SPE 62922

History Matching Geostatistical Reservoir Models with Gradual Deformation Method

Y. Le Gallo, SPE, M. Le Ravalec-Dupin, SPE, Institut Franais du Ptrole, and the HELIOS Reservoir Group

Copyright 2000, Society of Petroleum Engineers Inc.

This paper was prepared for presentation at the 2000 SPE Annual Technical Conference and
Exhibition held in Dallas, Texas, 14 October 2000.
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
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 an abstract of not more than 300
words; illustrations may not be copied. The abstract 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 presents a new history matching methodology to
constraint 3-D geostatistical reservoir model to well and
production data. This methodology is a general inversion
procedure based upon the gradual deformation method. It
allows for constraining simultaneously petrophysical,
geostatistical and reservoir parameters to dynamic production
data.
The gradual deformation algorithm creates realizations, which
evolve smoothly while preserving the global statistical features
of the model. The deformation process is coupled with an
optimization algorithm to automatically match production
history. After validating the inversion process on synthetic
data, we focused on real data. The inversion process involves
up to fourteen parameters constrained through a fifteen-year
production history. A coarse geostatistical model conditioned
to rock-types and porosities observed at well locations
describes the geological uncertainties. The petrophysical
uncertainties are summarized within the permeability-porosity
laws considered for the two dominant rock-types in the
reservoir. The main reservoir uncertainties are the strengths of
the edge aquifers and the critical gas saturations for each rocktype. The final match is obtained after several inversions and
is quite satisfactory with respect to well pressure, oil and water
flow rates.
Introduction
Geostatistical model enables fine geological interpretations
of the reservoir. But, they are seldom used during the history
match thus leading to a loss of geological information during
the match. However, reservoir characterization could be
improved through conditioning of the geological model to
dynamic production data. Moreover, the geological model
should be preserved and updated during the history match.

This lead to the development of a new history matching

methodology to constrain 3-D geostatistical reservoir models
to well and production data. This methodology is a general
inversion procedure based upon the gradual deformation
method1. It allows for constraining simultaneously
petrophysical, geostatistical and reservoir parameters to
dynamic production data. This paper presents the methodology
and its application, first to a synthetic case and, then to a real
field.
General inversion procedure
The Gradual Deformation Method1 (GDM) and the Fast
Fourier Transform-Moving Average (FFT-MA) algorithm2
yield the main components of the general inversion process.
FFT-MA algorithm
The FFT-MA algorithm is used to produce unconditional
Gaussian realization y with stationary covariance function C
from:
y = y o + g z .(1)
where yo is the mean of y and z is a Gaussian white noise.
Function g results from the decomposition of the covariance
function as:
C = g g (2)
Determining g and calculating the convolution product g*z
may be an arduous task. Translating the problem into the
spectral domain makes it much easier.
Gradual Deformation Method
The GDM is a geostatistical parameterization technique to
perturb a realization from a few parameters, termed
deformation parameters, while preserving the spatial
variability. It can be applied to the Gaussian white noise used
as input into the FFT-MA algorithm:
z = z 1cos( ) + z 2 sin ( ) (3)
This relation ensures that z is a Gaussian white noise if z1
and z2 are two independent Gaussian white noises. Varying
deformation parameter allows for describing a chain of

Y. LE GALLO, M. LE RAVALEC-DUPIN

Gaussian white noises. As the deformation rule is periodic,

ranges from 1 to 1. When it is 0.0 (0.5), z is the same as z1
(z2). Introducing z into Eq. 1 yields a Gaussian realization y.
Thus, smooth variations in induce smooth variations in y. In
other words, realization y can be modified, whatever its size,
from a single deformation parameter. All along the
deformation process, the spatial variability of y is unchanged.
Because of these properties, it is attractive to integrate the
GDM into optimization processes. Thus, the search process is
designed to assess an optimal deformation parameter to
minimize a given objective function. However, considering
solely the realization chain derived from z1 and z2 restricts our
investigation of the realization space. To explore other
directions, the building of realization chains is repeated. Each
of them is sequentially screened to estimate the corresponding
optimal deformation parameter. The Gaussian white noise
built from the optimal deformation parameter defined at step
(i) is used in place of z1 at step (i+1). Additionally, a new
Gaussian white noise z2 is randomly drawn for every new
chain.
This gradual deformation scheme could be extended to the
combination of more than two realizations3. The number of
deformation parameters equals the number of combined
realization minus one.
General Inversion Scheme
An objective function J is defined prior to any optimization
process. It measures the suitability of the suggested reservoir
model. Since the GDM preserves the spatial variability, the
objective function considered here is reduced to the mismatch
between the measured data and the ones simulated for the
studied reservoir model:
J ( , g , P ) =

1
( f (z( )) d obs )t C D1 ( f (z( )) d obs ) ...(4)
2

z is the Gaussian white noise vector characterizing the

reservoir model. dobs is the vector of measured data and CD is
the covariance matrix quantifying the experimental and
theoretical uncertainties. The objective function J depends on
the deformation parameters that define the Gaussian white
noise z (Eq. 3), on the structural (geostatistical) parameters g
depending on function g and mean yo (Eq. 1), and on the
production parameters P involved in the fluid flow simulation.
Parameters g and P are not explicitly written on the right-hand
side of Eq. 4: they are integrated into the operator f mapping
the Gaussian white noise space into the data space.
The main steps of the general inversion procedure4 are
depicted in Fig. 1.They are as follows:
1. Generate an initial Gaussian white noise z1.
2. Generate a complementary Gaussian white noise z2.
3. Gradual deformation of z1 (Eq. 3).
4. Compute the corresponding realization y (Eq. 1).
5. Compute the fluid flow simulation.
6. Determine the objective function.
7. Minimize the objective function by varying , g and P.

8.
9.

SPE 62922

Update the initial Gaussian white noise and of the

structural and production parameters.
Return to step 2 as long as the match is not satisfactory.

Application
Field and reservoir model
The method is applied to an offshore oil field with two
main reservoirs. They are produced through seven wells, five
of which are perforated in both reservoirs. The sediments were
deposited along an east-west direction. The upper reservoir has
two main layers with clean sand and feldspar-rich sandstone.
Two-edge aquifers provide some pressure support on the
eastern and western flanks. The lower reservoir has also two
main layers: one, which is mainly fine-grained sandstone with
some interbedded clay, the other one, which is mainly
dolomite. Active edge aquifers on the eastern and western
flanks support pressure in the former. The latter is a poor
quality reservoir except in its central zone. Clay and slitydolomite barriers isolate the two reservoirs. However, crossflow may occur between the reservoirs through the five
common production wells.
Field production started in mid 1982 through natural
depletion up to early 1983 when water injection took place.
The water injection was quite limited with respect to aquifer
water influx especially in the upper reservoir. By mid 1983, all
production wells were gas-lifted.
A no-flow barrier models the inter-reservoir. Hence, the
numerical model only includes four reservoir layers (two for
each of the reservoirs). A full-field 66x67x4 regular reservoir
grid is used. Carter and Tracy analytical approach is used to
model the aquifer behavior.
2-D seismic interpretations highlighted two main faults: an
east-west sealing fault, a major northeast-southwest sealing
fault. However, several minor faults could exist. The reservoir
model only considered the two major faults.
Reservoir models were established considering solely the
two main rock-types, referenced as the "good" one and the
"bad" one. Their mean porosities are 25% and 15% (Table 1),
respectively. The reservoir models were constrained to rocktypes and porosities at well locations, when these data were
available. The distributions of the rock-types within the layers
exhibit different trends (Table 2). The main trend observed in
the upper reservoir is that rock-type proportions vary along the
east-west depositional axis. For the lower reservoir, this trend
is submitted to a depth correlation. Thus, the good rock-type
is prominent in the upper reservoir model while it is
essentially observed in the central area of the lower reservoir.
Permeabilities (K) are correlated to porosities () with respect
to the relations log(K ) = 6.5 + 0.65 for the good rock-type
and log(K ) = 12. 0.8 for the bad one.
As the four layers are modeled independently, we proceed
as follows to build a reservoir model (Fig. 2):
1. Twelve Gaussian white noises were generated.
2. Four Gaussian white noises were turned into four
stationary Gaussian realizations (Eq. 1) using the
structural properties reported in Table 1 for the rock-type

SPE 62922

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

distributions.
These distributions were conditioned to the rock-types
observed at well locations.
4. These four realizations were truncated with respect to
proportion maps accounting for the observed trends, thus
yielding rock-type realizations.
5. The other Gaussian white noises were used as input to
create porosity realizations (Eq. 1). The structural
properties for the good and bad porosity distributions
are detailed in
Table 1. There are two porosity
realizations per layer, one for the good rock-type, and
the other one for the bad rock-type.
6. The porosity realizations were conditioned to porosity
measurements at well location.
7. The generated porosity values were used to fill the rocktype realizations.
The main fluid properties are summarized in Table 3. The
upper reservoir oil in place (STOOIP) is about 5.5 106 Sm3
while the lower reservoir oil in place (STOOIP) is about
4.2 106 Sm3.
3.

Synthetic case
Before applying the general inversion scheme to real data,
we focused on a synthetic case. A reference reservoir model,
characterized by the same geological features as above, was
built (Fig. 3). The fluid flow simulation was computed over a
15-year production history. The computed pressures, flow
rates, water cuts and gas oil ratios were considered as the
reference data. Then, we aimed at determining a reservoir
model capable to duplicate the reference dynamic behavior. At
this stage, the rock-type distribution, the porosity distributions,
the average porosities as well as the activity multiplier
coefficients for the two aquifers are assumed to be unknown.
History match parameters. Two experiments were planned.
For the first one, we considered 5 parameters to be optimized:
the two mean porosities, the two aquifer activity coefficients,
and one deformation parameter. The porosity and aquifer
parameters were submitted to constraints (inequalities) during
the optimization process (Table 4). The deformation
parameter was used to modify the rock-type distributions for
the four layers. For the second experiment, we considered 8
parameters to be optimized: again, the two mean porosities
and the two aquifer activity coefficients, plus 4 deformation
parameters instead of one. These 4 deformation parameters
allowed for varying the rock-type distributions for the four
layers independently. Every layer was attributed a deformation
parameter. Adding new deformation parameters increases the
number of degrees of freedom and makes the inversion
process more flexible. In both cases, we used porosity
distributions different from the reference ones, but we did not
try to constrain them. That way, we introduced some noise
into the inversion process.
Results. For the two experiments, the initial reservoir
models (Fig. 2), mean porosities and aquifer activity
coefficients were identical. With 5 parameters, satisfactory
match for the pressures (Fig. 5), oil flow rates (Fig. 6) and

water cuts (Fig. 7) was achieved after the successive

investigation of five realization chains. However, the gas oil
ratios obtained for the final reservoir model did not properly
reproduce the reference ones (Fig. 8). The mean porosity
parameters converge towards their reference values (Fig. 9a).
Unlike the east aquifer coefficient, the west one did not
converge towards its reference value: it was stopped by the
upper bound (Fig. 9b). Its value turned out to be correlated to
the water cut in well 127.
When eight optimization parameters were used instead of
5, the match improves significantly (Fig. 5, Fig. 6 and Fig. 7),
especially for the gas oil ratios (Fig. 8). Again, we observed
that the reference mean porosity values were reached (Fig. 9a)
while the behavior of the aquifer activity coefficients were no
longer restricted by the bounds (Fig. 9b). After screening three
realization chains, the objective function decreases below 10
(Fig. 9c). With 5 parameters, it was about 25 after the
investigation of 10 realizations.
This better match is due to additional deformation
parameters. For the first experiment, the rock-type
distributions for the four layers were controlled by a single
deformation parameter. This deformation parameter could
evolve only if its variation improved the whole reservoir
model. Thus, its influence on the inversion process was limited
by the size of the reservoir. With 8 parameters, each layer
depends on a deformation parameter. They can be modified
independently, which increases the flexibility of the inversion
process. Fig. 9 shows the influence additional deformation
parameters. With one deformation parameter, the final
reservoir model was not very different from the initial one.
With four deformation parameters, differences appear clearly.
That way, we had more degrees of freedom to determine rocktype distributions consistent with the reference ones. It was no
longer necessary to boost the west aquifer coefficient to match
the water cut at well 127.
Real Case
The method is used to improve the geological reservoir
model with dynamic production constraint. The goal of this
work is to obtain a history match of the reservoir production,
especially oil and water, by adjusting the reservoir parameters
and the geological model while maintaining the well
constraints (porosity and rock-type proportions at the wells).
History match parameters. The general inversion scheme
is applied to the same reservoir but with the actual 15-year
field production history: pressure, gas oil ratio, and water-cut.
To model the field gas production, gas relative
permeabilities are used in the history match. The relative
permeabilities are modeled as function of the fluid saturation
e.g. Corey model. The Corey exponents of the gas relative
permeability for each rock-type are added to the inversion
parameters as well as gas critical saturation for each of the
rock-type.
The different water production behaviors of two nearby
wells (127 and 137) indicate there may be a sealing fault
somewhere between the wells. Thus, a fault was added

Y. LE GALLO, M. LE RAVALEC-DUPIN

midway between the two wells. The fault transmissivity

coefficient was added to the inversion parameters to test if
such a fault may be justified by the production history.
To reproduce the different dynamic behavior of the two
reservoirs, each aquifer influence was modeled separately
using multiplier coefficient of the aquifer transmissivity.
As in the synthetic case and due to computational
constraints, only one global deformation parameter is used to
modify the rock-type distribution for the whole reservoir. The
average porosity of each rock-type is still used in the match, as
well as the intercept of the porosity-permeability relationship
for each of the rock-type.
All the above lead to 14 inversion parameters summarized
in Table 5.
The objective function (Eq. 4) used to quantify the match is a
cumulative weighted least-square function between simulation
and measurement. In this real case approach, oil production
and pressure measurements were the most important parameter
to match (heaviest weights). Given measurement uncertainties,
gas production was assigned the lightest weight. The water
production was assigned an intermediate weight. The choice of
weights affects the history match and influences the absolute
value of the objective function. However, this absolute value is
not important. Only its relative evolution indicates the quality
of the match: when the objective function decreases the match
quality improves as shown in Table 5.
Results. Table 5 summarizes the parameter evolutions with
the number of realization chains. In this case, seven realization
chains were computed. However, the main parameter
adjustments were obtained at the end of the second realization
chain. The following realization chains only resulted in minor
improvements (see Table 5). The seven realization chains
implied 164 reservoir simulations whereas only 91 reservoir
simulations were used to screen the first two realization
chains. It is important to note that 15 reservoir simulations are
required to compute the gradients with respect to each of the
14 inversion parameters, which lead us to only use one global
deformation parameter.
During the screening of realization chains, the overall
reservoir oil and water productions improve significantly (as
in Fig. 10). The fit between field data and model is quite good
(see left and middle graphs in Fig. 10). The gas production is
not so well predicted by the model (see the right graph in Fig.
10) due to the low weight assigned to these measurements in
the objective function. However, the match varies from well to
well since we use a single deformation parameter for the
whole reservoir.
To reproduce the very different water breakthroughs
between well 127 and 137, the waterfront must be slowed
down between the two wells. The history match indicates that
a partially sealing fault is necessary (a 0.29 transmissivity
multiplication coefficient is used in Table 5).
The average rock-type porosity (see Fig. 11 and Fig. 13)
converge towards 29 % for the good rock-type and 14 % for
the bad one as shown in Table 5. The intercepts of the
porosity-permeability law converge towards values quite

SPE 62922

different from those obtained in the synthetic case: 1.32 and

0.65 for the good rock-type in the real and synthetic case
respectively and, 0.98 and -0.8 for the bad rock-type in the
real and synthetic case respectively. These differences are due
to the production history used in the match. Obviously the
permeability are significantly increased during the history
match as illustrated in Fig. 12 and Fig. 14.
To best match the gas production profiles, the critical gas
saturations of the two rock-types have been switch by the
inversion process (see Table 5). The Corey exponent of the gas
relative permeability of the bad rock-type does not play any
significant role in this match: its values do not change during
the match. The Corey exponent of the gas relative
permeability of the good rock-type is slightly lower. These
gas relative permeability parameters play a significant role
towards improving the model match of the gas production.
During the history match, the eastern aquifers are only
slightly changed from their initial value (see Table 5).
However, the western aquifers are increased between 15 to
22 % depending on the reservoir. This suggested that the
initial western aquifer were not insuring enough pressure
support.
The well pressures (Fig. 15) are not well matched despite
the weight assigned to pressure measurements. The poor
pressure match explains the large absolute value of the
objective function even after seven realization chains.
However, for some wells, e.g. 127 and 238, the match is quite
good. For most of them, the match is quite approximate
especially those exhibiting a pressure increase combined with
a GOR increase towards the end of the available history (Fig.
18), e.g. 137, 254 and 467. The model could not display such a
behavior using the parameter chosen for the match. The
pressure match may be improved using several deformation
parameters instead of one global deformation parameter for
the whole reservoir as in the synthetic case.
The standard oil rates (Fig. 16) are well matched but
compensate for the other phase (gas and water) rate matches
Fig. 17 and Fig. 18) since in the reservoir simulator, the total
fluid flow rate was imposed for each well. Water cuts are
correctly predicted for wells 137 and 344. But water
breakthrough (Fig. 17) is too large in 254 well and too late in
the other wells. With the exception of 238 and 467 wells,
where no water production was computed, the history
matching improved significantly the model fit even in 127
well.
The GOR match is even more difficult than the water cut
match. In some wells, GOR increases despite a pressure
increase towards the end of the history. Thus, the goal of the
match is only to model the GOR trend. As illustrated in Fig.
18, the model fit is quite improved using the gradual
deformation history match.
Conclusions
The inversion methodology was successfully applied to a
synthetic case built from an actual field. In this case, the final
matching was pretty good. With the real field production data,

SPE 62922

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

the match quality is not as good. It could be improved by

increasing the number of gradual deformation parameters as
suggested by the synthetic study. We observed that with one
deformation parameter per layer, we gave more flexibility to
the inversion process. In such conditions, the rock-type
distributions evolved clearly. Anyway, an automatic match
was reached using several parameters. An exact match was
beyond the scope of this work. Additionally, the results could
have been refined performing some local deformation
parameters especially around 127 and 137 wells4.
Nomenclature
C = covariance
dobs = measured data
K = permeability, mD
J = objective function
g = structural (geostatistical) parameters
P = production parameters
z = Gaussian white noise
y = Gaussian realization
= porosity
= deformation parameter
Acknowledgements
The authors wish to thank Elf Exploration Production
(EEP) and Institut Franais du Ptrole (IFP) for their
permission to publish this paper and financial support. This
work was only possible through discussions and exchanges
with all the participants to the HELIOS reservoir group project
especially G. Vincent (EEP). All reservoir simulations were
carried out using ATHOS reservoir simulator, which is jointly
developed between IFP and BEICIP-FRANLAB.

Table 1 Structural properties of the reservoir model

Layer
1

Rock-type
distributions
anisotropic
Gaussian variogram
1st main axis:
(0;1;0)
1st main correlation
length: 1000m
2nd main axis:
(1;0;0)
2nd correlation
length: 500m
mean: 0.
variance: 1.

isotropic Gaussian
variogram
correlation length:
1000m
mean: 0.
variance: 1.

isotropic Gaussian
variogram
correlation length:
750m
mean: 0.
variance: 1.

isotropic Gaussian
variogram
correlation length:
750m
mean: 0.
variance: 1.

References
1. Hu, L. Y., Gradual deformation and iterative calibration of
Gaussian-related stochastic models, Math. Geol., 32(1): 87-108
(2000).
2. Le Ravalec, M., B. Noetinger, and L. Y. Hu, The FFT moving
average (FFT-MA) generator: An efficient tool for generating
and conditioning Gaussian simulations, Math. Geol., 32(6): 701723 (2000).
3. Roggero, F., and L. Y. Hu, Gradual deformation of continuous
geostatistical models for history matching, SPE 49004, New
Orleans, LA, 27-30 September 1998.
4. Le Ravalec, M., L. Y. Hu, and B. Noetinger, Stochastic
reservoir modeling constrained to dynamic data: Local
calibration and inference of the structural parameters, SPE
56556, Houston, TX, 3-6 October 1999.

SI Metric Conversion Factors

cp x 1.0
ft x 3.048
ft2 x 9.290 304
ft3 x 2.831 685
mD x 9.869 233
psi x 6.894 757
bbl/d x 1.589 873
scf/bbl x 1.801 175

E-03
E-01
E-02
E-02
E-04
E+00
E-01
E-01

= Pa.s
=m
= m2
= m3
= m2
= kPa
= m3/d
= St m3/ m3

Porosity distributions
anisotropic Gaussian
variogram
1st main axis: (0;1;0)
1st main correlation length:
500m
2nd main axis: (1;0;0)
second correlation length:
250m
good rock-type mean: 0.25
good rock-type variance:
6.25 10-4
bad rock-type mean: 0.15
bad rock-type variance:
6.25 10-4
isotropic Gaussian variogram
correlation length: 500m
good rock-type mean: 0.25
good rock-type variance:
6.25 10-4
bad rock-type mean: 0.15
bad rock-type variance:
6.25 10-4
isotropic Gaussian variogram
correlation length: 375m
good rock-type mean: 0.25
good rock-type variance:
6.25 10-4
bad rock-type mean: 0.15
bad rock-type variance:
6.25 10-4
isotropic Gaussian variogram
correlation length: 375m
good rock-type mean: 0.25
good rock-type variance:
6.25 10-4
bad rock-type mean: 0.15
bad rock-type variance:
6.25 10-4

Y. LE GALLO, M. LE RAVALEC-DUPIN

Table 2 Geological rock-type evolution in the reservoir

Reservoir
Layer
Rock-type drift

upper
1
NE SW

proportion good
rock-type (%)

50

upper lower
2
3
EW EW+
depth
50
30

lower
4
depth
10

Table 3 Fluid properties

Depth (m)
Water oil contact upper reservoir (m)
Water oil contact lower reservoir (m)
Initial reservoir pressure (kPa)
Bubble point pressure (kPa)
Gas gravity
Gas dissolution ratio (Rm3/Sm3)
API
Oil formation volume factor @ bubble point
pressure (Rm3/Sm3)
Oil viscosity @ bubble point pressure (cp)

1100
1145
1190
119 102
80 102
0.88
42
29
1.12
3.8

Table 4 Optimized parameters ( estimated from 5 parameters;

**
estimated from 8 parameters)

Parameter
Mean of
the good
rock-type
porosity
Mean of
the bad
rock-type
porosity
West
aquifer
activity
multiplier
coefficient
East
aquifer
activity
multiplier
coefficient

Initial
value

Constraint

0.29

0.21 < g < 0.29

Predicted
value
0.24*

0.25
0.24**
0.16

0.11

0.11 < b < 0.19

Reference
value

0.15
0.15**
3.00*

0.7

10

< Aw < 3.

1.00
2.19**
1.43*

1.3

106 < Ae < 3.

1.00
1.69**

SPE 62922

Table 5 Parameter evolution during the iterative history match

with Gradual Deformation Method

Realization number
Gradual deformation
coefficient ( x10-4)
Upper West aquifer
activity multiplier
coefficient
Lower West aquifer
activity multiplier
coefficient
Upper East aquifer
activity multiplier
coefficient
Lower East aquifer
activity multiplier
coefficient
Fault Transmissivity
multiplier coefficient
Good rock-type
porosity
Bad rock-type
porosity
Good rock-type
permeability coefficient
Bad rock-type
permeability coefficient
Good rock-type
critical gas saturation
Bad rock-type critical
gas saturation
Good rock-type gas
relative permeability
exponent
Bad rock-type gas
relative permeability
exponent
Objective function

initial
0

1
196

3
-174

5
-7.45

final
-1.54

1.02

1.15

1.14

1.15

.92

.97

.97

1.22

.95

1.07

1.07

1.07

1.04

1.

1.

1.

0.26

0.29

0.29

0.29

0.25

0.29

0.29

0.29

0.29

0.15

0.13

0.137 0.138 0.138

1.06

1.12

-1

-0.98

-0.98 -0.98

-0.98

0.02

0.031

0.03

0.03

0.03

0.026

0.021 0.022 0.022

0.5

0.42

0.42

0.42

0.42

0.5

0.51

0.51

0.51

0.51

165

82

79

75

74

1.22

0.03

1.32

SPE 62922

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

z1

z2

Gradual Deformation

optimization

z()
FFT-MA

K
flow
simulation

,g,P

Fig. 1 Flow chart of the general inversion loop.

Fig. 3 Porosity maps for the reference synthetic case.

Fig. 2 Porosity maps for the initial synthetic case.

Fig. 4 Porosity maps for the final synthetic case.

Y. LE GALLO, M. LE RAVALEC-DUPIN

7000

S t a n d a r d O i l R a t e ( m 3 /d )

P ressu r e (k P a )

8000

W e ll 1 2 7

referen ce
in itia l
5 p a r a m e te rs
8 p a r a m e te rs

10000

9000

9000

8000

7000

6000

6000

5000

5000

4000

4000

150

W e ll 1 3 7

11000

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

10000

P ressu r e (k P a )

150

12000

W e ll 1 2 7

11000

125

W e ll 1 3 7
S t a n d a r d O i l R a t e ( m 3 /d )

12000

SPE 62922

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

100

75

50

25

3000
0

1500

3000

4500

6000

0
0

1500

T im e (d a y s )

4500

6000

P ressu r e (k P a )

8000

7000

9000

8000

7000

6000

6000

5000

5000

4000

4000

4500

6000

3000
1500

3000

4500

6000

125

1500

100

75

50

4500

6000

10000

9000

9000

P ressu r e (k P a )

10000

8000

7000

50

6000

1500

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

8000

7000

3000

4500

6000

T im e (d a y s )
150

125

W e ll 3 4 7

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

100

75

50

125

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

100

75

50

25

4000

3000

4500

6000

T im e (d a y s )

0
0

1500

3000

4500

6000

1500

T im e (d a y s )

3000

4500

6000

T im e (d a y s )

12000

1500

3000

4500

6000

T im e (d a y s )

150

W e ll 4 6 7

11000

W e ll 4 6 7
S t a n d a r d O i l R a t e ( m 3 /d )

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

10000

P ressu r e (k P a )

75

W e ll 3 4 4

3000
1500

4500

25

3000
0

100

5000

referen ce
in itia l
5 p a ra m e te rs
8 p a ra m e te rs

4000

3000

150

6000

6000

5000

1500

W e ll 3 4 7

11000

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

T im e (d a y s )

S t a n d a r d O i l R a t e ( m 3 /d )

W e ll 3 4 4

11000

125

0
0

12000

6000

25

T im e (d a y s )

12000

P ressu r e (k P a )

3000

4500

W e ll 2 5 4

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

0
0

T im e (d a y s )

3000

150

S t a n d a r d O i l R a t e ( m 3 /d )

1500

T im e (d a y s )

25

3000

50

W e ll 2 3 8

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

10000

9000

3000

150

S t a n d a r d O i l R a t e ( m 3 /d )

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

10000

1500

W e ll 2 5 4

11000

75

T im e (d a y s )

12000

W e ll 2 3 8

11000

100

0
0

T im e (d a y s )

12000

P ressu r e (k P a )

3000

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

25

S t a n d a r d O i l R a t e ( m 3 /d )

3000

125

9000

8000

7000

6000

125

referen ce
in itia l
5 p a ra m e t e r s
8 p a ra m e t e r s

100

75

50

5000
25
4000

3000

0
0

1500

3000

4500

6000

T im e (d a y s )

Fig. 5 - Pressure variations: The black dots describe the reference

case, the solid thin lines are the initial simulation results, the
dashed lines are the final simulation results with 5 inversion
parameters, the solid thick lines are the final simulation results
with 8 inversion parameters.

1500

3000

4500

6000

T im e (d a y s )

Fig. 6 - Standard oil rate variations: The black dots describe the
reference case, the solid thin lines are the initial simulation
results, the dashed lines are the final simulation results with 5
inversion parameters, the solid thick lines are the final simulation
results with 8 inversion parameters.

SPE 62922

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

100

100

75

W e ll 1 2 7

75

W e ll 1 3 7

W e ll 1 2 7

W e ll 1 3 7

60

40

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

60

40

70

G a s O il R a t i o ( m 3 / m 3 )

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

W a ter cu t

W a ter cu t

80

G a s O il R a t i o ( m 3 / m 3 )

70
80

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

60

55

50

45
20

60

55

50

45

1500

3000

4500

6000

40

35

0
0

1500

T im e (d a y s )

3000

4500

35
0

6000

1500

T im e (d a y s )

3000

4500

6000

W e ll 2 3 8

40

60

40
r e fe re n c e
in itia l
5 p a ra m e te rs
8 p a ra m e te rs

20

20

60

55

50

1500

3000

4500

1500

T im e (d a y s )

3000

4500

6000

1500

3000

4500

6000

75

W e ll 3 4 4

40

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

60

40

4500

6000

W e ll 3 4 7
70

G a s O il R a t i o ( m 3 / m 3 )

W a ter cu t

60

G a s O il R a t i o ( m 3 / m 3 )

80

3000

75

W e ll 3 4 7

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

1500

T im e (d a y s )

70
80

50

T im e (d a y s )

100

W e ll 3 4 4

55

35
0

T im e (d a y s )

100

6000

60

40

35
0

6000

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

45

45

0
0

4500

W e ll 2 5 4

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

40

6000

70

G a s O il R a t i o ( m 3 / m 3 )

W a ter cu t

60

G a s O il R a t i o ( m 3 / m 3 )

80

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

4500

75

W e ll 2 5 4
70

80

3000

T im e (d a y s )

75

W e ll 2 3 8

1500

T im e (d a y s )

100

100

W a ter cu t

referen ce
in itia l
5 p a ra m e te r s
8 p a ra m e te r s

65

20
40

W a ter cu t

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

60

55

50

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

60

55

50

45

45

40

40

20

20

0
0

1500

3000

4500

6000

T im e (d a y s )

35
0

1500

3000

4500

6000

35
0

1500

T im e (d a y s )

3000

4500

6000

T im e (d a y s )

100

1500

3000

T im e (d a y s )

75

W e ll 4 6 7

W e ll 4 6 7

W a ter cu t

G a s O il R a t i o ( m 3 / m 3 )

70
80

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

60

40

r e fe re n c e
i n i t ia l
5 p a ra m e te rs
8 p a ra m e te rs

65

60

55

50

45
20
40

35
0

1500

3000

4500

6000

T im e (d a y s )

Fig. 7 - Water cut variations: The black dots describe the

reference case, the solid thin lines are the initial simulation
results, the dashed lines are the final simulation results with 5
inversion parameters, the solid thick lines are the final simulation
results with 8 inversion parameters.

1500

3000

4500

6000

T im e (d a y s )

Fig. 8 - Gas oil ratio variations: The black dots describe the
reference case, the solid thin lines are the initial simulation
results, the dashed lines are the final simulation results with 5
inversion parameters, the solid thick lines are the final simulation
results with 8 inversion parameters.

10

Y. LE GALLO, M. LE RAVALEC-DUPIN

SPE 62922

Fig. 9 - Parameter evolution during the inversion for the initial synthetic case.

2 .0

300

100

IN F & S U P R e s e r v o ir s
f ie l d
in itia l
f in a l

1 .6

W a te r c u t

1 .4

1 .0

0 .8

60

40

0 .6

0 .4

fie ld
i n i t ia l
fin a l

250

80

1 .2

IN F & S U P R e s e r v o ir s

G a s O il R a tio (m 3 /m 3 )

fie ld
i n i t ia l
fin a l

C u m u la tiv e O il P r o d u m
c t )i o n ( 1 0

IN F & S U P R e s e r v o ir s
1 .8

200

150

100

20
50

0 .2

0 .0

0
0

1500

3000

T im e (d a y s )

4500

6000

0
0

1500

3000

T im e (d a y s )

4500

6000

1500

3000

4500

6000

T im e (d a y s )

Fig. 10 Reservoir parameter variations during history matching (left: standard cumulative oil production, center: water cut, right: gas oil
ratio). The black dots describe the field measurements, the solid thin lines are the initial simulation results, and the solid thick lines are the
final simulation results.

SPE 62922

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

Fig. 11 Initial porosity maps for the real case.

Fig. 13 - Final porosity maps for the real case

Fig. 12 Initial permeability maps for the real case

Fig. 14 - Final permeability maps for the real case

11

12

Y. LE GALLO, M. LE RAVALEC-DUPIN

fie ld
in itia l
2 n d it e r a t i o n
fin a l

P ressu r e (k P a )

10000

9000

8000

7000

9000

8000

7000

6000

6000

5000

5000

4000

4000

250

W e ll 1 3 7
S t a n d a r d O i l R a t e ( m 3 /d )

fie ld
in itia l
2 n d it e r a t i o n
fin a l

10000

300

W e ll 1 2 7

W e ll 1 3 7

11000

S t a n d a r d O i l R a t e ( m 3 /d )

W e ll 1 2 7

11000

P ressu r e (k P a )

300

12000

12000

SPE 62922

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

50

1500

3000

4500

6000

1500

T im e (d a y s )

4500

100

1500

6000

3000

4500

6000

1500

T im e (d a y s)

6000

9000

8000

7000

6000

6000

5000

5000

W e ll 2 5 4

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

50

4000

4000

250

S t a n d a r d O i l R a t e ( m 3 /d )

7000

S t a n d a r d O i l R a t e ( m 3 /d )

P ressu r e (k P a )

8000

300

W e ll 2 3 8

fie ld
in itia l
2 n d it e r a t i o n
fin a l

10000

9000

300

W e ll 2 5 4

11000

fie ld
in itia l
2 n d it e r a t i o n
fin a l

10000

250

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

50

3000

3000
0

1500

3000

4500

6000

1500

T im e (d a y s )

3000

4500

6000

0
0

T im e (d a y s )

1500

3000

4500

6000

1500

T im e (d a y s)
12000

3000

4500

6000

T im e (d a y s)

12000
300

10000

P ressu r e (k P a )

9000

8000

7000

W e ll 3 4 4

fie ld
in itia l
2 n d it e r a t i o n
fin a l

9000

8000

7000

6000

6000

5000

5000

4000

4000

250

W e ll 3 4 7
S t a n d a r d O i l R a t e ( m 3 /d )

10000

300

W e ll 3 4 7

11000

fie ld
in itia l
2 n d it e r a t i o n
fin a l

S t a n d a r d O i l R a t e ( m 3 /d )

W e ll 3 4 4

11000

P ressu r e (k P a )

4500

12000

W e ll 2 3 8

11000

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

50

3000
1500

3000

4500

6000

T im e (d a y s )

1500

3000

4500

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

0
0

6000

250

50

3000
0

1500

3000

4500

6000

T im e (d a y s)

T im e (d a y s )

1500

3000

4500

T im e (d a y s)

300

12000

W e ll 4 6 7
S t a n d a r d O i l R a t e ( m 3 /d )

W e ll 4 6 7

11000

fie ld
in itia l
2 n d it e r a t i o n
fin a l

10000

P ressu r e (k P a )

3000

T im e (d a y s)

T im e (d a y s )

12000

P ressu r e (k P a )

3000

150

0
0

3000
0

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

50

3000

250

9000

8000

7000

6000

5000

250

fie ld
in itia l
2 n d it e r a t i o n
fin a l

200

150

100

50

4000
0
0

3000
0

1500

3000

4500

6000

1500

3000

4500

6000

T im e (d a y s)

T im e (d a y s )

Fig. 15 - Pressure variations during history matching. The black

dots describe the field measurements, the solid thin lines are the
initial simulation results, the dashed lines are the second iteration
simulation results, and the solid thick lines are the final
simulation results.

Fig. 16 - Standard oil flow variations during history matching. The

black dots describe the field measurements, the solid thin lines
are the initial simulation results, the dashed lines are the second
iteration simulation results, and the solid thick lines are the final
simulation results.

6000

HISTORY MATCHING GEOSTATISTICAL RESERVOIR MODELS WITH GRADUAL DEFORMATION METHOD

100

20

60

40

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

20

0
1500

3000

4500

6000

1500

60

40

4500

1500

3000

4500

350
300
250
200
150

40

20

3000

4500

1500

3000

4500

6000

T im e (d a y s )

1500

3000

4500

1500

3000

350

W e ll 3 4 7

450

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

300
250
200
150

400

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

350
300
250
200
150
100
50
0

1500

T im e (d a y s )

3000

4500

6000

T im e (d a y s )

1500

3000

T im e (d a y s )

500

W e ll 4 6 7

W e ll 4 6 7

450

G a s O i l R a t io ( m 3 / m 3 )

80

W a ter cu t

150

500

100

6000

200

W e ll 3 4 4

400

6000

4500

250

T im e (d a y s )

0
0

300

6000

50
0

6000

350

0
1500

100

20

4500

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

400

50

450

40

6000

100

500

60

4500

W e ll 2 5 4

450

T im e (d a y s )

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

3000

500

G a s O i l R a t io ( m 3 / m 3 )

W a ter cu t

60

1500

T im e (d a y s )

fie ld
in it i a l
2 n d ite ra tio n
fin a l

400

6000

W e ll 3 4 7
80

150

50

100

W e ll 3 4 4

200

6000

W e ll 2 3 8

T im e (d a y s )

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

4500

T im e (d a y s )
100

3000

100

f ie l d
in itia l
2 n d ite ra tio n
f in a l

1500

250

450

40

300

T im e (d a y s )

60

6000

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

350

50

400

100

500

20

80

150

6000

G a s O i l R a t io ( m 3 / m 3 )

W a ter cu t

W a ter cu t

4500

80

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

20

W a ter cu t

3000

W e ll 2 5 4

3000

200

50

100

W e ll 2 3 8

1500

250

T im e (d a y s )

100

300

0
0

T im e (d a y s )

80

350

100

0
0

400

W e ll 1 3 7

450

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

G a s O i l R a t io ( m 3 / m 3 )

40

G a s O i l R a t io ( m 3 / m 3 )

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

60

W e ll 1 2 7

450

80

W a ter cu t

W a ter cu t

80

W e ll 1 3 7

13

500

500

W e ll 1 2 7

G a s O i l R a t io ( m 3 / m 3 )

100

G a s O i l R a t io ( m 3 / m 3 )

SPE 62922

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

60

40

20

400

fie ld
i n i t ia l
2 n d ite ra tio n
fin a l

350
300
250
200
150
100
50

0
0

1500

3000

4500

6000

T im e (d a y s )

Fig. 17 - Water cut variations during history matching. The black

dots describe the field measurements, the solid thin lines are the
initial simulation results, the dashed lines are the second iteration
simulation results, and the solid thick lines are the final
simulation results.

1500

3000

4500

6000

T im e (d a y s )

Fig. 18 - GOR variations during history match. The black dots

describe the field measurements, the solid thin lines are the initial
simulation results, the dashed lines are the second iteration
simulation results, and the solid thick lines are the final
simulation results.