model. Results
show that the most accurate upscaling scheme should be de-
fined using the two-phase dynamic upscaling technique on the
model with the smallest upscaling ratio.
Introduction
Reservoir models generated by geostatistical techniques, high-
resolution fine scale models (10
7
-10
8
grid blocks), are capa-
ble of with great precision reservoir characterization as for
compartmentalization, heterogeneity, connectivity and struc-
ture. However, the main drawback of high-resolution models
is the significant computational cost when performing reser-
voir simulation. Upscaling reservoir properties allows taking
high-resolution models to coarse scale models (10
4
-10
5
grid
blocks) reducing computational costs during flow simulation
for history matching and forecast. Permeability upscaling plays
and important role in reservoir characterization
5
, as shown in
Figure 1. The importance of using an appropriate upscaling
technique consists in preserving the geological realism of high-
resolution, fine-scale models, thus preserving the flow response
in reservoirs
1,2
. Permeability upscaling is an active research
topic, numerous studies on upscaling have been conducted by
university researchers
6,7
and industry
2,8
.
In this work, the effectiveness of different permeability upsca-
ling techniques is evaluated using reservoir simulation. Ana-
lytical and numerical single-phase upscaling techniques were
used with different upscaling ratios for a conceptual fluvial
reservoir. Public available software was used for this purpose.
In addition, an alternative technique based on two-phase nume-
rical upscaling was developed and implemented. Using these
techniques, flow simulation results of upscaled models were
compared with the reference fine scale model in terms of flow
2 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
production, well pressure and saturation distribution.
The sensitivity study of different permeability upscaling tech-
niques is referred to a conceptual fluvial reservoir based on the
Stanford V model
4
. As much for the fine scale model as upsca-
led models the numerical simulation of an oil-water inmiscible
displacement process was effected, specifically a waterflooding
process. This work includes the evaluation of static upscaling
techniques (arithmetic, harmonic, and geometric) and dynamic
upscaling techniques (one-phase and two-phase). The fine scale
reference model was upscaled to four different upscaling ratios.
This work allows to understand the impact of using different
upscaling techniques on the simulation results. Investigation on
gridding techniques was not considered in this study.
Theory
Darcy’s Law Darcy’s law express the relationship between
fluid velocity and pressure gradient in a porous media. The
Darcy velocity can be written in matrix notation for a Cartesian
system (neglecting gravity) as:
u = −
1
µ
k · ∇p (1)
In Equation 1, u is the velocity vector, k permeability tensor
and ∇p the pressure gradient.
Permeability Tensor The permeability of the porous media is
a property that can vary at any point and on any direction in the
three-dimensional space
5
. It is mathematically represented by
the full permeability tensor k (Equation 2). Each component of
the permeability tensor represents the directional permeability
at one point in space. The permeability tensor is usually taken
to be locally symmetric (k
ij
= k
ji
).
k =

¸
k
xx
k
xy
k
xz
k
yx
k
yy
k
yz
k
zx
k
zy
k
zz
¸

(2)
If there is an orientation such that u and ∇p are parallel for
a full tensor k, the principal orientation of permeability is ob-
tained. This leads to the principal values and directions of the
permeability tensor (eigenvalues and eigenvectors):
k

=

¸
k

xx
0 0
0 k

yy
0
0 0 k

zz
¸

(3)
The tensor k

is the diagonal permeability tensor.
Governing Equations In single-phase incompressible flow,
neglecting gravity and capillary effects, the conservation Equa-
tion is written as:
∇· u = 0 (4)
using the Darcy’s law, Equation 4 becomes:
∇· (k∇p) = 0 (5)
In two-phase incompressible flow, Darcy’s velocity is written
as:
u
j
= −
k
rj
µ
j
k · ∇p (6)
where, k
rj
is the relative permeability of phase j, and µ
j
is the
viscosity of phase j. The conservation equation is written as:
∇· u
t
= 0 (7)
subsituting Equation 6 in Equation 7 yields:
∇(λ
t
(s) k · ∇p) = 0 (8)
where λ
t
is the total mobility, defined as:
λ
t
(s) =
k
rw
µ
w
+
k
ro
µ
o
(9)
Classification of Techniques Upscaling techniques can be
classified in terms of the parameters to be upscaled. In
one-phase parameter upscaling technique, the fine-scale
permeability tensor (k) is upscaled to a coarse-scale effective
permeability tensor (k

) while retaining the fine-scale rela-
tive permeabilities. Analytic and numerical methods are used
for this purpose. On the other hand, in two-phase parameter
upscaling technique, fine-scale relative permeabilities (k
rj
) are
also upscaled to curves of different shapes (k

rj
). These curves
are usually referred as effective or pseudo-r elative permeabi-
lity curves and their generation is accomplished by numerical
methods. Figure 2 illustrates these upscaling techniques.
Analytic methods Analytic methods for computing one-
phase parameter upscaling involve the solution of Equation 5
with no-flow boundary conditions in the non-communicative
layered system shown in Figure 3, with n horizontal layers of
permeability k
i
and dimensionless thickness h
i
. Flow in x di-
rection is referred as parallel flow and upscaled permeability
results in the arithmetic mean of permeability values in each
layer:
k

xx
=
n
¸
i=1
k
i
h
i
(10)
Flow in the z direction is referred as series flow and results in
the harmonic mean of permeability values in each layer:
k

zz
=

n
¸
i=1
h
i
k
i

−1
(11)
For this 2D system, the effective permeability tensor is:
k

=

k

xx
0
0 k

yy

(12)
SPE 106679 J. R. Villa, M. O. Salazar 3
where k

xx
and k

yy
are the principal values of the permeability
tensor. Although permeability values were locally isotropic,
effective permeability values in Equation 12 are constant and
anisotropic, meaning that when upscaling fine-scale permea-
bility values, coarse-scale permeabilities are anisotropic even
with isotropic fine-scale permeability.
An estimate for systems with spatially random permeability can
be obtained using the geometric average of permeability values
where there is no particular assumption of flow direction:
k
g
= exp
1
n
n
¸
i=1
log k
i
(13)
In systems with constant h
i
, Equation 13
9
can be also ex-
pressed as:
k
g
=

n
¸
i=1
k
i
1
n
(14)
A generalization of these averages techniques is known as the
Power Law
10
. This empirical relation is written as:
k
ω
=

1
n
n
¸
i=1
k
ω
i

1/ω
(15)
The Power Law allows to determine the effective permeability
k
ω
of a set of volume elements for different values of ω ranging
between -1 and 1. For values of ω = 1, 0, −1, Equation 15
results in arithmetic, geometric and harmonic averages, respec-
tively. In general, arithmetic average provides an upper bound
to k

, and the harmonic average provides a lower bound. Fi-
gure 4 shows the effective permeability computed for differents
values of ω in a heterogeneous system
10
.
Numerical methods Numerical methods involve the solu-
tion of Equation 5 or Equation 8 for the pressure distribution
using finite-difference methods. In one-phase dynamic upsca-
ling, each component of the diagonal permeability tensor is cal-
culated separately depending of the flow direction. To solve
the pressure distribution, arbitrary boundary conditions are as-
signed in the target coarse grid block (Figure 5) and the pre-
ssure of each fine grid block inside the target coarse grid block
is computed by the solution of the single phase incompressible
flow steady state Equation 5. This equation can be expressed in
finite difference form and written as a matrix equation as:
Tp = b (16)
where, T is the transmissibility matrix, b a vector representing
the source / sink term and p the unknown pressure vector. Once
the pressure vector is calculated, the effective permeability of
the target coarse grid block in the x direction (k

xx
) is computed
as follows:
k

xx
=
n
x
n
y
n
z
(p
in
−p
out
)
nz
¸
k=1
ny
¸
j=1
k
1jk
(p
1jk
−p
in
) (17)
where, n
x
, n
y
, and n
z
are the number of fine grid blocks in
the x, y, and z direction respectively, p
in
is the pressure in the
inlet of the coarse gridblock, p
out
is the pressure in the outlet of
the coarse grid block, k
1jk
is the absolute permeability in each
fine grid block next to the inlet, and p
1jk
is the pressure in each
fine grid block next to the inlet. The terms, k

yy
and k

zz
can be
obtained in a similar fashion.
In two-phase dynamic upscaling, it is recognized that it is not
enough to upscale the absolute permeability to characterize
transport in porous media under inmiscible displacement
processes
2
. Therefore, the fine-grid relative permeability curves
(k
rj
) are upscaled to different curves (k

rj
) and thus the fluid-
rock interaction in the coarse-scale model is considered. This
kind of upscaling performed by generating pseudo-relative per-
meability curves allows to better represent fluid flow when the
fine-grid relative permeability curves function are not able to
represent. The main methods for generating pseudo-relative
permeability functions
11
are Kyte and Berry method, Stone
method, weighted porous volume method, weighted relative
permeabilities method, and the Kirchoff’s Law method. All
these methods use numerical flow simulation results on the
high-resolution fine-scale model to generate the relative per-
meability curves of the coarse-scale model.
Upscaling ratio The upscaling ratio is defined as:
r =
n
N
(18)
where, r is the upscaling ratio, n the number of fine grid blocks,
and N the number of coarse grid blocks. The upscaling ratio
represents a measure of how coarse is the coarse model. The
larger upscaling ratio the coarser the upscaled model. Figure 6
shows two coarse models at different upscaling ratios.
Methodology
The methodology propossed for this work involves the use of
analytic and numerical methods to upscale a fine-grid model to
four different upscaling ratios. Figure 7 illustrates the metho-
dology followed in this work. Flow simulations are performed
on the reference model and on the coarse-scale models and
comparisons are made based on the calculated error. Analytic
upscaling and single-phase dynamic upscaling of the fine-
grid model was performed using the Fortran-based program
flowsim
12
. For two-phase dynamic upscaling, a new Fortran-
based program flowsim2p was coded for this porpuse, following
a similar structure to GSLIB suite of programs
12
. Porosity ups-
caling was performed using program upscaler
12
and program
gsl2ecl
12
was used for output compatibility to the black-oil
reservoir simulator ECLIPSE
13
. An automated workflow and
post-processing of results were implemented using the applica-
tion MATLAB
14,15
.
4 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
flowsim2p The computational tool developed in this work and
termed flowsim2p was coded in Fortran90
16
and implements the
Kyte and Berry method for generating pseudo-relative permea-
bility curves
11,17
. In the Kyte and Berry method, the fine-scale
properties are weighted to obtain coarse-scale properties and
later used in the Darcy’s law to obtain pseudo-relative permea-
bilities curves corresponding for each phase (j) and for each
coarse-grid block:
k

rj
= −
µ

j
q

j
T

∆p

j
−∆ρ

j
g∆D

(19)
where, k

rj
is the pseudo-relative permeability of a coarse-grid
block for phase j, µ

j
the phase viscosity of a coarse-grid block,
q

j
the phase rate in one specific direction associated to a coarse
grid block, T

the weighted transmissibility between a coarse
grid block and its neighboring grid block, ∆p

j
the pressure
difference between the target coarse grid block and its neigh-
boring grid block, ρ

j
the phase density associated to a coarse
grid block, g the gravity term, and ∆D

the thickness diffe-
rence between the target coarse grid block and its neighboring
grid block.
As shown in Equation 19, time-dependent phase pressure and
saturation is computed in each fine-grid block inside the target
coarse-grid block. Some drawbacks associated with the Kyte
and Berry method can be found
18,19
. These include: proble-
matic gridblocks due to flow restrictions, unability to generate
flow direction-dependent curves, and high computational cost
for global boundary conditions. In flowsim2p, an inmiscible
oil-water displacement in a target coarse grid block is formu-
lated and specific boundary conditions are imposed in the target
coarse grid block (Figure 8). The displacement process is fo-
cused locally and the global boundary conditions are ignored,
the oil phase has no dissolved gas, the gravitational and capi-
lar effects are negligible, the rock and fluid compressibilities
are dismissed, series flow is assumed between neighboring fine
grid blocks, the fluid viscosities are taken constant, and the
duration time of the process is variable depending of the coarse
grid block size.
Figure 9 shows the workflow of program flowsim2p. First,
the input data is stored, then the program computes the pre-
ssure and saturation distribution of each phase inside the target
coarse grid block, and finally the pseudo-relative permeability
curves are generated. This process is repeated a certain time
step number depending of the coarse-grid block size.
Reference fine-scale model In this work, the fine-scale model
used is a conceptual fluvial reservoir based on the Stanford V
model
4
. This is a Cartesian model with 130,000 grid blocks
(100x130x10) as shown in Figure 10. The heterogeneous
porosity and locally isotropic permeability distributions were
generated using nonconditional sequential Gaussian simulation
20,21
. The reservoir has four oil production wells and one water
injector well. Permeability and porosity distributions are shown
in Figures 11 and 12, respectively.
Cases Four cases were defined in this work. Each case co-
rresponds to a different upscaling ratio. Different upscaling
techniques were used for all cases. Table 1 shows the diffe-
rent cases proposed in this study. Case 1 is the finest upscaled
model with 50x65x10 (32,500 grid blocks), Case 2 has 50x26x5
(6,500 grid blocks), Case 3 has 10x13x5 (650 grid blocks) and
Case 4 is the coarsest upscaled model with 5x5x5 (125 grid
blocks). Upscaling ratios r are 4, 20, 200, and 1,040 respec-
tively.
Results and Discussion
A waterflooding process was performed in the fine-grid model
and in each coarse-scale model using the black-oil reservoir
simulator ECLIPSE
13
. This program is a fully-implicit, three
dimensional, three phase, general purpose black-oil reservoir
simulation. There are four vertical producing wells located at
the reservoir boundaries. Water is injected through one ver-
tical wells located in the middle on the reservoir. All wells
are open to flow over the entire thickness. All wells exhibit
no formation damage. Well constraints include maximum oil
production of 20,000 STB/d and maximum bottom-hole pre-
ssure of 1050 psia at the injection well. No restriction were
defined to the field production or well water cut. Simulation
results were post-processed for analysis and visualization using
MATLAB
14
. First, discussion of numerical simulation results
is presented and later, an evaluation of the upscaling errors is
considered.
Simulation results Figures 13 to 17 shows the permeability
field and water saturation distribution for different upscaling
ratios (r) and at the lower layer of each corresponding model.
In the case of two-phase dynamic upscaling (Figure 17), flow
simulation of the coarse-scale models was performed with the
correspondent pseudo-relative permeability curves. It is impor-
tant to emphasize that the end-points of the pseudo-relative per-
meability curves are the same as the end-points of the relative
permeability curves of the fine-scale model, since displacement
efficiency of the reference fine-scale model is preserved
3
. The
increase in the upscaling ratio, the difference in shape between
the pseudo-relative permeability curves and the fine-scale re-
lative permeability curves. When upscaling absolute permea-
bility, the capacity of capturing heterogeneities is lost as the
upscaling ratio increases, and the saturation profile loses repre-
sentativety with respect to the reference fine-scale model. It is
precisely the shape of the pseudo-relative permeability curves
that compensates this lost of representativity. Also, this diffe-
rence between the pseudo-curves and the original curves does
not indicate that the pseudo-curves are better or worst, this fact
only indicate the upscaling ratio used. For the cases of arith-
metic, harmonic, geometric and one-phase dynamic upscaling
(Figures 13 to 16) it is maintained the tendency of losing the
representativity on the permeability field and saturation profiles
as the upscaling ratio is increased.
SPE 106679 J. R. Villa, M. O. Salazar 5
Figures 18 to 22 show the bottomhole pressure and water cut at
the four production wells for the differents techniques and ups-
caling ratios. Each well have early breakthrough times respect
to the reference fine-scale model. Performance of the bottom-
hole pressure response is also. This difference less marked in
the case of two-phase dynamic upscaling (Figure 22), which
is the technique that presents the best fit of the flow response
by well respect to reference fine-scale model. On the contrary,
the harmonic static upscaling (Figure 19) is the technique that
generates the flow response less representative.
Figures 23 and 24 show the responses of field water cut and cu-
mulative oil production, corresponding to different techniques
and upscaling ratios. Once again, it can be observed how as
it increases the upscaling ratio the flow responses become less
representative of what is happening in the reference fine-scale
model. For instance, in the cumulative oil production graphics
for the models of greater upscaling ratios the technique of har-
monic upscaling begins to fail remarkably on representing the
flow performance.
Figure 25 presents the CPU time required to perform the flow
simulation on the studied models. Here, one motivation of per-
forming upscaling is demonstrated by reducing the simulation
time decreasing the number of grid blocks of the model. In fact,
it can be observed that increasing on the magnitude order of the
number of grid blocks generate increasings on the magnitude
order of the simulation time. However, it should be established
a balance between the decreasing of the simulation time and
the representativity maintenance of the flow performance that
happens on the reference fine-scale model.
Upscaling errors This sections presents an evaluation of the
upscaling error computed for each technique used. Error is is
defined by Equations 20 and 21:
e =
¸
k

1 −
¯
Sw
k,r
Sw
k,r

n
r
(20)
and,
¯ s
w
k,r
=
¸
i
s
wi
n
r
, ∀i k (21)
where r is the upscaling ratio, s
wi
the water saturation of a fine
grid block i inside a grid block k, s
w
k,r
the water saturation of
a coarse grid block k at upscaling ratio r, ¯ s
w
k,r
the weighted
water saturation of fine grid blocks respect to a coarse grid
block k, n
r
the number of coarse grid blocks at upscaling ratio
r, and e the upscaling error. Equations 20 and 21 basically
establish the difference between the block saturation on the
upscaled model and the average fine-grid block saturations in
the corresponding coarse-grid block. Figure 26 illustrates the
procedure for the calculation of the upscaling error.
Figures 27 to 31 show the error maps in gray scale for each
one of the upscaling methods. Larger errors are located in grid
blocks on the water-oil interfase of the waterflooding process,
indicating that upscaling errors increment in blocks with partial
water and oil saturation. When two-phase dynamic upscaling
method is used, errors are significantly reduced on the oil-
water interfase (Figure 31), indicating that upscaling errors are
reduced in those grid blocks.
Table 2 shows the total upscaling errors in each case for diffe-
rent techniques and upscaling ratios. When two-phase dynamic
upscaling method is used in Case 1, the smallest upscaling error
is obtained. The harmonic static upscaling in the Case 4 ge-
nerates larger errors. Figure 32 shows the upscaling errors for
each case and method evaluated.
In general, analytic and numerical upscaling techniques can be
evaluated in terms of a total error. Figure 33 shows that nume-
rical techniques generate smaller upscaling error that analytic
techniques. The difference in the upscaling error can reach
up to 10%, for the studied model, when upscaling ratio is in-
creased.
Conclusions
Based on the oil and water production, well pressures, and
saturation results obtained from the simulation results using
dynamic and static upscaling, it can be see that by upscaling,
earlier water breakthrough times result. According to the repro-
duction of the reference water cut, the most accurate dynami-
cally upscaled result came from 50 ×65 ×10 model.
Numerical upscaling techniques provides better prediction than
does analytic upscaling. Dynamic upscaling will provide more
accurate results relative to static for a given set of boundary
conditions because different flow regimes can be accounted for
in a single model. Lastly, reliable future predictions can only be
obtained when geologic models depict the fine-scale case ac-
curately and precisely. Regardless of the upscaling method, the
averaging process will alter the original permeability field. The
degree of difference between the resulting effective permeabi-
lity and the true reference controls how different the grid block
pressure and water saturation, well water cuts and bottom-hole
pressures will be in the upscaled models relative to the fine
scale. When analytic techniques were used, geometric static
upscaling generated the best flow representativity. Upscaling
relative permeabilities showed an improvement in upscaling
results. Two-phase dynamic upscaling generated better flow
response than the one-phase dynamic upscaling.
An important aspect is the upscaling ratio, which determines
the accuracy of production predictability. With a high upsca-
ling ratio, the accuracy of the production prediction decreases.
There is a limit on how coarse a model can be without intro-
ducing significant errors. This limit is important particularly
when extreme permeability features are present in the reser-
voir, which directly affect the fluid flow.
Results show that there is no an optimal upscaling technique
suitable for any fine-scale model. For each fine-scale model,
6 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
a sensitivity study of different upscaling techniques with diffe-
rent upscaling ratios should be performed to determine the least
upscaling error technique.
Acknowledgements
The authors would like to thank Universidad Central de
Venezuela and PDVSA Intevep for the use of its computa-
tional facilities, and the Department of Petroleum Engineering
at Stanford University for the use of its computational tools.
Nomenclature
u = velocity vector
p = pressure vector
s = saturation vector
k = permeability tensor
k

= diagonal permeability tensor
T = transmissibility matrix
b = source/sink vector
∇p = pressure gradient
n = total number of fine grid blocks
N = total number of coarse grid blocks
ip = identificator number of fine grid blocks
k = absolute permeability of a fine grid block
h = thickness of a fine grid block
λ
t
= total mobility
ω = power parameter
k
ω
= effective permeability obtained by Power Law
k
g
= geometric average of permeability values
p
in
= pressure at the inlet of a coarse grid block
p
out
= pressure at the outlet of a coarse grid block
p
1,j,k
= pressure on fine grid blocks next to the inlet
k
1,j,k
= permeability on fine grid blocks next to the inlet
v
i,j,k
= volume of a fine grid block
∆t = timestep size
γ, m = Corey’s terms
A = cross-sectional area to the flow
References
1. Durlofsky, L. Advanced Reservoir Engineering PE222.
Stanford University, California, USA, 2002.
2. Christie, M. Upscaling for Reservoir Simulation. Jour-
nal of Petroleum Technology, November 1996.
3. Westhead, A. Upscaling for Two-Phase Flow in Porous
Media. California Institute of Technology, 2005.
4. Mao, S. and Journel, A. Generation of a Reference
Petrophysical-Seismic Data Set: The Stanford V Reser-
voir. Technical report, Stanford University, California,
USA, 1999.
5. Villa, J. R. Simulaci´ on de Yacimientos. Universidad Cen-
tral de Venezuela, Caracas, Venezuela, 2005.
6. Durlofsky, L. “Upscaling of Geocellular Models for
Reservoir Flow Simulations: A Review of Recent
Progress”. Paper presented at 7 International Forum on
Reservoir Simulation, Germany, June 2003.
7. Holden, L. and Nielsen, B. F. Global Upscaling of Per-
meability.
8. Stern, D. “Practical Aspects of Scaleup of Simulation
Models”. Paper SPE 89032, 2005.
9. Ahmed, T. and McKinney, P. D. Reservoir Engineering
Handbook. Gulf Professional Publishing, Houston, TX,
USA, 2001.
10. Journel, A., Deutsch, C., and Desbarats, A. “Power Ave-
raging for Block Effective Permeability”. Paper SPE
15128 presented at the 56 California Regional Meeting,
California, USA, September 1986.
11. Cao, H. Evaluation of Pseudo Functions. Master’s the-
sis, Stanford University, 1988.
12. Deutsch, C. and Journel, A. GSLIB - Geostatistical
Software Lybrary and User’s Guide. Oxford University
Press, New York, USA, 1998.
13. Schlumberger. ECLIPSE Reference Manual, 2003.
14. The Mathworks, Inc. MATLAB, The Language of Tech-
nical Computing, 2005.
15. Hanselmanz, D. and Littlefield, B. Mastering Matlab
6: A Comprenhensive Tutorial and Reference. Prentice
Hall, New Jersey, USA, 2001.
16. Chapman, J. FORTRAN 90/95 for Scientists and
Engineers. McGraw-Hill, New York, USA, 2005.
17. Inanc, O. A Sensitivity Study on the Effectiveness of
the Pseudo Relative Permeability Concept, PE224 Class
Project. Technical report, Stanford University, Califor-
nia, USA, 2000.
18. Pickup, G. and Stephen, K. An Assessment of Steady-
State Scale-Up for Small-Scale Geological Models.
Technical report, Heriot-Watt University, 2000.
19. Barker, J. and Thibeau, S. “A Critical Review of the Use
of Pseudo Relative Permeabilities for Upscaling”. Paper
SPE 35491 presented at European 3-D Reservoir Mod-
elling Conference, Stavanger, Norway, April 1996.
20. Deutsch, C. V. Geostatistical Reservoir Modeling. Ox-
ford University Press, New York, USA, 2002.
21. Stanford University. Geostatistical Earth Modeling Soft-
ware, 2004.
22. Kleppe, J. Reservoir Simulation. Technical report, Nor-
wegian University of Science and Tecnology, January
2006.
23. Vaca, P. Simulaci´ on de Yacimientos. Universidad Central
de Venezuela, Caracas, Venezuela, 2003.
24. de la Garza, F. R. Simulaci ´ on Num´ erica de Yacimientos.
Technical report, PEMEX, M´ exico, 2000.
25. Mattax, C. and Dalton, R. Reservoir Simulation. SPE,
Richardson, Texas, USA, 1990.
26. Fanchi, J. Principles of Applied Reservoir Simulation.
Butterworth-Heinemann, Texas, USA, 2001.
27. Crotti, M. and Cobenas, R. “Scaling Up of Labo-
ratory Relative Permeability Curves. An Advantageous
Approach Based on Realistic Average Water Satura-
tions”. Paper SPE 69394 presented at the SPE Latin
SPE 106679 J. R. Villa, M. O. Salazar 7
American and Caribbean Petroleum Engineering Confer-
ence, Buenos Aires, Argentina, March 2001.
Appendix A: flowsim2p Algorithm
The first step flowsim2p executes is to read and store the input
data defined in the parameter file, including porosity and per-
meability data (Figure A-1). Considering the fine-grid model
showed in Figure A-2, an index number is assigned to each fine
grid block:
ip = (k −1)n
x
n
y
+ (j −1)n
x
+ i (A-1)
where, ip is the index number, (i, j, k) the Cartesian coordinates
of a fine grid block in x, y, and z, and n
x
, n
y
, n
z
the number
of grid blocks in each direction. The next step is to calculate
the effective permeability of the target coarse grid block by
harmonic mean (Equation 11).
Next, fine grid block pressures and saturations inside a coarse
grid block are computed. The pressure equation is defined and
solved by the IMPES method
22
as shown in Equation A-2:
[T
o
+ T
w
]
α
p
α+1
±[b
o
+ b
w
]
α
= 0 (A-2)
where, T
o
and T
w
are the transmissibility matrix for oil and wa-
ter respectively, p the pressure vector, b
o
and b
w
the source/sink
terms of oil and water respectively, α represents the previous
timestep, α + 1 represents the current timestep, and 0 the null
vector. In flowsim2p, the capillary effects are neglected and
therefore p
o
=p
w
=p. This equation is solved by the LSOR
method
23,24
.
Once the pressure distribution inside the target coarse grid
block is calculated, the phase saturation is calculated by the
following equations
25,26
:
T
o
α
p
o
α+1
±b
o
α
=
φv
i,j,k
∆t

s
o
α+1
−s
o
α

(A-3)
T
w
α
p
w
α+1
±b
w
α
=
φv
i,j,k
∆t

s
w
α+1
−s
w
α

(A-4)
where, φ is the porosity of a fine grid block, v
i,j,k
the volume
of the respective fine grid block, ∆t the timestep size, and s
o
,
s
w
the phase saturation vectors.
The phase relative permeability value of each fine grid block
inside the target coarse grid block can be determinated by the
Corey relations
9
:
k
rw
= γ
1

s
w
−s
wc
1 −s
wc
−s
or

m1
(A-5)
k
ro
= γ
2

1 −s
w
−s
or
1 −s
wc
−s
or

m2
(A-6)
where, k
rw
and k
ro
are the relative permeabilities of water and
oil respectively, s
w
the water saturation in the current time
step, s
wc
the connate water saturation, and s
or
the residual oil
saturation. The terms γ
1
, γ
2
, m
1
, and m
2
are real numbers
determined from the original relative permeability curves and
solving a linear system of equations.
The flow rate in each fine grid block is determined using
Darcy’s law for two-phase flow
27
:
q
j
= −

k
rj
µ
j

kA

∂p
∂x

(A-7)
where, (j) represents the respective phase, q
j
the phase flow
rate at one specific direction, k
rj
the phase relative permeabi-
lity, µ
j
the phase viscosity, k the absolute permeability, A the
cross-sectional area to the flow, and ∂p/∂x the pressure gradi-
ent.
The phase pressure difference and phase saturation of the target
coarse grid block is determined by the following equations
11,17
:
∆p

j
= p

j
inlet
−p

j
outlet
(A-8)
p

j
=
¸
k
rj
k h p
j

¸
k
rj
k h
(A-9)
s

j
=
¸
v s
j
v

(A-10)
where, (j) represents the respective phase, ∆p

j
the effective
phase pressure difference at the target coarse grid block, p

j
inlet
and p

j
outlet
the effective phase pressures at the inlet and out-
let of the target coarse grid block respectively, k
rj
the phase
relative permeability of each fine grid block inside the target
coarse grid block, k the absolute permeability of each fine grid
block inside the target coarse grid block, h the thickness of each
fine grid block inside the target coarse grid block, p
j
the phase
pressure of each fine grid block at the respective face (inlet or
outlet) inside the target coarse grid block, s

j
the effective phase
saturation of the target coarse grid block, s
j
the phase satura-
tion of each fine grid block inside the target coarse grid block,
v the volume of each fine grid block inside the target coarse
grid block, and v

the volume of the target coarse grid block.
flowsim2p calculates the phase difference pressure considering
solely local boundary conditions and therefore only the fine
grid blocks at the inlet and outlet of the target coarse grid block
are used. If global boundary conditions were considered, it
would take into account the central sides of two neighboring
coarse blocks.
Finally, the phase pseudo-relative permeabilities of the target
coarse grid block are determined by Equation 19 adapted to the
premises and suppositions of flowsim2p:
8 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
k

rw
=
µ

w
q

w
D

x
D

y
D

z
k

∆p

w
(A-11)
k

ro
=
µ

o
q

o
D

x
D

y
D

z
k

∆p

o
(A-12)
where, k

rj
are the phase pseudo-relative permeabilities, D

x
,
D

y
, D

z
, the coarse dimensions at each main direction, k

the
effective permeability of the target coarse grid block, ∆p

j
the
phase pressure difference at the target coarse grid block, q

j
the
effective phase rate of the target coarse grid block calculated by
the summation of the phase rates of the fine grid blocks at the
central side of the target coarse grid block, and µ

j
the effec-
tive phase viscosity of the target coarse grid block calculated by
volume weighted mean of the viscosities of the fine grid blocks.
This process is iteratively repeated for each time step. The num-
ber of time steps dependes on the size of the coarse grid block.
Finally, an output file is generated (Figure A-3), containing the
pseudo-relative permeability curves of a coarse-scale model for
a defined upscaling ratio.
SPE 106679 J. R. Villa, M. O. Salazar 9
Table 1: Cases
Cases Number Upscaling fine/coarse
of blocks ratio n
i
/N
i
Fine: 100x130x10 130,000 1 [1 1 1]
Case 1: 50x65x10 32,500 4 [2 2 1]
Case 2: 50x26x5 6,500 20 [2 5 2]
Case 3: 10x13x5 650 200 [10 10 2]
Case 4: 5x5x5 125 1,040 [20 26 2]
Table 2: Upscaling errors
Upscaling Case 1 Case 2 Case 3 Case 4
method
Arithmetic 3.5% 12.3% 14.1% 18.1%
Harmonic 3.9% 17.6% 19.9% 23.9%
Geometric 3.2% 9.2% 11.6% 17.2%
Single-phase 2.9% 6.9% 10.9% 16.6%
Two-phase 2.4% 5.2% 10.5% 15.8%
Figure 1: Reservoir modeling workflow
Figure 2: One-phase and two-phase parameter upscaling
Figure 3: a) Parallel flow, b) Series flow
Figure 4: Power Law
10
10 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
Figure 5: Flow scheme considered in one-phase dynamic
upscaling
Figure 6: Two reservoir models at different upscaling ratios
Figure 7: Methodology for upscaling the reference model
Figure 8: Boundary conditions and flow scheme at the target
coarse grid block considered on flowsim2p
Figure 9: Workflow for flowsim2p
Figure 10: Permeability distribution and well locations in the
reference fine-scale model
SPE 106679 J. R. Villa, M. O. Salazar 11
Figure 11: Histogram of permeability distribution in the re-
ference fine-scale model
Figure 12: Histogram of porosity distribution in the reference
fine-scale model
Figure 13: Permeability and water saturation distribution for
different upscaling ratios - arithmetic static upscaling
Figure 14: Permeability and water saturation distribution for
different upscaling ratios - harmonic static upscaling
12 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
Figure 15: Permeability and water saturation distribution for
different upscaling ratios - geometric static upscaling
Figure 16: Permeability and water saturation distribution for
different upscaling ratios - single-phase dynamic upscaling
Figure 17: Permeability and water saturation distribution for
different upscaling ratios - two-phase dynamic upscaling
Figure 18: Bottom-hole pressure and well water cut for
different upscaling ratios using arithmetic upscaling
SPE 106679 J. R. Villa, M. O. Salazar 13
Figure 19: Bottom-hole pressure and well water cut for
different upscaling ratios using harmonic upscaling
Figure 20: Bottom-hole pressure and well water cut for
different upscaling ratios using geometric upscaling
Figure 21: Bottom-hole pressure and well water cut for
different upscaling ratios using single-phase upscaling
Figure 22: Bottom-hole pressure and well water cut for
different upscaling ratios using two-phase upscaling
14 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
Figure 23: Field water cut for different upscaling ratios and
upscaling methods
Figure 24: Field cumulative oil production for different
upscaling ratios and upscaling methods
Figure 25: CPU time vs. number of blocks
Figure 26: Upscaling error calculation
SPE 106679 J. R. Villa, M. O. Salazar 15
Figure 27: Upscaling error distribution for different
upscaling ratios using arithmetic upscaling
Figure 28: Upscaling error distribution for different
upscaling ratios using harmonic upscaling
Figure 29: Upscaling error distribution for different
upscaling ratios using geometric upscaling
Figure 30: Upscaling error distribution for different
upscaling ratios using single-phase upscaling
16 Permeability Upscaling Techniques for Reservoir Simulation SPE 106679
Figure 31: Upscaling error distribution for different
upscaling ratios using two-phase upscaling
Figure 32: Upscaling error vs. upscaling ratio for different
upscaling methods
Figure 33: Upscaling error vs. upscaling ratio for analytic
and numerical methods
Figure A-1: Parameter file for flowsim2p
SPE 106679 J. R. Villa, M. O. Salazar 17
Figure A-2: Fine grid blocks arrange inside a coarse grid
block - flowsim2p
Figure A-3: Output file of flowsim2p

Sign up to vote on this title
UsefulNot useful