Professional Documents
Culture Documents
SPE 38653 A Procedure To Integrate Well Test Data, Reservoir
SPE 38653 A Procedure To Integrate Well Test Data, Reservoir
A Procedure to Integrate Well Test Data, Reservoir Performance History and 4-D
Seismic Information into a Reservoir Description
Jorge L. Landa, SPE, and Roland N. Horne, SPE, Stanford University
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 Introduction
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
Devising the optimal strategy for the development of an oil or
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at gas reservoir is an important and difficult task. Many
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 mathematical techniques for optimization can be used to deal
for commercial purposes without the written consent of the Society of Petroleum Engineers is with problems in engineering and economics systems. These
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 techniques assume that we have a fairly complete
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. understanding of the problem and also that we can construct a
mathematical model that predicts the system’s performance
Abstract accurately in time under different scenarios; this is not a
This paper deals with the problem of estimating the serious concern in most engineering problems since the
distributions of permeability and porosity in heterogeneous parameters that define the system may not be very difficult to
and multiphase petroleum reservoirs by matching the dynamic obtain by direct measurement. Unfortunately this is not the
behavior. The dynamic data is in the form of field case in reservoir engineering, where the system, that is the oil-
measurements from well testing, production history, gas reservoir, is physically inaccessible many thousands of feet
interpreted 4-D seismic information, and other data such as underground. Thus, any serious attempt at optimization of
correlations between permeability and porosity, geostatistics in reservoir development first requires the determination of the
the form of a variogram model and the inference of large scale parameters of the reservoir and the only way to obtain them is
geological structure. through indirect measurement. The process of inferring the
parameters from the indirect measurements is an inverse or
The issue was posed as an inverse problem and solved by parameter estimation problem. Such is the focus of this work.
using nonlinear parameter estimation. The procedure
developed here is capable of processing all the information Permeability and porosity are the parameters that have the
simultaneously and this results in a fast and efficient method. largest influence in determining the performance of the
The procedure is also able to determine the uncertainty reservoir, and thus, this work addresses the problem of
associated with the estimated permeability and porosity fields. estimating permeability and porosity from a variety of
measurements that are only indirectly related to them.
Examples of different parameter types that may be Estimating permeability and porosity is difficult for the
estimated by this approach include: (a) individual block following reasons:
permeabilities and porosities; (b) geological objects such as
channels and faults; (c) pilot points that form the basis of a • Permeability and porosity have spatial variability.
kriged distribution; and (d) seismic attenuation values from 3- • There are very few sampling locations (wells)
D seismic images. compared to the areal extent of the reservoir.
• Information (data) is scarce.
An important conclusion of this work is that the value of • Measurements are obtained with different
each piece of information does not reside in its isolated use but technologies.
in the value it adds to integrated analysis of the complete set of • The mathematical model of the reservoir is very
information. Thus data that traditionally was considered to be complex, usually consisting of a numerical reservoir
of low information content for reservoir characterization simulation.
2 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
• What are the data we need to determine the reservoir (measurements) under the same set of external conditions. In
parameters. this work the physical system under study is a reservoir. The
• What parameters can be resolved with a given data set. following fundamental laws are relevant to the dynamics of the
• What is the meaning of the calculated parameters. reservoir:
• What is the uncertainty associated with the parameter
estimates. 1. Mass conservation law.
2. Darcy’s law.
Theory 3. Equation of state.
The process of inversion to determine values of reservoir 4. Relative permeability and capillary pressure relationships.
parameters, such as permeability and porosity, from indirect
measurements is referred to as a parameter estimation The mathematical model is constructed by combining these
problem (also referred to as an inverse problem). The usual laws and results in a system of differential equations. In a few
approach to solve the parameter estimation problem is by cases, such as in traditional well testing theory, it is possible to
going through three major steps, not only for the specific case obtain an explicit solution to the differential equations, but in
of this work but for any general problem. These steps are as the general multiphase, multiwell, heterogeneous case this is
follows: not possible and therefore it is necessary to resort to numerical
methods to obtain a solution. In this work the mathematical
model consisted of a numerical reservoir simulation. The
1. To construct a mathematical model. r
2. To define an objective function. behavior d cal of the system is represented by Eqn. 1.
3. To apply a minimization (referred to in this work as the
parameter estimation algorithm). r r r
d cal = d cal ( α ) (1)
Once the mathematical model has been constructed, the r
objective function has been defined, and the minimization where α ∈ R npar is the vector of the parameters of the
algorithm has been chosen, the procedure for inversion works mathematical model. Most or all of the parameters are directly
by the following steps: related to the distribution of permeability and porosity in the
reservoir.
1. Assign an arbitrary, but reasonable, value to the unknown
r
set of parameters α , this is referred to as the first guess. Parameterization of the mathematical model
2. Compute the response of the system with the mathematical Since in our approach we use a numerical simulator our
model. parameter estimation problem is to find the permeability and
3. Compute the objective function, which compares the porosity we should assign each cell of the simulation grid such
calculated response of the system to the actual set of that the calculated data replicates the field observations. This
measurements. If the objective function is less than a problem can be translated to the inverse problem formulation
from two perspectives: pixel and object modeling27,28.
certain predetermined value then STOP. r
4. Use the minimization algorithm to compute a change in Pixel modeling. In this approach the parameters α for the
the set of parameters. If the change in the set of inverse problem are the permeability and porosity at each cell
parameters is less than a certain predetermined value then of the simulation grid. Thus the number of parameters is twice
STOP the number of simulation cells.
5. Return to Step 2. Object modeling. In this approach the permeability and
porosity at each cell of the simulation grid is a function of a set
r
Mathematical Model of parameters α , that is
The physical system under study is represented by a
r
mathematical model that is constructed by applying the ki = ki ( α ) (2)
fundamental physical laws that are relevant to the problem. r
φi = φi ( α ) (3)
The purpose of the mathematical model is to predict with
reasonable accuracy the behavior of the system under different
conditions. The problem of computing the response of the Fig. 5 (a) shows an example of a channel reservoir that can
mathematical model to an external perturbation is referred to be modeled by either of the approaches. There are two main
as the forward problem. The physical properties that remain purposes in object modeling. The first is to preserve the large
invariant for different problems are referred to as parameters scale geological information, that is if we model the reservoir
of the system. The ones that change are referred to as as a channel the result of the inverse problem will always be a
variables. The opposite problem, the inverse problem, consists channel. The second purpose is to reduce the dimension of the
of finding a set of parameters for a given model such that the inverse problem, for example the channel reservoir shown in
predicted behavior of the system replicates the true behavior Fig. 5 (a) can be parameterized with only eight parameters
4 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
description.
The parameter estimation algorithm will converge to
r
Parameter Estimation Algorithms α * when the following conditions are met:
Common to all parameter estimation algorithms is that they try
r
to minimize a discrepancy function (objective function). ∇E( α * ) ≤ ε1 (13)
r
One of the characteristics of reservoir parameter estimation E( α * ) ≤ ε 2 (14)
is that the objective function E is nonlinear with respect to the r
parameters; consequently all the algorithms rely on iterative where ε1 and ε2 are small positive numbers. α * is referred
procedures that minimize by a succession of changes to a to as an optimal point and should provide a set of parameters
given first set of parameters. that results in a good match of the data.
There are many methods30, 34, 35 to minimize the objective Computation of the Sensitivity Coefficients
function. These methods are usually classified depending on As shown in Eqns. 8, 9, 11, and 12. The computation of G
whether they use the gradient of the objective function or not. (Eqn. 10) is crucial to the Gauss-Newton method. The
The gradient of E is defined as: efficient evaluation of this matrix has been the subject of
intensive research2, 3, 4, 5, 14, 21, 23, 25, 27, 28.
∂E
T
r r r
yk Cov{α * } = G − g Cov{d obs }G − gT (28)
Zαk = (20)
i
∂α i
r k +1 Eqns. 27 and 28 can be modified to take into account data
∂f
J = r k +1 (21) weights. The weight matrix W is usually:
∂y
r r −1
∂f k +1 W = Cov{d obs } 2 (29)
D = rk (22)
∂y
rk
[ ]
T Applications
y = p1k , S wk _1 , p2k , S wk _ 2 ,K , pnbloc
k
, S wk _ nblock (23) Some practical examples of the application of the procedure
are shown next.
Variance and Resolution of the Parameter Estimates
Since the reservoir parameter estimation problem we are Channel Model
dealing with is nonlinear (Eqn. 1), it is not possible to have a The data (field observations) are shown in Fig. 2. The
simple way to calculate the covariance matrix of the reservoir was discretized with a 40 × 30 Cartesian mesh. The a
parameters estimates. The importance of the covariance matrix priori large scale information consisted of the knowledge of a
is in that it provides information about the uncertainty channel, thus we used the dynamic object approach. The
associated with each parameter and also about the correlation object channel can be parameterized with eight parameters as
between the parameters. One way to overcome the difficulty shown in Fig. 3.
associated with the nonlinearity is by making a linear In the object model approach the parameters are no longer
approximation to the problem. This type of analysis will the permeability and porosity at each cell of the simulation
provide very valuable qualitative information not only about grid (pixel model) but the parameters that define the object.
the level of uncertainty in the parameter estimates but also in Thus in this approach the object is allowed to “float” in the
the information content in each type of data. The linear reservoir, that is the channel can change shape, translate, and
analysis is performed by assuming that the nonlinear system rotate. Also, the permeability inside and outside the channel
r
can be modeled in the neighborhood of an optimal point α * are considered as parameters. Fig. 4 shows a sequence of
with a first order approximation as: frames that depicts the evolution of the parameters that define
the channel, which reveals two main features of the method.
r r First, is the capability to handle objects in the space. Second,
d cal = Gα + constant (24)
is the speed of the method, in this case we were able to match
r r the data and converge to the “true” reservoir in only 41
where the sensitivity matrix G is computed at α = α * iterations. Fig. 2 shows the match of the data.
6 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
The same data were used for another inverse problem, this was able to recover the true location of the fault after 51
time without using the information about the existence of a iterations.
channel. We kept the 40 × 30 simulation grid but now we
parameterized with 100 parameters, each parameter Kriging Model
representing the permeability of 12 adjacent cells. We refer to We can use the object model approach to solve another type of
this method of parameterization as the “large pixel” model problem by using a technique presented by Fasanino et al.37.
because of the resemblance to the pixel approach. The initial With this technique we can introduce geostatistical
guess was a homogeneous reservoirs. Fig. 5 (b) shows the information into our inverse problem without the need to use
calculated permeability distribution. Fig. 5 (c) shows also the the generalized least square formulation (Eqn. 5). The
boundaries of the channel that was used originally to generate technique works as follows.
the data. This problem was much harder to solve because of If there were a relatively large number of permeability and
the larger number of parameters. It took approximately 400 porosity samples in the reservoir, then it would be possible to
iterations to obtain a good match of the data. Thus a first estimate the permeability in rest of the reservoir by using a
impression is that the object modeling approach was 10 times linear estimation technique, such as kriging, that takes into
faster, but it must be remembered that each iteration of the 100 consideration a predefined spatial correlation of the parameters
parameters requires much longer CPU time. Hence the object given in the form of a covariance matrix. In our case we do not
model approach was actually 1000 times faster than the large have the sample points, therefore we make the sampled
pixel approach. permeability values the unknown parameters in the inverse
problem. The covariance matrix used by the kriging estimator
Black and White 4-D Seismic Data allows us to introduce the a priori information about the
One observation in the previous example is that we were using spatial correlation.
“exact” data. This can be considered unrealistic, especially in The permeability k at simulation cell j is calculated as:
the case of the 4-D seismic data since with the current
npar
technology it is not possible for the geophysicists to prepare an
“exact” map of change of saturation. Most likely they will be kj = ∑λ k
i,j i (30)
i =1
able to prepare a map where they can assert the areas of the
reservoir where there were changes in the saturation but the
where λi,j is the kriging weight calculated from a
magnitude of the changes will be unknown. Thus we used the
predefined variogram model at the location of the npar
data shown in Fig. 1 but instead of using the exact change of
hypothetical sample points.
saturation maps in Fig. 1 (f) and (g) we used the coarse maps
Thus the problem now can be stated as to find a set of
shown in Fig. 6 (b) and (d). Fig. 7 (a) shows the true
permeability values that we should assign to the pilot sample
permeability field, (b) the calculated permeability when
points that result in a kriged reservoir that match the actual
“exact” data is used, and (c) the calculated permeability when
dynamic data.
the 4-D seismic data is used in a coarse “black and white”
format. We see from (c) that the “black and white” data can be
Figs. 9 (a) and (b) show an example of the use of this
used in the approach we developed here and still provide a
approach. In this case we used 48 hypothetical sample points
reasonable description of the reservoir.
in the reservoir. (a) shows the true reservoir and (b) shows the
calculated reservoir when exact data are used. Since in real
Fault Model
reservoirs the appropriate locations or number of pilot points
We can use the object model approach to find the location of
and the variogram model are unknown, we can use this
faults in a reservoir by using a diversified data set similar to
procedure to generate several realizations of the reservoir by
the one depicted in Fig. 1. As a first approximation, we can
using different pilot points and variogram models that are
model a sealing-fault as a rectangle, where the permeability
considered plausible for that particular reservoir. This idea was
inside is very low (10-5 md) and the width is small compared to
tested by solving the same inverse problem with three different
the dimensions of the reservoir. We set free all the other
variogram models that differed from the true variogram in the
parameters that define the rectangle object. Thus the rectangle
azimuth angle. Figs. 9 (c) and (d) show the results when the
can change its length, rotate and translate in space. Fig. 8
variogram is rotated 90° and 45° respectively. Figs. 9 (b), (c),
shows an example of finding the location of a single fault. This
figure shows the location of the fault and the water saturation and (d) all provided a good match of the dynamic data
distribution in the reservoir at the end of the simulated time. generated with the true reservoir shown in (a) even though
We do not use the saturation information in the match. The they look substantially different from each other. The first
first frame shows the first guess for the fault. The first guess is guess in all cases was a homogeneous reservoir. This could be
substantially different than the “true case” (last frame) from considered as the most unfavorable condition because it
the point of view of water saturation (see the well closest to the presumes an unrealistic ignorance of the major features in the
right bottom corner). The procedure developed in this work reservoir.
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 7
Another procedure closely related to the kriging-pilot We start the analysis by first computing the matrix G of the
r r
points procedure is the one referred to as “the self-calibrated sensitivity coefficients at α = α * , then we compute the
method” by Wen, Gomez-Hernandez, Capilla and singular value decomposition of G (or GTG)28, next we
Sahuquillo38. This method was developed in the ground water compute the covariance matrix of the parameter estimates
modeling field but can be adapted for use in the multiphase r
(Eqn. 28). The diagonal elements of Cov{ α * }, that is the
problem of petroleum reservoir modeling. The essence of the
variances of the parameter estimates ( σ α2i ) provide
method is, given a first guess of the reservoir or realization
obtained with a geostatistical tool, to calculate a perturbation information about the uncertainty. We can visualize this
to the field such that its response matches the dynamic data information by translating σ α−1i to the reservoir map. This is
from the real reservoir. The perturbation is calculated as a shown in Fig. 11. The red colors show which section of the
kriging estimate from perturbations at certain predefined reservoir are best resolved by the data, next are the areas in
control or pilot points. Mathematically it is: green. The areas in blue or black are the areas not well defined
by the data, that is the permeability in such areas can take any
npar
k j = k 0j + ∑ λi , j ∆ki
arbitrary value within a large range without affecting the
(31)
i =1
quality of the match. This is a basic analysis, next we calculate
the variance for each parameter and for each data set, both
3-D Seismic Model individually and in combination. The results are summarized
The object model approach can also be used to integrate in Fig. 13, which maps the inverse of the normalized variance
information from 3-D seismic interpretations. For example σ αi
in the reservoir.
Fig. 10 (a) shows a 3-D interpretation (adapted from Ref. 39) . αi
This image was discretized as a 16 color map and then we The following combinations of data were analyzed:
assigned a permeability/porosity value to each color. Thus the
colors themselves become the “objects” in the problem, and 1. DST pressure only.
we use the procedure to determine the permeability/porosity 2. Long term pressure only.
values we should assign to each color in order to match the 3. Pressure only (DST pressure and long term pressure).
data. Fig. 10 shows in (a) the true reservoir and in (b) the 4. Water cut only.
calculated reservoir. In this case we simulated the field 5. Change of saturation only (two 4-D seismic
considering 10 wells. The data used in the inverse problem interpretations).
consisted of field observations from one DST and permanent 6. Pressure and water cut.
gauges at each well over a 200 day period. The simulation grid 7. Pressure and change of saturation.
consisted of a 30 × 30 mesh. 8. Water cut and change of saturation.
9. All (pressure and water cut and change of saturation).
Variance of Parameter Estimates and Value of the
Information The following remarks and observations can be drawn
The previous section showed some examples of solutions to from Fig. 13:
reservoir inverse problems by the integrated use of data of • Combinations of data always reduce the uncertainty. That
different nature. The inverse problem is not completely solved is, adding more uncertain information to a data set does
by providing sets of permeability and porosity values that not downgrade the solution, on the contrary, it always
result in a good match of the data. There are other issues that produces a reduction in the level of uncertainty.
come along with the inverse problem such as the uncertainty
• The addition of 4-D seismic information produced a
associated with each parameter estimate and the value of the
remarkable improvement despite this information having a
information. These issues can be explored with the theory used
relative high uncertainty. This is because the 4-D seismic
to develop Eqns. 24 to 29. A simple example will illustrate the data provides information that is spatially distributed.
use of this analysis.
• The water cut information seems to help little in this
specific problem.
For the sake of simplicity let us assume that we have a
• The long term pressure history seems to provide strong
homogeneous (but unknown) reservoir, and a data set similar
information. The highest sensitivity in the long term
to the one illustrated in Fig. 1. The reservoir was discretized
pressure is concentrated at the time when the water front
with 40 × 30 mesh and parameterized with 100 large pixel
r reaches the wells28. This is confirmed in the color map,
parameters. After a permeability distribution ( α * ) that here the best area determined by the long term pressure is
matches the data is obtained, the first question we would like the area behind the water front.
to answer what is the uncertainty associated to each of the 100 • The DST information alone will not help much in the
parameters we have calculated. inverse problem in the way it was posed, that is the DST
data alone cannot be used to compute 100 spatially
8 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
porosity. The information that can be integrated includes: well The more interesting conclusions obtained from this work
test data, shut-in pressure, long term pressure (from permanent are about the roles of well testing, 4-D seismic data and
gauges), production history (water cut, rates), interpreted 4-D integration.
seismic maps, permeability-porosity correlations, variogram
models and large scale geological information (object Well testing has often been considered as an isolated tool
modeling). This seems to cover most of the information that to solve relatively simple reservoir models in the
can be realistically obtained from a program for reservoir neighborhood of the wells and to improve the productivity.
monitoring with current technology, but it is not necessary to These are very important considerations from the engineering
have all of them to apply the method described in this paper. and economical point of view, but since the well tests are
performed generally during the earlier part of the life of the
In this work the interpreted 4-D seismic information was reservoir then this information is not used later mainly because
assumed to be in the form of maps of change of saturation in of the assumption that the area investigated is small and also
the reservoir. The procedure can be adapted to the case when because of the lack of a mathematical tool to be used in a more
the 4-D seismic interpretation is not related to a simple change complex environment. The analysis performed here shows that
of saturation but to a more complex relationship that could be the well test information can be used to resolve much more
a function of not only saturation but pressure, density, rock complex reservoirs models later in the life of the reservoir, not
type, etc. as a standalone approach but as a piece of information in a
larger data set. The addition of well test information enhances
The method developed in this work relies on the the resolution value of the other information, thus well tests
computation of the sensitivity coefficients for the field early in the life of the reservoir will also be useful later for
observations with respect to the parameters of the inverse reservoir characterization.
problem. This computation might be considered an
unnecessary burden since there are other methods to solve the The value of 4-D seismic data is due to the fact that it
inverse problem without the sensitivity coefficients, but we provides information that is spatially distributed in the
found such an effort worthwhile for the following reasons: reservoir in contrast to the other data that are localized at the
• The sensitivity coefficients are necessary to perform the wells. Even so, it was found that the 4-D information does not
variance analysis, and this is key to understanding the assist much in reservoir characterization when it is considered
meaning of what it is being calculated and to plan alone. The value of this type of information appears when it is
reservoir monitoring. The sensitivity coefficients are also combined with other more traditional data such as production
needed to find the most adequate parameterization of the history. Also it was found that this kind of information does
reservoir for a given amount of information. not need to be very accurate since data in coarse “black and
• The method to compute the sensitivity coefficients, as white” format can still be useful. In all the examples shown in
developed in this work, is relatively simple to implement this work it was assumed that two consecutive 3-D seismic
when it is possible to have access to the numerical surveys were performed after production had begun in the
reservoir simulator code. Thus the complexity of reservoir. If the first survey were to be performed before any
modeling the reservoir is left to the simulation part and production then it is very likely that the 4-D information will
not to the inverse problem as it would be the case for the provide even more value to the reservoir inverse problem.
optimal control approach.
• The Gauss-Newton approach is very efficient as a The value of data integration can be evaluated with the
parameter estimation algorithm, compared to direct search variance analysis. It was shown that the addition of data
methods that do not require sensitivity coefficients. always enhances the overall resolution of the parameters. The
• The computation of the sensitivity can be performed with addition of data that seems to provide little information value
high efficiency, specially in the case of object or large when considered in isolation results in a large increase in the
pixel modeling. The efficiency can also be enhanced by resolution of the other data. A dramatic example is in the case
parallel computing. of the shut-in pressure data, which includes only a very small
number of pressure measurements obtained when the wells are
From the results of our numerical experiments it is possible closed at the surface for a short period. This information when
to conclude that the method can be utilized as a tool for considered by itself cannot be used to solve even the simplest
reservoir characterization. The method seems to be more models, but its value is considerable when it is used in an
useful when the dynamic data set consists of a large number of integrated procedure such as the one developed in this work.
measurements that are difficult to honor with the existing
geostatistical approaches; this may be the case in mature fields Acknowledgments
that have already gone through a secondary recovery process The authors gratefully acknowledge the financial support from
and in which data have been gathered over many years. the Stanford University Research Consortium for Innovation in
Well Test Analysis (SUPRI-D).
10 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
References
Nomenclature 1. Seymour, R.H and Barr, F.J.: “Seabed-Seismic 4D-Collection
r
α = parameter model vector Method for Reservoir Monitoring,” JPT (January 1997) Vol. 49,
r No. 1, 40-41.
α * = optimal parameter model vector
r 2. Jacquard, P., and Jain, C.: “Permeability Distribution from Field
α prior = a priori parameter model vector Pressure Data,” Soc. Pet. Eng. J. (Dec. 1965) 281; Trans.,
k= permeability AIME, Vol. 234.
k0 = permeability from realization 3. Carter, R.D., Kemp, L.F., Pierce, A.C., and Williams, D.L:
“Performance Matching With Constraints,” Soc. Pet. Eng. J.
λ= kriging weight
(Apr. 1974) 187; Trans., AIME, Vol. 257.
φ= porosity 4. Dogru, A.H., and Seinfeld, J.H.: “Comparison of Sensitivity
σ2 = variance Coefficient Calculation Methods in Automatic History
∇E = gradient of E Matching,” Soc. Pet. Eng. J. (Oct. 1981) 551.
Cd= data covariance matrix 5. Carter, R.D., Kemp, L.F., and Pierce, A.C.: “Discussion of
Cα = a priori parameter covariance matrix Comparison of Sensitivity Coefficient Calculation Methods in
Cov{ } = covariance matrix operator Automatic History Matching,” Soc. Pet. Eng. J. (Apr. 1982)
205.
DST = drill stem test
r 6. Chen, W.H., Gavalas, G.R., Seinfeld, J.H., and Wasserman,
d = data vector M.L: “A New Algorithm for Automatic History Matching,” Soc.
E= objective function Pet. Eng. J. (Dec. 1974) 593; Trans., AIME, Vol. 257.
r
f = material balance equations vector 7. Chavent, G., Dupuy, M., and Lemonnier, P.: “History Matching
by Use of Optimal Theory,” paper SPE 4627 presented at the
G= matrix of sensitivity coefficients 1973 SPE Annual Technical Conference and Exhibition, Las
H= Gauss-Newton approximation to the Hessian matrix Vegas, NV, Sep. 30 - Oct. 3.
of objective function 8. Watson, A.T., Seinfeld, J.H., Gavalas, G.R., and Woo, P.T.:
J= Jacobian of material balance equations “History Matching in Two-Phase Petroleum Reservoirs,” paper
R= resolution matrix SPE 8250 presented at the 1979 SPE Annual Technical
Sw = water saturation Conference and Exhibition, Las Vegas, NV, Sep. 23-26.
9. Yang, P.H, and Watson, A.T.: “Automatic History Matching
S= singular value matrix
With Variable-Metric Methods,” paper SPE 16977 presented at
t= time the 1987 SPE Annual Technical Conference and Exhibition,
U= factor from singular value decomposition Dallas, TX, Sep. 27-30.
V= factor from singular value decomposition 10. Kamal, M.M.: “Use of Pressure Transients to Describe
W = weighting matrix Reservoir Heterogeneity,” JPT, (Aug. 1979) 1061, Trans.,
r
y = vector of pressure and saturation in the simulation AIME, Vol. 257, SPE Reprint Series No. 14.
11. Tang, Y.N., and Chen, Y.M.: “Application of GPST Algorithm
mesh
r to History Matching of Single-Phase Simulator Models,” paper
Z = sensitivity vector SPE 13410, 1985.
12. Tang, Y.N., and Chen, Y.M.: “Generalized Pulse-Spectrum
Technique for 2-D and 2-Phase History Matching,” Applied
Subscripts Numerical Mathematics (1989) Vol. 5, 529-539.
obs = observed value 13. Oliver, D.S.: “Incorporation of Transient Pressure Data Into
calc = calculated value Reservoir Characterization,” In Situ (1994) Vol. 18, 243-275.
npar = number of parameters 14. Chu, L., Reynolds, A.C., and Oliver, D.S.: “ Computation of
Sensitivity Coefficients for Conditioning The Permeability Field
nblocks = number of blocks in simulation grid
to Well-Test Pressure Data,” In Situ (1995) Vol. 19, 179-223.
p = number of nonzero singular values 15. Oliver, D.S.: “Multiple Realizations of the Permeability Field
T = transpose of matrix From Well Test Data,” paper SPE 27970 presented at the 1994
University of Tulsa Centennial Petroleum Engineering
Symposium, Tulsa, OK, Aug. 29-31.
Superscripts 16. Chu, L., Reynolds, A.C., and Oliver, D.S.: “Reservoir
r Description From Static and Well-Test Data Using Efficient
= vector
-g = generalized inverse of a matrix Gradient Methods,” paper SPE 29999 presented at the 1995
SPE International Meeting on Petroleum Engineering, Beijing,
k = iteration index
PR China, Nov. 1995.
p = pressure 17 Reynolds, A.C., He, N., Chu, L., and Oliver, D.S.:
wc = water cut “Reparameterization Techniques for Generating Reservoir
∆Sw= change of water saturation Descriptions Conditioned to Variograms and Well-Test Pressure
Data,” paper SPE 30588 presented at the 1995 SPE Annual
Technical Conference and Exhibition, Dallas, TX, Oct. 22-25.
18. Oliver, D.S.: “A Comparison of the Value of Interference and
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 11
Well-Test Data for Mapping Permeability and Porosity,” In Situ Theory, Academic Press, Inc., San Diego, CA, 1989.
(1996) Vol. 20, 41-59. 34. Gill, P.E., Murray, W., and Wright, M.H.: Practical
19. Oliver, D.S., He, N. and, Reynolds, A.C.: Conditioning Optimization, Academic Press, San Diego, CA, 1981.
Permeability Fields to Pressure Data,” paper presented at the 5th 35. Bard, Y.: Nonlinear Parameter Estimation, Academic Press,
European Conference on the Mathematics of Oil Recovery, New York, NY, 1970.
Leoben, Austria, (1996), Sep. 3-6. 36. Jackson, D.: “Interpretation of Inaccurate, Insufficient and
20. He, N., Reynolds, A.C., and Oliver, D.S.: “Three-Dimensional Inconsistent Data,” Geophysical Journal of the Royal
Reservoir Description from Multiwell Pressure Data,” paper Astronomical Society,(1972), Vol. 28, 97-109.
SPE 36509 presented at the 1996 SPE Annual Technical 37. Fasanino, G., Molinard, J., and de Marsily, G.: “Inverse
Conference and Exhibition, Denver, CO, Oct. 6-9. Modeling in Gas Reservoirs,” paper SPE 15592 presented at the
21. Anterion, F., Eymard, R., and Karcher, B.: “Use of Parameter 1986 SPE Annual Technical Conference and Exhibition, New
Gradients for Reservoir History Matching,” paper SPE 18433 Orleans, LA, Oct. 5-8.
presented at the 1989 SPE Symposium on Reservoir Simulation, 38. Wen, X., Gomez, J., Capilla, J. and Sahuquillo, A:
Houston, TX, Feb. 6-8. “Significance of Conditioning to Piezometric Head Data for
22. Bissell, R., Sharma, Y., and Killough, J.E.: “History Matching Predictions of Mass Transport in Groundwater Modeling,”
Using the Method of Gradients: Two Case Studies,” paper SPE Mathematical Geology, (1996), Vol. 28, No. 7, 951-968.
28590 presented at the 1994 SPE Annual Technical Conference 39. Sheriff, R.E, and Geldart, L.P.: Exploration Seismology,
and Exhibition, New Orleans, LA, Sep. 25-28. Cambridge University Press, New York, NY, 1995.
23. Tan, T.B., and Kalogerakis, N.: “A Fully Implicit, Three-
Dimensional, Three-Phase Simulator with Automatic History-
Matching Capability,” paper SPE 21205 presented at the 1991
SPE Symposium on Reservoir Simulation, Anaheim, CA, Feb.
17-20.
24. Tan, T.B.: “A Computational Efficient Gauss-Newton Method
for Automatic History Matching,” paper SPE 29100 presented
at the 1995 SPE Symposium on Reservoir Simulation, San
Antonio, TX, Feb. 12-15.
25. Killough, J.E., Sharma, Y., Dupuy, A., and Bissell, R.: “A
Multiple Right Hand Side Solver for History Matching,” paper
SPE 29119 presented at the 1995 SPE Symposium on Reservoir
Simulation, San Antonio, TX, Feb. 12-15.
26. Bissell, R.: “History Matching A Reservoir Model by the
Positioning of Geological Objects,” paper presented at the 5th
European Conference on the Mathematics of Oil Recovery,
Leoben, Austria, (1996), Sep. 3-6.
27. Landa, J.L., Kamal, M.M., Jenkins, C.D., and Horne, R.N.:
“Reservoir Characterization Constrained to Well Test Data: A
Field Example,” paper SPE 36511 presented at the 1996 SPE
Annual Technical Conference and Exhibition, Dallas, TX, Oct.
6-9.
28. Landa, J.L.: “Reservoir Parameter Estimation Constrained to
Pressure Transients, Performance History and Distributed
Saturation Data,” Ph.D. dissertation, Stanford University,
Stanford, CA, 1997.
29. Sultan, A.J., Ounes, A. and, Weiss, W.W.: “ Automatic History
Matching for an Integrated Reservoir Description and
Improving Oil Recovery,” paper SPE 27712 presented at the
1994 SPE Permian Basin Oil and Gas recovery Conference,
Midland, TX, Mar. 16-18
30. Ounes, A, Bhagavan, S., Bunge, P.H., and Travis, B.J.:
“Application of Simulated Annealing and Other Global
Optimization Methods to Reservoir description: Myths and
Realities,” paper SPE 28415 presented at the 1994 SPE Annual
Technical Conference and Exhibition, New Orleans, Sep. 25-28.
31. Tarantola, A., and Valette, B.: “Generalized Nonlinear Inverse
Problems Solved Using The Least Squares Criterion,” Reviews
of Geophysics and Space Physics (May 1982) Vol. 20, No. 2,
219-232
32. Tarantola, A.: Inverse Problem Theory - Methods for Data
Fitting and Model Parameter Estimation, Elsevier Science
Publishers, Amsterdam 1987.
33. Menke, W.: Geophysical Data Analysis: Discrete Inverse
12 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653
8000
c c 6000
Permeability
? 4000
log(k)
c c 2000
φ - Porosity
0
(a) (h)
c c Pressure
Water Cut Production History
DST Well # 1
Well # 1
c c
c c
(b) (i) (m)
c c c c
(f) Production History
Well # 2
c c DST
c c Well # 2
c c c c
DST Production History
(g) Well # 3 Well # 3
c c
(d) (k) (o)
c c 0.6
DST
Saturation
Well # 4
0.4
Injector Injection History
c c 0.2
Well # 4 - Injector
(a): Reservoir geometry. (b)-(e): Water saturation distribution as a function of time. (f): Change of water saturation between (b)
and (c). (g): Change of water saturation between (c) and (d). (h): Permeability-Porosity relationship. (i)-(l): Pressure vs. time for
DST’s. (m)-(p): Long term pressure and watercut vs. time.
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 13
a a a a a a
True Reservoir Calculated Reservoir 15000
a a a a a a a a a a
Permeability
10000
Permeability
Iter #7 E= 15,617 Iter #11 E= 14,506 Iter #16 E= 14,340
5000
a a Field
a a a a a a a a
a a a a
a a a a a a
Change of
Match a a Saturation #1
a a
0.30
Iter #19 E= 14,360 Iter #22 E= 13,680 Iter #25 E= 13,510
0.25
of the Data
∆Sw
a a a a a a
a a a a
0.20
0.15
Change of a a a a a a
a a Saturation #2
a a Iter #28 E= 13,252 Iter #31 E= 12,756 Iter #34 E= 10,530
a aa aaaaa aaaaaa aa
a
a aa a a a a a a
a
True Pressure
a DST
True Water Cut Pressure & Water-Cut History
aaaa Well # 1 a
a Calculated Values Well # 1 a
aa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a aaaaaa a a a a a a
a aaa
a a a aaa
aaaaa a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a
aaaa
a Iter #36 E= 2,778 Iter #38 E= 71 Iter #41 E= 0.01
aa
15000
a
a
a a a a aa a
aaaa
aaa a a a a a a
a a aa a a a a
a a a a
10000
aaaaaaaaaa
aa
Pressure & Water-CutaHistory
aa aa a a aa a
aaaaa a
md
Well # a2 a a
a a a a a
DST a a a 5000
a a a a a a
Well # 2 a
a a
a a a a a a a a a a a a a a a a
aaaaa a a a a a a a a a a a a a a a a a a a a a
0
aaaa
aa
a aa aaaa aaaaaaa
a a
DST aaaaaaa
Pressure & Water-Cut History a a a a a a a a Fig. 4 Channel model. Evolution of parameters.
a
a Well # 3 a a a
a a aa
Well # 3 a a
aaa a a a a a a a a a a a a a a a a a a a a a a
a a a a a aaa
a a a a a a a a
a a a a
aaaaa a a a a a a a a a a a a a a a a a a a a a a a a a a True Permeability
aa
a
a DST
aa a a a a
a a a a a a a
a a
Well # 4 a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a Injector Pressure History
a aaaa aaaa
a Well # 4 - Injector
a a aa a a aaa aaa
aa a
(a)
(c)
Calculated Permeability
15000
a a
Curve Analysis - Parametric Channel Model 10000
md
α5
5000
α2 = kmatrix a a
α3
0
(b)
-1
π α4
Calculated Permeability
α1 =
kch
anne
l
a a
α8
α7
a a
α6 (c)
c c c c c c
0.3 c c c c c c
∆Sw
0.2
Iter #7 Iter #11 Iter #15
0.1 c c c c c c
0.0
(a) (b)
c c c c c c
∆Sw
0.2
0.1 c c c c c c
0.0
(c) (d) Iter #35 Iter #42 Iter #45
c c c c c c
Fig. 6 4-D seismic data in “black and white” format.
c c c c c c
a a
∆Sw
0.4
c c c c c c 0.2
(a)
True Permeability - α=30° Calculated Permeability - α=30°
Calculated Permeability - Exact ∆Sw 6000
a a a a 103
a a 5000
md
4000
10
md
3000
2000 a a a a
a a 1000
(a) (b)
0
(b)
Calculated Permeability - α=120° Calculated Permeability - α=75°
a a
a a a a
a a (c) (d)
True k Calculated k
True Permeability
104
10000
c c c c
a
c 103
k md
5000
c
a c
k - md
102 c c c c
c c c (a) (c)
c 10
c c c c
1
k/σk md/md
(a)
1
Calculated Permeability
c c c c
a 10-1
c (b) (d)
c Fig. 12 Variance maps: (a): True reservoir . (b): k/σ map for true
a c
reservoir. (c): Calculated reservoir. (d): k/σ map for calculated reservoir
c c c
c
c
(b)
Map of σ-1
1
c c
10-1
1/σ md-1
10-2
c c
c well location
c c c c
c c c c
k/σ scale
Data: Pressure
10
c c
k/σk md/md
1
c c
10-1
Data: Pressure + Water Cut Data: Pressure + Water Cut + ∆Sw Data: Pressure + ∆Sw
c c c c c c
c c c c c c
c c c c c c
c c c c c c
k/σ scale
Data: DST Pressure
10
c c
k/σk md/md
1
10-1
10-2
c c
10-3
Data: DST + Shut-in Pressure Data: DST + Shut-in + Long Term Pressure Data: DST + Long Term Pressure
c c c c c c
c c c c c c
Data: Shut-in Pressure Data: Shut-in + Long Term Pressure Data: Long Term Pressure
c c c c c c
c c c c c c