This action might not be possible to undo. Are you sure you want to continue?

# SPE 38653 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

Copyright 1997, Society of Petroleum Engineers, Inc. This paper was prepared for presentation at the 1997 SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, 5–8 October 1997. 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.

becomes useful and enhances the value of the data set as a whole. Introduction Devising the optimal strategy for the development of an oil or gas reservoir is an important and difficult task. Many mathematical techniques for optimization can be used to deal with problems in engineering and economics systems. These techniques assume that we have a fairly complete understanding of the problem and also that we can construct a mathematical model that predicts the system’s performance accurately in time under different scenarios; this is not a serious concern in most engineering problems since the parameters that define the system may not be very difficult to obtain by direct measurement. Unfortunately this is not the case in reservoir engineering, where the system, that is the oilgas reservoir, is physically inaccessible many thousands of feet underground. Thus, any serious attempt at optimization of reservoir development first requires the determination of the parameters of the reservoir and the only way to obtain them is through indirect measurement. The process of inferring the parameters from the indirect measurements is an inverse or parameter estimation problem. Such is the focus of this work. Permeability and porosity are the parameters that have the largest influence in determining the performance of the reservoir, and thus, this work addresses the problem of estimating permeability and porosity from a variety of measurements that are only indirectly related to them. Estimating permeability and porosity is difficult for the following reasons: • • • • • Permeability and porosity have spatial variability. There are very few sampling locations (wells) compared to the areal extent of the reservoir. Information (data) is scarce. Measurements are obtained with different technologies. The mathematical model of the reservoir is very complex, usually consisting of a numerical reservoir simulation.

Abstract This paper deals with the problem of estimating the distributions of permeability and porosity in heterogeneous and multiphase petroleum reservoirs by matching the dynamic behavior. The dynamic data is in the form of field measurements from well testing, production history, interpreted 4-D seismic information, and other data such as correlations between permeability and porosity, geostatistics in the form of a variogram model and the inference of large scale geological structure. The issue was posed as an inverse problem and solved by using nonlinear parameter estimation. The procedure developed here is capable of processing all the information simultaneously and this results in a fast and efficient method. The procedure is also able to determine the uncertainty associated with the estimated permeability and porosity fields. Examples of different parameter types that may be estimated by this approach include: (a) individual block permeabilities and porosities; (b) geological objects such as channels and faults; (c) pilot points that form the basis of a kriged distribution; and (d) seismic attenuation values from 3D seismic images. An important conclusion of this work is that the value of each piece of information does not reside in its isolated use but in the value it adds to integrated analysis of the complete set of information. Thus data that traditionally was considered to be of low information content for reservoir characterization

2

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

The data set defining a reservoir is derived from measurements that are related to the permeability and porosity according to different laws or complex relationships. It is common to try to solve the inverse problem by working on each set of measurements independently of the others. In general each individual interpretation may not be fully consistent with the others, and as a result the final interpretation comes after a long process of iteration to ensure compatibility. As a final step the interpretations are validated by history matching, since it would not be reasonable to rely on a model that could not predict the past performance. Since data is being collected almost continuously, the process of updating the reservoir model never ends. During the producing life of a reservoir, data of different nature are always being collected. These data can be classified as static or dynamic depending on their association with the movement or flow of fluids in the reservoir. Data that have originated from geology, electrical logs, core analysis, fluid properties, seismic and geostatistics can be generally classified as static, whereas the information originating from well testing, pressure shut-in surveys, production history, bottom hole pressure from permanent gauges, water-cut, and gas-oil ratio (GOR) can be classified as dynamic. A special consideration needs to be applied in the case of 4-D1 seismic information, which is a relatively new technology developed in the field of geophysics. With this process it is possible to estimate the areal distribution of change of saturation in the reservoir due to the production or injection of fluids. Since 4-D seismic information is related to the movement of fluids in the reservoir it can be classified as dynamic data. One of the outstanding features of the 4-D seismic information is that it is areally distributed whereas the other dynamic data are available only at the location of the production or injection wells. The parameter estimation problem would not only be faster but also more reliable if it were performed with a process that uses all or at least most of the information in the reservoir data set simultaneously. The process of handling different data simultaneously is known as data integration. So far, most of the success in data integration has been obtained with static information. Remarkably, it has not yet become common to completely or systematically integrate dynamic data with static data and it is currently the subject of major research effort in several places2-30. This work addresses this specific problem and represents a number of steps in the direction of full integration. Problem Statement An example of the type of problem we want to solve has been summarized in Fig. 1, where (a) shows a two-dimensional

reservoir map in which the permeability and porosity are the unknowns to be determined. There are four wells (marked with triangles), one of them is injecting water and the others are producing. The production rates and injection rates are known. Figs. 1 (b)-(e) show the water distribution in the reservoir as a function of time, this is not known to us but is shown here only to illustrate the problem. Figs. 1 (f)-(p) show graphically the field observations that can be available to us. These observations are: • 4-D seismic interpretation1. Fig. 1 (f) and (g) show two changes of water saturation (∆Sw) maps. These maps are assumed to have been made available after three consecutive 3-D seismic surveys. The seismic surveys were performed at the time when the reservoir was in states shown in (b), (c) and (d). The geophysical interpretation cannot provide us with the value of the water saturation at each time, but it can provide the change of saturation in the reservoir between two surveys. • Permeability-porosity correlation. Fig. 1 (h) shows such a correlation. This correlation may be obtained from measurements in cores from the wells. The maximum and minimum values possible for the permeability and porosity are also specified, these values may come from a priori geological information • DST pressures. Drill stem tests (DST) have been performed at each well and the information is available (Figs. 1 i-l). The information includes a relatively large number of pressure measurements over a small period of time. Each DST consisted of a single flow followed by a build-up. The DST’s are performed early in the productive life of the reservoir, thus only a single phase (oil) was produced. • Pressure from permanent gauges. The solid lines in Figs. 1 (m)-(p) show the bottom hole pressure at each well as a function of time. At late time the permanent gauges recorded a simulated shut-in pressure at each well. These events can be planned or may be due to operational reasons. The shut-in is of shorter duration than the DST, and fewer measurements are made • Water cut. The dotted lines in Fig. 1 (m)-(p) show the water cut measured at each well. This assumes that production of each well is being measured individually and continuously. Other information that is available includes: • Fluid and rock properties. Except for the permeability and porosity, properties such as viscosity, density, relative permeability and compressibility are known. • Large scale geological information. • Geostatistical information. Such as a variogram model. Our interest is not only in finding distributions of permeability and porosity in the reservoir that result in a match of the data but also in other issues that are as important as the match. Thus we look at:

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

3

• • • •

What are the data we need to determine the reservoir parameters. What parameters can be resolved with a given data set. What is the meaning of the calculated parameters. What is the uncertainty associated with the parameter estimates.

(measurements) under the same set of external conditions. In this work the physical system under study is a reservoir. The following fundamental laws are relevant to the dynamics of the reservoir: 1. 2. 3. 4. Mass conservation law. Darcy’s law. Equation of state. Relative permeability and capillary pressure relationships.

Theory The process of inversion to determine values of reservoir parameters, such as permeability and porosity, from indirect measurements is referred to as a parameter estimation problem (also referred to as an inverse problem). The usual approach to solve the parameter estimation problem is by going through three major steps, not only for the specific case of this work but for any general problem. These steps are as follows: 1. 2. 3. To construct a mathematical model. To define an objective function. To apply a minimization (referred to in this work as the parameter estimation algorithm).

The mathematical model is constructed by combining these laws and results in a system of differential equations. In a few cases, such as in traditional well testing theory, it is possible to obtain an explicit solution to the differential equations, but in the general multiphase, multiwell, heterogeneous case this is 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 r behavior d cal of the system is represented by Eqn. 1.

r r r d cal = d cal ( α )

(1)

Once the mathematical model has been constructed, the objective function has been defined, and the minimization algorithm has been chosen, the procedure for inversion works by the following steps: 1. 2. 3. Assign an arbitrary, but reasonable, value to the unknown r set of parameters α , this is referred to as the first guess. Compute the response of the system with the mathematical model. Compute the objective function, which compares the calculated response of the system to the actual set of measurements. If the objective function is less than a certain predetermined value then STOP. Use the minimization algorithm to compute a change in the set of parameters. If the change in the set of parameters is less than a certain predetermined value then STOP Return to Step 2.

r where α ∈ R npar is the vector of the parameters of the mathematical model. Most or all of the parameters are directly related to the distribution of permeability and porosity in the reservoir.

4.

5.

Mathematical Model The physical system under study is represented by a mathematical model that is constructed by applying the fundamental physical laws that are relevant to the problem. 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 mathematical model to an external perturbation is referred to as the forward problem. The physical properties that remain invariant for different problems are referred to as parameters of the system. The ones that change are referred to as variables. The opposite problem, the inverse problem, consists of finding a set of parameters for a given model such that the predicted behavior of the system replicates the true behavior

Parameterization of the mathematical model Since in our approach we use a numerical simulator our parameter estimation problem is to find the permeability and porosity we should assign each cell of the simulation grid such that the calculated data replicates the field observations. This problem can be translated to the inverse problem formulation from two perspectives: pixel and object modeling27,28. r Pixel modeling. In this approach the parameters α for the inverse problem are the permeability and porosity at each cell of the simulation grid. Thus the number of parameters is twice the number of simulation cells. Object modeling. In this approach the permeability and porosity at each cell of the simulation grid is a function of a set r of parameters α , that is

r ki = ki ( α ) r φi = φi ( α )

(2) (3)

Fig. 5 (a) shows an example of a channel reservoir that can be modeled by either of the approaches. There are two main purposes in object modeling. The first is to preserve the large scale geological information, that is if we model the reservoir as a channel the result of the inverse problem will always be a channel. The second purpose is to reduce the dimension of the inverse problem, for example the channel reservoir shown in Fig. 5 (a) can be parameterized with only eight parameters

4

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

which is fewer than and independent of the number of cells in the simulation mesh. This parameterization is illustrated in Fig. 3. Objective Function The objective function is a measure of the discrepancy between the information, that is the data, and the response calculated by the mathematical model using the current set of parameters. Eqns. 4 and 5 show two forms of the objective function that are commonly used.

r H GN ∆α = −∇E

(7)

HGN is Gauss-Newton approximation to the Hessian matrix of E (matrix of the second derivatives of E). When Eqn. 4 is used, ∇E and HGN are calculated as follows: r r ∇E = −2G TW ( d obs − d cal ) (8)

H GN = 2G TWG

(9)

r r r r T E = d obs − d calc W d obs − d calc

(

) (

T

)

)

(4)

E=

r 1 r d obs − d calc 2 1 r r α − α prior 2

(

(

)

r − d calc + d T r −1 r C α α − α prior

−1 obs

) C (dr

(5)

**r where G is the matrix of the first derivatives of d cal , also referred to as sensitivity coefficients. r ∂d cal G= r (10) ∂α
**

When Eqn. 5 is used ∇E and HGN are calculated as follows:

Eqn. 4 is referred to as the weighted least square problem. Eqn. 5 is referred to as the generalized least square problem, this form was developed from theory of probability 31, 32, 33 and was used by Oliver13 to integrate well test data into reservoir description. Parameter Estimation Algorithms Common to all parameter estimation algorithms is that they try to minimize a discrepancy function (objective function). One of the characteristics of reservoir parameter estimation is that the objective function E is nonlinear with respect to the parameters; consequently all the algorithms rely on iterative procedures that minimize by a succession of changes to a given first set of parameters. There are many methods30, 34, 35 to minimize the objective function. These methods are usually classified depending on whether they use the gradient of the objective function or not. The gradient of E is defined as:

**r r r r − − ∇E = −G T C d 1( d obs − d cal ) + Cα 1( α − α prior )
**

H GN = G C G + C

T −1 d −1 α

(11) (12)

The parameter estimation algorithm will converge to r α * when the following conditions are met:

r ∇E( α * ) ≤ ε1 r E( α * ) ≤ ε 2

(13) (14)

r where ε1 and ε2 are small positive numbers. α * is referred to as an optimal point and should provide a set of parameters that results in a good match of the data.

Computation of the Sensitivity Coefficients As shown in Eqns. 8, 9, 11, and 12. The computation of G (Eqn. 10) is crucial to the Gauss-Newton method. The efficient evaluation of this matrix has been the subject of intensive research2, 3, 4, 5, 14, 21, 23, 25, 27, 28. Because of the different types of field observations it is necessary to compute the sensitivity of the calculated data with respect to the parameters of the problem, that is:

∂E ∇E = r ∂α

T

(6)

In this work we used the Gauss-Newton method, which is classified as a gradient method. The Gauss-Newton algorithm was combined with the Marquardt stabilization method to compute the direction for the change in the parameters, and with a linear search procedure to calculate the size of the change of parameters at each iteration. The constraints on the parameters were included by using penalty functions. At each iteration of the Gauss-Newton algorithm a linear systems of equations is solved (Eqn. 7):

**r ∂p ( t ) si p ( t ) = i r ∂α r wc ∂wci ( t ) si ( t ) = r ∂α ∂∆S w j ( t ) r s j∆S ( t ) = r ∂α
**

w

(15) (16) (17)

where pi is the pressure measured in the i well, wci is the

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

5

water cut measured in the i well, and ∆Sw j is the change of water saturation in the j cell of the simulation grid All the field observations are directly related to the pressure and saturation calculated by the numerical reservoir simulator. Thus we first compute the sensitivity of the calculated pressure and saturation with respect to the parameters of the problem, and then convert to the data dimension. In this work the sensitivity coefficients are calculated with a multiphase extension of the Jacobian method shown in Appendices A and B of Ref. 27 and detailed in Ref. 28. The method is a variant of the “gradient simulator” approach described by Anterion et al.21 The method is relatively straightforward to implement and can be applied to both pixel and object modeling. r Thus the vector of sensitivities Z with respect to the parameter αi at the time step k+1 of the time discretization in the simulator is obtained from Eqn. 1828.

After this approximation it is possible to apply the theory of resolution and variance developed for linear models33,36. This theory is based on the singular value decomposition of G.

G = USV T = U p S pV pT

(25)

where U and V are orthogonal matrices and S is a diagonal matrix of the singular values of G. p is the number of nonzero singular values. Up, Vp and Sp are the matrices constructed with the columns of U, V and S corresponding to the nonzero singular values. The generalized inverse G-g is calculated as:

− G − g = V p S p 1U T p

(26)

Thus the resolution matrix R is calculated as:

**rk rk r JZα +1 = DZα + Wαk +1
**

i i i

R = V pV pT

(27)

(18) (19) (20) (21) (22)

with

**r0 r Zα = 0 r rk yk Zα = ∂α i r k +1 ∂f J = r k +1 ∂y r ∂f k +1 D = rk ∂y rk k k k k k y = p1k , S w _1 , p2 , S w _ 2 ,K , pnbloc , S w _ nblock
**

i

i

and the covariance matrix of the parameter estimates r Cov{ α * } as:

r r Cov{α * } = G − g Cov{d obs }G − gT

(28)

Eqns. 27 and 28 can be modified to take into account data weights. The weight matrix W is usually: r −1 W = Cov{d obs } 2 (29) Applications Some practical examples of the application of the procedure are shown next. Channel Model The data (field observations) are shown in Fig. 2. The reservoir was discretized with a 40 × 30 Cartesian mesh. The a priori large scale information consisted of the knowledge of a channel, thus we used the dynamic object approach. The object channel can be parameterized with eight parameters as shown in Fig. 3. In the object model approach the parameters are no longer the permeability and porosity at each cell of the simulation grid (pixel model) but the parameters that define the object. Thus in this approach the object is allowed to “float” in the reservoir, that is the channel can change shape, translate, and rotate. Also, the permeability inside and outside the channel are considered as parameters. Fig. 4 shows a sequence of frames that depicts the evolution of the parameters that define the channel, which reveals two main features of the method. First, is the capability to handle objects in the space. Second, is the speed of the method, in this case we were able to match the data and converge to the “true” reservoir in only 41 iterations. Fig. 2 shows the match of the data.

[

]

T

(23)

Variance and Resolution of the Parameter Estimates Since the reservoir parameter estimation problem we are dealing with is nonlinear (Eqn. 1), it is not possible to have a simple way to calculate the covariance matrix of the parameters estimates. The importance of the covariance matrix is in that it provides information about the uncertainty associated with each parameter and also about the correlation between the parameters. One way to overcome the difficulty associated with the nonlinearity is by making a linear approximation to the problem. This type of analysis will provide very valuable qualitative information not only about the level of uncertainty in the parameter estimates but also in the information content in each type of data. The linear analysis is performed by assuming that the nonlinear system r can be modeled in the neighborhood of an optimal point α * with a first order approximation as:

r r d cal = Gα + constant

r r where the sensitivity matrix G is computed at α = α *

(24)

6

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

The same data were used for another inverse problem, this time without using the information about the existence of a channel. We kept the 40 × 30 simulation grid but now we parameterized with 100 parameters, each parameter representing the permeability of 12 adjacent cells. We refer to this method of parameterization as the “large pixel” model because of the resemblance to the pixel approach. The initial guess was a homogeneous reservoirs. Fig. 5 (b) shows the calculated permeability distribution. Fig. 5 (c) shows also the boundaries of the channel that was used originally to generate the data. This problem was much harder to solve because of the larger number of parameters. It took approximately 400 iterations to obtain a good match of the data. Thus a first impression is that the object modeling approach was 10 times faster, but it must be remembered that each iteration of the 100 parameters requires much longer CPU time. Hence the object model approach was actually 1000 times faster than the large pixel approach. Black and White 4-D Seismic Data One observation in the previous example is that we were using “exact” data. This can be considered unrealistic, especially in the case of the 4-D seismic data since with the current technology it is not possible for the geophysicists to prepare an “exact” map of change of saturation. Most likely they will be able to prepare a map where they can assert the areas of the reservoir where there were changes in the saturation but the magnitude of the changes will be unknown. Thus we used the data shown in Fig. 1 but instead of using the exact change of saturation maps in Fig. 1 (f) and (g) we used the coarse maps shown in Fig. 6 (b) and (d). Fig. 7 (a) shows the true permeability field, (b) the calculated permeability when “exact” data is used, and (c) the calculated permeability when 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 used in the approach we developed here and still provide a reasonable description of the reservoir. Fault Model We can use the object model approach to find the location of faults in a reservoir by using a diversified data set similar to the one depicted in Fig. 1. As a first approximation, we can model a sealing-fault as a rectangle, where the permeability inside is very low (10-5 md) and the width is small compared to the dimensions of the reservoir. We set free all the other parameters that define the rectangle object. Thus the rectangle can change its length, rotate and translate in space. Fig. 8 shows an example of finding the location of a single fault. This figure shows the location of the fault and the water saturation distribution in the reservoir at the end of the simulated time. We do not use the saturation information in the match. The first frame shows the first guess for the fault. The first guess is substantially different than the “true case” (last frame) from the point of view of water saturation (see the well closest to the right bottom corner). The procedure developed in this work

was able to recover the true location of the fault after 51 iterations. Kriging Model We can use the object model approach to solve another type of problem by using a technique presented by Fasanino et al.37. With this technique we can introduce geostatistical information into our inverse problem without the need to use the generalized least square formulation (Eqn. 5). The technique works as follows. If there were a relatively large number of permeability and porosity samples in the reservoir, then it would be possible to estimate the permeability in rest of the reservoir by using a linear estimation technique, such as kriging, that takes into consideration a predefined spatial correlation of the parameters given in the form of a covariance matrix. In our case we do not have the sample points, therefore we make the sampled permeability values the unknown parameters in the inverse problem. The covariance matrix used by the kriging estimator allows us to introduce the a priori information about the spatial correlation. The permeability k at simulation cell j is calculated as:

kj =

npar i =1

∑λ

i,j i

k

(30)

where λi,j is the kriging weight calculated from a predefined variogram model at the location of the npar hypothetical sample points. Thus the problem now can be stated as to find a set of permeability values that we should assign to the pilot sample points that result in a kriged reservoir that match the actual dynamic data. Figs. 9 (a) and (b) show an example of the use of this approach. In this case we used 48 hypothetical sample points in the reservoir. (a) shows the true reservoir and (b) shows the calculated reservoir when exact data are used. Since in real reservoirs the appropriate locations or number of pilot points and the variogram model are unknown, we can use this procedure to generate several realizations of the reservoir by using different pilot points and variogram models that are considered plausible for that particular reservoir. This idea was tested by solving the same inverse problem with three different variogram models that differed from the true variogram in the azimuth angle. Figs. 9 (c) and (d) show the results when the variogram is rotated 90° and 45° respectively. Figs. 9 (b), (c), and (d) all provided a good match of the dynamic data generated with the true reservoir shown in (a) even though they look substantially different from each other. The first guess in all cases was a homogeneous reservoir. This could be considered as the most unfavorable condition because it presumes an unrealistic ignorance of the major features in the reservoir.

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

7

Another procedure closely related to the kriging-pilot points procedure is the one referred to as “the self-calibrated method” by Wen, Gomez-Hernandez, Capilla and Sahuquillo38. This method was developed in the ground water modeling field but can be adapted for use in the multiphase problem of petroleum reservoir modeling. The essence of the method is, given a first guess of the reservoir or realization obtained with a geostatistical tool, to calculate a perturbation to the field such that its response matches the dynamic data from the real reservoir. The perturbation is calculated as a kriging estimate from perturbations at certain predefined control or pilot points. Mathematically it is:

We start the analysis by first computing the matrix G of the r r sensitivity coefficients at α = α * , then we compute the singular value decomposition of G (or GTG)28, next we compute the covariance matrix of the parameter estimates r (Eqn. 28). The diagonal elements of Cov{ α * }, that is the 2 variances of the parameter estimates ( σ αi ) provide information about the uncertainty. We can visualize this −1 information by translating σ αi to the reservoir map. This is shown in Fig. 11. The red colors show which section of the reservoir are best resolved by the data, next are the areas in 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 arbitrary value within a large range without affecting the quality of the match. This is a basic analysis, next we calculate the variance for each parameter and for each data set, both individually and in combination. The results are summarized in Fig. 13, which maps the inverse of the normalized variance σ αi in the reservoir. αi The following combinations of data were analyzed: 1. 2. 3. 4. 5. 6. 7. 8. 9. DST pressure only. Long term pressure only. Pressure only (DST pressure and long term pressure). Water cut only. Change of saturation only (two 4-D seismic interpretations). Pressure and water cut. Pressure and change of saturation. Water cut and change of saturation. All (pressure and water cut and change of saturation).

k j = k 0 + ∑ λi , j ∆ki j

i =1

npar

(31)

3-D Seismic Model The object model approach can also be used to integrate information from 3-D seismic interpretations. For example Fig. 10 (a) shows a 3-D interpretation (adapted from Ref. 39) . This image was discretized as a 16 color map and then we assigned a permeability/porosity value to each color. Thus the colors themselves become the “objects” in the problem, and we use the procedure to determine the permeability/porosity values we should assign to each color in order to match the data. Fig. 10 shows in (a) the true reservoir and in (b) the calculated reservoir. In this case we simulated the field considering 10 wells. The data used in the inverse problem consisted of field observations from one DST and permanent gauges at each well over a 200 day period. The simulation grid consisted of a 30 × 30 mesh. Variance of Parameter Estimates and Value of the Information The previous section showed some examples of solutions to reservoir inverse problems by the integrated use of data of different nature. The inverse problem is not completely solved by providing sets of permeability and porosity values that result in a good match of the data. There are other issues that come along with the inverse problem such as the uncertainty associated with each parameter estimate and the value of the information. These issues can be explored with the theory used to develop Eqns. 24 to 29. A simple example will illustrate the use of this analysis. For the sake of simplicity let us assume that we have a homogeneous (but unknown) reservoir, and a data set similar to the one illustrated in Fig. 1. The reservoir was discretized with 40 × 30 mesh and parameterized with 100 large pixel r parameters. After a permeability distribution ( α * ) that matches the data is obtained, the first question we would like to answer what is the uncertainty associated to each of the 100 parameters we have calculated.

The following remarks and observations can be drawn from Fig. 13: • Combinations of data always reduce the uncertainty. That is, adding more uncertain information to a data set does not downgrade the solution, on the contrary, it always produces a reduction in the level of uncertainty. • The addition of 4-D seismic information produced a remarkable improvement despite this information having a relative high uncertainty. This is because the 4-D seismic data provides information that is spatially distributed. • The water cut information seems to help little in this specific problem. • The long term pressure history seems to provide strong information. The highest sensitivity in the long term pressure is concentrated at the time when the water front reaches the wells28. This is confirmed in the color map, here the best area determined by the long term pressure is the area behind the water front. • The DST information alone will not help much in the 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

•

•

distributed parameters. DST’s resolve simple models only. On the other hand the combination of the apparently poor information from DST with the long term pressure data (permanent gauges) produced a remarkable improvement in the quality of the data. The DST plot reveals how this data should be used in reservoir inverse problems. If no other data are available then the number of parameters should be small enough to ensure a low variance. Later in the life of the reservoir, when other data are available, such as production history and 4-D seismic, then the DST information can be used again but without the need to constrain the reservoir to a small number of parameters. The appropriate number of parameters, or level of parameterization (size of the model) can be adjusted with the amount of data, that is when data are scarce the model should be kept small. When the data are abundant then certain areas of the reservoir can be parameterized with a more complex model.

k point of view, they are seem to have σ essentially the same uncertainty. Both maps reveal that the permeability at the middle of the South and West boundaries will be ill-determined, consequently any value there cannot taken seriously, since any arbitrary but reasonable value assigned to those regions would provide an acceptable match of the data. That is the same as saying that the available data are not sufficient to resolve the parameters in these regions of the reservoir.

evaluated from the Both variance maps are similar from the qualitative point of view, this implies that it is not necessary to know the true reservoir values to perform the variance analysis, a reservoir description that provides a good match of the field observations is sufficient for the analysis. Application of Variance Analysis in Reservoir Monitoring The variance analysis shown in the previous sections can be used to organize and optimize the data collection policy in a reservoir. In general it is possible to obtain a first approximation to the reservoir model before production and/or dynamic data are collected in the field. Then it is possible to use this approximated reservoir to perform the variance analysis28 and to prepare plots similar to those shown Figs. 13 and 14 before any data is actually collected in the field. Then the best strategy for data collection can be devised considering that each piece of information will be used later in conjunction with the others. The analysis can also be used to find the best level of parameterization at a given time considering the data available. The level of parameterization will change as more field observations are made available. Both processes, reservoir parameter estimation and future data collection policy, have to be updated continually since every time new field data arrives the reservoir model is updated via the inverse problem solution, then the variance analysis is updated and the cycle starts over again. Since the variance analysis uses a covariance model for the data (field observations) which is generally assumed to be diagonal and which is related to the accuracy or reliability of the field measurements, then this method can be used to investigate how the accuracy of the instruments used to collect the information has an effect on the calculated permeability and porosity. Conclusions We showed procedures; (1) to estimate the distribution of permeability and porosity in the reservoir by using dynamic data; (2) to assess the uncertainty associated with the calculated permeability and porosity and; (3) to determine the value of each data type from the point of view of the inverse problem. The procedures allow us to integrate information from several sources to compute distributions of permeability and

From these analyses it was observed that the pressure data play an important role in the reservoir parameter problem. Therefore a more detailed analysis was performed in the case of the pressure data. The analysis was performed by breaking down the pressure into major components, which are: DST, shut-in, and data from permanent gauges (referred to here as long term pressure only). The results are shown in Fig. 14. The focus here is on the shut-in pressure information. The apparently poor shut-in pressure information enhances the quality of the data set. The implication is important. If permanent gauges are installed, then any sporadic or unpredicted shut-off of a well will produce a short transient, and indirectly will enhance the quality of the rest of the data. Analysis of the Results from a Reservoir Inverse Problem The previous sections presented a systematic procedure to analyze the issue of how each data type contributes to the reduction of the uncertainty attached to each parameters. To perform such analysis we assumed that we knew the true values of the parameters. Here we will use the variance approach to analyze the results of an inverse problem, that is after the data have been matched to a calculated set of parameters. Then a new question arises, namely how good are such calculated parameters. This issue was investigated by using the variance analysis. This case used the same heterogeneous reservoir and data shown in Fig. 1. The first guess for the inverse problem was a homogeneous distribution of permeability. The result of the match and the variance calculation are shown along with the true case in Fig. 12. Both true and calculated sets of parameters provide a good match to the data, the “true case” provides the exact match because the data was generated with this distribution of permeability. The true and the calculated permeability fields seen to be different, but when both are

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

9

porosity. The information that can be integrated includes: well test data, shut-in pressure, long term pressure (from permanent gauges), production history (water cut, rates), interpreted 4-D seismic maps, permeability-porosity correlations, variogram models and large scale geological information (object modeling). This seems to cover most of the information that can be realistically obtained from a program for reservoir monitoring with current technology, but it is not necessary to have all of them to apply the method described in this paper. In this work the interpreted 4-D seismic information was assumed to be in the form of maps of change of saturation in the reservoir. The procedure can be adapted to the case when the 4-D seismic interpretation is not related to a simple change of saturation but to a more complex relationship that could be a function of not only saturation but pressure, density, rock type, etc. The method developed in this work relies on the computation of the sensitivity coefficients for the field observations with respect to the parameters of the inverse problem. This computation might be considered an unnecessary burden since there are other methods to solve the inverse problem without the sensitivity coefficients, but we found such an effort worthwhile for the following reasons: • The sensitivity coefficients are necessary to perform the variance analysis, and this is key to understanding the meaning of what it is being calculated and to plan reservoir monitoring. The sensitivity coefficients are also needed to find the most adequate parameterization of the reservoir for a given amount of information. • The method to compute the sensitivity coefficients, as developed in this work, is relatively simple to implement when it is possible to have access to the numerical reservoir simulator code. Thus the complexity of modeling the reservoir is left to the simulation part and not to the inverse problem as it would be the case for the optimal control approach. • The Gauss-Newton approach is very efficient as a parameter estimation algorithm, compared to direct search methods that do not require sensitivity coefficients. • The computation of the sensitivity can be performed with high efficiency, specially in the case of object or large pixel modeling. The efficiency can also be enhanced by parallel computing. From the results of our numerical experiments it is possible to conclude that the method can be utilized as a tool for reservoir characterization. The method seems to be more useful when the dynamic data set consists of a large number of measurements that are difficult to honor with the existing geostatistical approaches; this may be the case in mature fields that have already gone through a secondary recovery process and in which data have been gathered over many years.

The more interesting conclusions obtained from this work are about the roles of well testing, 4-D seismic data and integration. Well testing has often been considered as an isolated tool to solve relatively simple reservoir models in the neighborhood of the wells and to improve the productivity. These are very important considerations from the engineering and economical point of view, but since the well tests are performed generally during the earlier part of the life of the reservoir then this information is not used later mainly because of the assumption that the area investigated is small and also because of the lack of a mathematical tool to be used in a more complex environment. The analysis performed here shows that the well test information can be used to resolve much more complex reservoirs models later in the life of the reservoir, not as a standalone approach but as a piece of information in a larger data set. The addition of well test information enhances the resolution value of the other information, thus well tests early in the life of the reservoir will also be useful later for reservoir characterization. The value of 4-D seismic data is due to the fact that it provides information that is spatially distributed in the reservoir in contrast to the other data that are localized at the wells. Even so, it was found that the 4-D information does not assist much in reservoir characterization when it is considered alone. The value of this type of information appears when it is combined with other more traditional data such as production history. Also it was found that this kind of information does not need to be very accurate since data in coarse “black and white” format can still be useful. In all the examples shown in this work it was assumed that two consecutive 3-D seismic surveys were performed after production had begun in the reservoir. If the first survey were to be performed before any production then it is very likely that the 4-D information will provide even more value to the reservoir inverse problem. The value of data integration can be evaluated with the variance analysis. It was shown that the addition of data always enhances the overall resolution of the parameters. The addition of data that seems to provide little information value when considered in isolation results in a large increase in the resolution of the other data. A dramatic example is in the case of the shut-in pressure data, which includes only a very small number of pressure measurements obtained when the wells are closed at the surface for a short period. This information when considered by itself cannot be used to solve even the simplest models, but its value is considerable when it is used in an integrated procedure such as the one developed in this work. Acknowledgments The authors gratefully acknowledge the financial support from 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 r α = parameter model vector r α * = optimal parameter model vector r α prior = a priori parameter model vector k= k0 = λ= φ= σ2 = ∇E = Cd= Cα = Cov{ } = DST = r d = E= r f = G= H= J= R= Sw = S= t= U= V= W = r y =

r Z =

1. Seymour, R.H and Barr, F.J.: “Seabed-Seismic 4D-Collection Method for Reservoir Monitoring,” JPT (January 1997) Vol. 49, No. 1, 40-41. 2. Jacquard, P., and Jain, C.: “Permeability Distribution from Field Pressure Data,” Soc. Pet. Eng. J. (Dec. 1965) 281; Trans., AIME, Vol. 234. 3. Carter, R.D., Kemp, L.F., Pierce, A.C., and Williams, D.L: “Performance Matching With Constraints,” Soc. Pet. Eng. J. (Apr. 1974) 187; Trans., AIME, Vol. 257. 4. Dogru, A.H., and Seinfeld, J.H.: “Comparison of Sensitivity Coefficient Calculation Methods in Automatic History Matching,” Soc. Pet. Eng. J. (Oct. 1981) 551. 5. Carter, R.D., Kemp, L.F., and Pierce, A.C.: “Discussion of Comparison of Sensitivity Coefficient Calculation Methods in Automatic History Matching,” Soc. Pet. Eng. J. (Apr. 1982) 205. 6. Chen, W.H., Gavalas, G.R., Seinfeld, J.H., and Wasserman, M.L: “A New Algorithm for Automatic History Matching,” Soc. Pet. Eng. J. (Dec. 1974) 593; Trans., AIME, Vol. 257. 7. Chavent, G., Dupuy, M., and Lemonnier, P.: “History Matching by Use of Optimal Theory,” paper SPE 4627 presented at the 1973 SPE Annual Technical Conference and Exhibition, Las Vegas, NV, Sep. 30 - Oct. 3. 8. Watson, A.T., Seinfeld, J.H., Gavalas, G.R., and Woo, P.T.: “History Matching in Two-Phase Petroleum Reservoirs,” paper SPE 8250 presented at the 1979 SPE Annual Technical Conference and Exhibition, Las Vegas, NV, Sep. 23-26. 9. Yang, P.H, and Watson, A.T.: “Automatic History Matching With Variable-Metric Methods,” paper SPE 16977 presented at the 1987 SPE Annual Technical Conference and Exhibition, Dallas, TX, Sep. 27-30. 10. Kamal, M.M.: “Use of Pressure Transients to Describe Reservoir Heterogeneity,” JPT, (Aug. 1979) 1061, Trans., AIME, Vol. 257, SPE Reprint Series No. 14. 11. Tang, Y.N., and Chen, Y.M.: “Application of GPST Algorithm to History Matching of Single-Phase Simulator Models,” paper SPE 13410, 1985. 12. Tang, Y.N., and Chen, Y.M.: “Generalized Pulse-Spectrum Technique for 2-D and 2-Phase History Matching,” Applied Numerical Mathematics (1989) Vol. 5, 529-539. 13. Oliver, D.S.: “Incorporation of Transient Pressure Data Into Reservoir Characterization,” In Situ (1994) Vol. 18, 243-275. 14. Chu, L., Reynolds, A.C., and Oliver, D.S.: “ Computation of Sensitivity Coefficients for Conditioning The Permeability Field to Well-Test Pressure Data,” In Situ (1995) Vol. 19, 179-223. 15. Oliver, D.S.: “Multiple Realizations of the Permeability Field From Well Test Data,” paper SPE 27970 presented at the 1994 University of Tulsa Centennial Petroleum Engineering Symposium, Tulsa, OK, Aug. 29-31. 16. Chu, L., Reynolds, A.C., and Oliver, D.S.: “Reservoir Description From Static and Well-Test Data Using Efficient Gradient Methods,” paper SPE 29999 presented at the 1995 SPE International Meeting on Petroleum Engineering, Beijing, PR China, Nov. 1995. 17 Reynolds, A.C., He, N., Chu, L., and Oliver, D.S.: “Reparameterization Techniques for Generating Reservoir 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

permeability permeability from realization kriging weight porosity variance gradient of E data covariance matrix a priori parameter covariance matrix covariance matrix operator drill stem test data vector objective function material balance equations vector matrix of sensitivity coefficients Gauss-Newton approximation to the Hessian matrix of objective function Jacobian of material balance equations resolution matrix water saturation singular value matrix time factor from singular value decomposition factor from singular value decomposition weighting matrix vector of pressure and saturation in the simulation mesh sensitivity vector

Subscripts obs = observed value calc = calculated value npar = number of parameters nblocks = number of blocks in simulation grid p = number of nonzero singular values T = transpose of matrix

Superscripts = vector -g = generalized inverse of a matrix k = iteration index p = pressure wc = water cut ∆Sw= change of water saturation

r

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

11

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

Well-Test Data for Mapping Permeability and Porosity,” In Situ (1996) Vol. 20, 41-59. Oliver, D.S., He, N. and, Reynolds, A.C.: Conditioning Permeability Fields to Pressure Data,” paper presented at the 5th European Conference on the Mathematics of Oil Recovery, Leoben, Austria, (1996), Sep. 3-6. He, N., Reynolds, A.C., and Oliver, D.S.: “Three-Dimensional Reservoir Description from Multiwell Pressure Data,” paper SPE 36509 presented at the 1996 SPE Annual Technical Conference and Exhibition, Denver, CO, Oct. 6-9. Anterion, F., Eymard, R., and Karcher, B.: “Use of Parameter Gradients for Reservoir History Matching,” paper SPE 18433 presented at the 1989 SPE Symposium on Reservoir Simulation, Houston, TX, Feb. 6-8. Bissell, R., Sharma, Y., and Killough, J.E.: “History Matching Using the Method of Gradients: Two Case Studies,” paper SPE 28590 presented at the 1994 SPE Annual Technical Conference and Exhibition, New Orleans, LA, Sep. 25-28. Tan, T.B., and Kalogerakis, N.: “A Fully Implicit, ThreeDimensional, Three-Phase Simulator with Automatic HistoryMatching Capability,” paper SPE 21205 presented at the 1991 SPE Symposium on Reservoir Simulation, Anaheim, CA, Feb. 17-20. 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. 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. 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. 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. Landa, J.L.: “Reservoir Parameter Estimation Constrained to Pressure Transients, Performance History and Distributed Saturation Data,” Ph.D. dissertation, Stanford University, Stanford, CA, 1997. 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 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. 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 Tarantola, A.: Inverse Problem Theory - Methods for Data Fitting and Model Parameter Estimation, Elsevier Science Publishers, Amsterdam 1987. Menke, W.: Geophysical Data Analysis: Discrete Inverse

Theory, Academic Press, Inc., San Diego, CA, 1989. 34. Gill, P.E., Murray, W., and Wright, M.H.: Practical Optimization, Academic Press, San Diego, CA, 1981. 35. Bard, Y.: Nonlinear Parameter Estimation, Academic Press, New York, NY, 1970. 36. Jackson, D.: “Interpretation of Inaccurate, Insufficient and Inconsistent Data,” Geophysical Journal of the Royal Astronomical Society,(1972), Vol. 28, 97-109. 37. Fasanino, G., Molinard, J., and de Marsily, G.: “Inverse Modeling in Gas Reservoirs,” paper SPE 15592 presented at the 1986 SPE Annual Technical Conference and Exhibition, New Orleans, LA, Oct. 5-8. 38. Wen, X., Gomez, J., Capilla, J. and Sahuquillo, A: “Significance of Conditioning to Piezometric Head Data for Predictions of Mass Transport in Groundwater Modeling,” Mathematical Geology, (1996), Vol. 28, No. 7, 951-968. 39. Sheriff, R.E, and Geldart, L.P.: Exploration Seismology, Cambridge University Press, New York, NY, 1995.

12

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

8000

c

?

(a) (b)

4000

Permeability

c

c c

6000

2000 0

log(k)

φ - Porosity (h)

c c

c c

DST Well # 1

Pressure Water Cut

Production History Well # 1

c c

(f)

c

(i) (m)

c c

(c)

c c

c

DST Well # 2 (j) Production History Well # 2

c c

(g)

c c

(n)

c c

(d)

c c

DST Well # 3

Production History Well # 3

(k)

(o)

0.4

Saturation

c c

(e)

c c

0.6

DST Well # 4 Injector

Injection History Well # 4 - Injector

0.2

(l)

(p)

Fig. 1: Problem statement. (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

Iter #1 E= 197,097

Iter #2 E= 113,596

Iter #3 E= 41,512

a

True Reservoir Calculated Reservoir

15000

a a

Iter #7 E= 15,617

a a

Iter #11 E= 14,506

a a

a a

Iter #16 E= 14,340

a a

10000

Permeability

a a

a a

Field

a a

a a a a a

5000

Permeability

a

a

a

a

a

a a

a a

Iter #19 E= 14,360

a a

Iter #22 E= 13,680

a a

a a

Iter #25 E= 13,510

a a

Change of

∆Sw

Match of the Data

a a a

a a

Saturation #1

a a

0.30

0.25

a a

Iter #28 E= 13,252

a a

a a

Iter #31 E= 12,756

a a

a a

Iter #34 E= 10,530

a a

0.20

0.15

Change of

a

Saturation #2

a

a a aaaa

a aa aaaaa aaaaaa

DST Well # 1

a a

True Pressure True Water Cut Calculated Values Pressure & Water-Cut History Well # 1

a a a a

a a

Iter #36 E= 2,778

a a

a a

Iter #38 E= 71

a a

a a

Iter #41 E= 0.01

a a

15000

DST Well # 2

a

a

a

a

a

a

5000

0

a a a a aa

a aa aaaa aaaaaaa

DST Well # 3

a a a a a a a aa a a a a a a a a a a a a aaaa 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

DST Well # 4 Injector

aaa a a a a a a a a a a a a a a a a a a a

a aaa aaa Pressure & Water-Cut History a a a a a a a a a a a a Well # 3 a a a a

a

**Fig. 4 Channel model. Evolution of parameters.
**

True Permeability

aa a

a

a

a

a aaaa

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 a a a a a a

Pressure History Well # 4 - Injector

a

a

a a aa

a a aaa

a a

a a aa a a a a

Fig. 2 Channel model - Data and match of the data.

a

(a) (c)

a

Calculated Permeability

15000

**Curve Analysis - Parametric Channel Model
**

α5

α2 = kmatrix

a

a

10000

5000

α3 π

-1 α4

a

(b)

a

0

α1 = kch

Calculated Permeability

anne

l

a

a

α8

α7

a

a

(c)

α6

Fig. 3 Parameterization of a channel.

Fig. 5 Channel reservoir. (a): True reservoir. Calculated reservoir using large pixel model. Comparison of pixel and true reservoir.

md

(b): (c):

md

aaaaaaaaaa

aa

aaaaa

aa a

a a aa aaa a aaa a a a a a a aaaa 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 a aa a aa a a aaa aaa a aaa a a a a a a a a a a a a a Pressure & Water-Cut History a a a a a a a a a a a a a a a Well # a 2 a a a a a a a a a a a a a a a a a a a a a aaaa 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 aa a a

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

a

a

a

a

a

a

a

10000

14

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

Iter #1

Iter #2

Iter #3

c

c

c

c

c

c

True ∆Sw #1

Filtered ∆Sw #1

0.4

0.3

c

c

c

c

c

c

0.2

∆Sw

Iter #7

Iter #11

Iter #15

0.1

c

c

c

c

c

c

0.0

(a) True ∆Sw #2

(b) Filtered ∆Sw #2

c

0.4

c

c

c

c

c

Iter #20

0.3

Iter #25

Iter #30

c

0.2

c

c

c

c

c

∆Sw

0.1

c

c

c

c

c

c

0.0

(c)

(d)

c

Iter #35

Iter #42

Iter #45

Fig. 6 4-D seismic data in “black and white” format.

c

c

c

c

c

c

c

c

c

c

c

True Permeability

c

Linear Search

Iter #48

Iter #51

c

c

c

c

c

0.6

a

a

c c c c c c

0.4

0.2

Fig. 8 Fault model. Evolution of parameters.

a

(a)

a

True Permeability - α=30°

6000

**Calculated Permeability - α=30°
**

a a

Calculated Permeability - Exact ∆Sw

∆Sw

a

a

a

5000

a

103

4000

3000

md

10

a a a a

2000

a

(b)

a

1000

(a)

0

(b)

**Calculated Permeability - α=120°
**

a a

**Calculated Permeability - α=75°
**

a a

Calculated Permeability - Filtered ∆Sw

a

a

a

a

a

a

a

(c)

a

(c)

(d)

Fig. 7 Comparison of results. (a): True reservoir. (b): Calculated reservoir using exact data. (c): Calculated reservoir using 4-D seismic data in “black and white” format.

Fig. 9 Kriging model. (a): True reservoir. (b): Calculated reservoir using exact data. (c): Calculated reservoir variogram rotated 90°. (d): Calculated reservoir variogram rotated 45°.

md

SPE 38653

INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION

15

True Permeability

c a c c c c c c a

True k

Calculated k

104

c

103

c

c

c

10000

k - md

c

102

5000

c

(a)

c

c

(c)

c

10

k/σk - True k

k/σk - Calculated k

10

1

c

c

c

c

1

(a)

Calculated Permeability

c a c c c c c c a c

c

(b)

c

c

(d)

c

10-1

Fig. 12 Variance maps: (a): True reservoir . (b): k/σ map for true reservoir. (c): Calculated reservoir. (d): k/σ map for calculated reservoir

(b)

Fig. 10 3-D seismic model. (a): True reservoir. (b): Calculated reservoir.

Map of σ-1

1

c

c

10-1

10-2

c

c

c

Fig. 11 Map of σ . Homogeneous reservoir.

-1

well location

1/σ md-1

k/σk md/md

k md

16

JORGE L. LANDA AND ROLAND N. HORNE

SPE 38653

Data: DST Pressure

Data: Long Term Pressure

c

c

c

c

c

c

c

c

Data: Pressure

k/σ scale 10

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

Data: Water Cut

Data: Water Cut + ∆Sw

Data: ∆Sw

c

c

c

c

c

c

c

c

c

c

Fig. 13 Data analysis.

Data: DST Pressure

k/σ scale 10 1 10-1

c

c

10-2 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

Fig. 14 Analysis of the pressure information.

k/σk md/md

c c

c

c

k/σk md/md

c c

c

c