Вы находитесь на странице: 1из 7

SPE 108180 A More General Capillary Pressure Curve and Its Estimation From Production Data

Tao Gang, SPE, PetroTel Inc., and Mohan Kelkar, SPE, U. of Tulsa

Copyright 2007, Society of Petroleum Engineers This paper was prepared for presentation at the 2007 SPE Rocky Mountain Oil & Gas Technology Symposium held in Denver, Colorado, U.S.A., 1618 April 2007. 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, Texas 75083-3836 U.S.A., fax 01-972-952-9435.

Abstract Capillary pressure plays a very important role in naturally fractured reservoirs. The exchange of fluids between matrix and fracture is essentially dependent on capillary pressure expressed by the capillary pressure curve. Capillary force may contribute to the displacement process through imbibition process, or may oppose it in the drainage displacement process. It is essential to represent the capillary pressure curves properly. Many capillary pressure correlations have been suggested in the literature. Capillary pressure data were normally measured using core data. However these methods suffer limitation due to the scale over which the data were collected. Therefore it will be very useful if the capillary pressure curve could be extracted from production data through the history matching process. This paper presents a simple and generalized capillary pressure model which captures both the spontaneous and forced parts of the imbibition capillary pressure curve. In addition, the capillary pressure was estimated by the production data. The sensitivities of production data with respect to capillary pressure model parameters are required when the gradient-based optimization algorithm is used to minimize the objective function. How to calculate the sensitivities of the production data with respect to the capillary pressure model parameters is provided. By minimizing the objective function which describes the mismatch of the observed and simulated production data, the water-oil capillary pressure curves are estimated. All of implementations are incorporated into a commercial simulator (ECLIPSE) and iterated in the automatic history matching scheme. The method is validated using many synthetic cases. Introduction Naturally fractured reservoirs represent an important percentage of the worldwide hydrocarbon reserves and production. The major characteristic of naturally fractured

reservoirs is that the matrix blocks, which have high porosity and low permeability, were separated by interconnected fractures, which have high permeability and low porosity. In other words, the fractures are interconnected as a continuum and form a flow network connected to the wellbore. The matrix blocks, on the other hand, are presumed not to be interconnected; therefore, there is no matrix-to-matrix flow, but there is matrix-to-fracture flow which is mainly controlled by the gravity and capillary pressure forces though other forces such as expansion, diffusion and viscous may also influence the recovery process depending upon the pressure, temperature and the composition of the fluids involved in the process. Therefore, the capillary pressure curve plays a much more important role in a naturally fractured reservoir than in a conventional reservoir.The dynamic role of the capillary forces in a conventional reservoir is more limited.[1] In naturally fractured reservoirs capillary forces may contribute to the displacement process through imbibition process, or may oppose it in the drainage displacement process. So a proper capillary pressure curve is required to reasonably model the interaction between the matrix and the fractures. There are four different methods of measuring capillary pressure curves: porous plate, centrifuge, air mercury injection test and water vapor desorption.[2] The most common data is the air-brine (porous plate) test and the air mercury injection test. However, these methods are based on the availability of the core data and suffer limitation due to scale over which the data are collected. Many capillary pressure correlations have been suggested in the literature. The widely used model in petroleum field is the Brooks-Corey capillary pressure model [3]. Brooks and Corey conducted analysis for the capillary pressure curves of a large number of consolidated core samples. They found that the relationship between the capillary pressure and the normalized wetting phase saturation was a straight line on loglog data. The mathematical expression of this relationship was also known as the Brooks-Corey capillary pressure model. Residual wetting-phase saturation must be known to calculate the normalized wetting-phase saturation. The Brooks-Corey capillary pressure model works satisfactorily in many cases. Pooladi-Darvish and Firoozabadi[4] presented another capillary pressure model using the capillary pressure constant and the normalized wetting phase saturation. Li [5] derived a more general capillary pressure model from fractal modeling of a porous medium. He suggested that the more general capillary pressure model could be reduced to the frequently used Brooks-Corey capillary pressure model when

www.petroman.ir

SPE 108180

the fractal dimension of a porous medium takes a limiting value. This paper presents a new capillary pressure model which captures both the spontaneous and forced parts of the imbibition capillary pressure curves. Furthermore, the capillary pressure curve can be obtained through estimating the capillary pressure model parameters using the production data. The procedure to estimate the capillary pressure using the production data is an inversion problem or a history matching problem. In other words, we will adjust the model parameters in order to minimize the objective function, which defines the mismatch of the calculated data and observed data. The model parameters in this paper are the parameters controlling the capillary pressure curve. When a gradientbased optimization algorithm is used to minimize the objective function, for example, the Levenberg-Marquardt [6] method in this paper, the sensitivities of the production data with respect to the model parameters are required. This paper presents how to calculate the sensitivity coefficients when the adjoint method is used. All calculations are incorporated within the framework of a commercial simulator ECLIPSE.[7] The proposed capillary pressure model is compared with other commonly used models, and the comparison indicates this model is reasonable in properly capturing the imbibition capillary pressure curve. The method is validated by conducting history match for synthetic data sets to estimate the capillary pressure curve. The results indicate that the capillary pressure curve can be properly estimated using the production data. Analytical Capillary Pressure Model The mathematical expression for the Brooks-Corey capillary pressure model is:[3]
Pcow = Pc (S wD ) , (1) where is the pore size distribution index, and Pc is the entry capillary pressure. S wD is the normalized saturation of the wetting phase and is expressed as S w S iw , (2) S wD = 1.0 S iw S or
1

Pe b = 1 (5) , P max where = 3 D f , D f is the fractal dimension, which is a

representation of the heterogeneity of rock. The greater the fractal dimension, the greater the heterogeneity. Note that the pore size distribution index in the Brooks-Corey capillary pressure model is also a representation of the heterogeneity. For D f < 3 , if Pmax approaches infinity, then Equation 4 can be reduced to Equation 1, which is the frequently used Brooks-Corey model. Li also pointed out that Equation 4 can be reduced to the imbibition capillary pressure model proposed by Li and Horne in 2001.[8] In this paper, a new capillary pressure curve was presented as follows: S (S wD Sc )C1 Pc + C2 , Pcow = ln c (6) C1 S + (S S )C1 Sc + Sc c wD c ln S S C1 c c where Pc is a capillary pressure constant, S c is the cross water saturation, C1 is a positive odd integer and C2 is a real number. When C1 = 3 and C 2 = 0.0 , Equation 6 simplifies to
S (S wD S c )3 . ln c (7) 3 S + (S S )3 Sc + Sc c wD c ln S S3 c c It is known that ln( x) requires that x is positive. So S c has the range constraint when Equation 6 or 7 is used to estimate the capillary pressure curve. From Equation 6, by Pcow = Pc

assuming C1 = 3 , we got, S c + (S wD S c )3 0 for 0 < S c < 1 and 0 < S wD < 1 . So S c has to satisfy the following Equation

S c (S wD S c )3 > 0.0 , which means S c > 0.32 . So the

range for the cross point is 0.32 < S c < 1.0 when C1 = 3 . For more details please see Gang.[9]
Comparison of Analytical Capillary Pressure Curves Figures 1 to 4 plot the capillary pressure curves with different Pc , S c , C1 and C2 respectively when the capillary pressure curves are determined using Equation 6. Figure 1 illustrates the capillary pressure curve for different Pc and same S c , C1 and C2 . Figure 2 illustrates the capillary pressure curve for different S c and same Pc , C1 and C 2 . Figure 3 illustrates the capillary pressure curve for different C 2 and same S c , Pc and C1 . Figure 4 illustrates the capillary pressure curve for different C1 and same S c , Pc and C 2 . Figures 1 to 4 imply that Pc controls the magnitude of the capillary pressure. S c is the cross point where the capillary pressure is zero. C1 impacts the curvature of the capillary pressure curve. Figure 5 plots the capillary pressure curves using the Brooks-Corey, Pooladi-Darvish and Firoozabadi and the new model. Table 1 shows the model parameters used in the curves

where S iw is the irreducible water saturation and S or is the residual oil saturation. The Brooks-Corey capillary pressure model works satisfactorily in many cases and has been utilized widely for several decades in petroleum and other industries. Pooladi-Darvish and Firoozabadi [4] presented another equation for capillary pressure as Pcow = Pc ln S wD , (3) where Pc is the capillary pressure constant. S wD is the normalized saturation of the wetting phase defined in Equation 2. Li [5] suggested the following capillary pressure model

Pcow = Pmax (1 bS wD ) , (4) where Pmax is the capillary pressure at the residual nonwetting phase saturation in the imbibition case and the capillary pressure at the residual wetting phase saturation in the drainage case. b is a constant and expressed as

www.petroman.ir

SPE 108180
n f of ,1 m1 f n of ,1 = m 2 # f n of ,1 mN m n f wff ,Nw m1 , n f wff ,Nw " m2 " # n f wff ,Nw " mN m

at Figure 5. It is obvious that the new model can capture the spontaneous and forced parts of the imbibition capillary pressure curve.
Objective Function and Optimization Algorithm We let m denotes the N m dimensional column vector of model parameters. d obs is the measured data and g (m ) represents the corresponding vector of predicted data with dimension N d . C d denotes the covariance matrix for production data measurement error. The objective function describing the mismatch between the predicted and observed production data can be defined as[10]
O(m ) =
1 2 1 (g (m) d obs )T C D (g (m ) d obs ) .

n f wf ,1

n f gf ,1

n f oma ,1

m f

( )

n T

m1 n f wf ,1 m2 # n f wf ,1 mN m

m1 n f gf ,1 m2 # n f gf ,1 mN m

" m2 m2 # " # n n f gma f oma ,1 ,N " mN m mN m

m1 n f oma ,1

"

n f gma ,N

n f wff ,1

m1 n f gma ,N

m1 n f wff ,1 m2 # n f wff ,1 mN m

"

(14)

and
P (15) m = c . Sc Actually, in the flow equations only the mass transfer from the matrix into fracture involves the matrix capillary pressure which is estimated in this paper. Here, we present formula to compute the derivatives of the capillary pressure with respect to Pc and S c . When Equation 7 is used to determine the capillary pressure curve, the partial derivative of the Pcow with respect

(8)

Normalized objective function is defined as ON (m ) = 2O(m ) N d . (9) The value of the normalized objective function can be used as useful convergence criteria in the optimization algorithm. In this paper, a gradient based optimization method, LevenbergMarquardt method, is used to minimize the objective function defined in Equation 8, the gradient and Hessian of the objective function is T 1 O(mk ) = Gk C D (g (m ) d obs ) , (10) where O(mk ) and H (mk ) is the gradient and Hessian of the objective function evaluated at mk . Gi , j = g i m j is the sensitivity coefficient matrix, which represents the relation between the production data d with respect to the model parameter m . When the Levenberg-Marquardt method is used, the mk +1 at each optimization iteration is given by (H k + I )mk +1 = O(mk ) . (12) where is a damping factor controlling the convergence speed. The Levenberg-Marquardt step is computed as a weighted combination of the Gauss-Newton step and the steepest descent step at each iteration.
Sensitivity of Production Data to Capillary Pressure Parameters In this paper, the production data will be used to estimate only S c and Pc by assuming C1 = 3 and C 2 = 0.0 for simplicity. When the capillary pressure model parameters can be estimated, the corresponding capillary curve can be obtained using Equation 6 or 7. Next we will discuss how to calculate the sensitivity of the production data to capillary pressure model parameters, Pc and S c . When the adjoint equation is used to calculate the adjoint variables . The formulation to calculate sensitivity of J to capillary pressure parameters is
f n m J = m + m i =1
T 1 H (m k ) G k C D Gk ,

to Pc and S c is expressed as
S (S wD Sc )3 ln c S + (S S )3 Pcow c wD c = , 3 Pc Sc + S c ln S S3 c c
Pcow S c S + (S S )3 wD c ln c S (S S )3 Pc wD c c = 3 S c Sc + Sc ln S S3 c c
3 Sc + Sc 1 ln 3 Sc Sc S c + (S wD S c ) + Pc ln S (S S )3 S c wD c c 3

(16)

(11)

(17)

The flow equations f n can be written as


n n n fnf ,i , j ,k = Fn,i , j ,k C ,i , j ,k A ,i , j ,k q ,i , j ,k ,
n n fnma ,i , j ,k = C ,i , j ,k Ama ,i , j ,k ,

(18) (19)

where, = o, w, g ,

Fn,i , j ,k
n

represents flow term between

adjacent gridblocks, C ,i , j ,k is the mass transfer between


n matrix and fracture, A ,i , j ,k is the accumulation term and

n n q ,i , j ,k is the sink or source term. Only C ,i , j , k involves the

capillary pressure of matrix blocks. Using the chain rule, we can get Ci , j , k d (Pcow ) f wf ,i , j , k d (Pcow ) fi, j ,k . (20) = = (Pcow ) d (m ) (Pcow ) d (m ) (m ) When the relation between the coupling term and the capillary pressure Pcow is known, we can evaluate f i , j , k (m ) by

) (

the matrix m

( ) ( ) ( f ) is defined by
N T

n ,

(13)

n T

substituting Equations 16, 17 and 20 into Equation 13. Considering m = 0 , we can obtain m J , which is the sensitivity of production data J with respect to model parameter m . For more details please see Gang.[9]

www.petroman.ir

SPE 108180

Results All discussed calculations are implemented in a full-featured implicit black-oil simulator, ECLIPSE. [7] For simplicity, we assume C1 = 3 and C 2 = 0.0 when Equation 6 is used to estimate the water-oil capillary pressure curve. In this section, a simply synthetic 2D case with 5 wells is used to estimate the capillary pressure curve through history matching the production data observed at the wellbore. The reservoir description model consists of a 21 21 1 regular mesh with grid block dimension of 100 100 40 feet. The gridblock permeability of fracture system of this heterogeneous reservoir and the well locations are shown in Figure 6. The matrix system has a constant permeability 20 mD. The fracture and matrix system has a constant porosity 0.1 and 0.2 respectivly. We assume the true capillary pressure curve was generated by assuming Pc = 20 and S c = 0.5 . This reservoir model with heterogeneous horizontal permeability shown in Figure 6 was simulated for 8 years to generate water cut data at the producers using the true capillary pressure curve. The noises were added to the simulated water cut, and the resulting water cut data were used as the observed data. The observed water cut data are used to estimate the capillary pressure curve. We start with Pc = 0.1 and S c = 0.4 , which is referred as the initial model. And the water cut data calculated with the initial model are referred as the initial estimation. When the optimization algorithm reaches the minimum, the model is referred as the estimated model and the water cut data obtained from the simulation results are referred as the estimated production data. Figure 7 plots the convergence behavior of the objective function; it suggests that after 17 iterations, the normalized objective function decreases from about 17,000 to less than 1.0, which indicates the convergence of the objective function. Figures 8 to 11 show the matching results of the water cut at 4 producers. Different colors represent the initial, matched and observed water cut, with blue being the initial, pink being the matched and green being the observed. It is obvious that all the water cut at 4 producers have been matched reasonably. Since the initial model has a small Pc , so the initial water breakthrough is earlier than the actual data as shown in Figures 8 to 11. The initial, estimated and the true water-oil capillary pressure curves are shown in Figure 12. It is obvious that the estimated water-oil capillary pressure curve agrees with the true curve properly. Table 2 shows the initial, estimated and the true model parameters used in the tested example. The sensitivity analysis indicates that production data are more sensitive to S c than Pc . The estimated results imply that we have a better estimation of the S c compared to Pc . The initial, estimated and true capillary pressure parameters are shown in Table 2. Conclusions Capillary pressure plays an important role in naturally fractured reservoirs compared to the single porosity model. Several experimental approaches are available to estimate the

capillary pressure curve, but all of them are limited to the availability of the core data. In this paper, a more general model has been developed to represent capillary pressure curves. The new capillary pressure model can capture both the spontaneous and forced parts of the imbibition capillary pressure curve. The sensitivity of the production data with respect to the the capillary pressure model has been derived. The capillary pressure model parameters can be estimated by the production data. These estimated parameters can then be used to obtain the capillary pressure curve. The synthetic case result indicates the estimated capillary pressure curve agrees with the true curve reasonably well.
Acknowledgements The authors would like to thank Dr. Dixon for his helpful discussions and the financial support from the TUCRS (Tulsa University Center Reservoir Studies). Nomenclature C = covariance matrix G = sensitivity coefficient matrix H = Hessian matrix m = model parameter O = objective function Reference 1. T.D. van Golf-Racht: Fundamentals of Fractured Reservoir Engineering, Elsevier, 1982 2. Mike Carlson: Practical Reservoir Simulation Penwell, 2003 3. Brooks, R.H. and Corey, A. T.: Properties of Porous Media Affecting Fluid Flow, J. Irrig. Drain. Div. 1966, 6, 61 4. Pooladi-Darvish, M. and Firoozabadi, A.: Cocurrent and Countercurrent Imbibition in Water-Wet Matrix Block, SPEJ, March, 3 2000 5. Li, K.: Generalized Capillary Pressure and Relative Permeability Model Inferred from Fractal Characterization of Porous Media, SPE 89874, presented at the 2004 SPE Annual Technical Conference and Exhibition, Houston, Texas, USA, 26-29 September, 2004 6. Fletcher, R.: Practical Methods of Optimization, John Wiley&Sons, New York, Second Edition, 1987. 7. Schlumberger GeoQuest: ECLIPSE Reference Manual 2004A 8. Li, K. and Horne, R.N.: An Integrated Approach to Saturation Study of Steam-Water Capillary Pressure, SPEREE (December 2001), p. 477-482 9. Gang, T.: Data Integration and Model Improvement for Naturally Fractured Reservoirs, PhD Dissertation, University of Tulsa, 2006 10. Tarantola, A.: Inverse Problem Theory: Methods for Data Fitting and Model Parameters Estimation, Elsevier, Amsterdam, The Netherlands, 1987

www.petroman.ir

Table 1 Model Parameters for Different Capillary Pressure Curves


Model Brooks-Corey Curve Pooladi-Darvish and Firoozabadi Curve New Model Curve 1 New Model Curve 2 Pc 2.5 5 40 40 Lamda 1.5 ---Sc --0.9 0.5 C1 --3 3 C2 -Pcow

180

Pc=200, Sc=0.8, C1=3, C2=-20 Pc=200, Sc=0.8, C1=3, C2=0 Pc=200, Sc=0.8, C1=3, C2=20

130

80

-0 0

30

-20

Table 2: Parameter Estimation Results


Initial Estimated True
40 30 20 10 Pcow 0 -10 -20 -30 Pc=10, Sc=0.5, C1=3, C2=0 Pc=20, Sc=0.5, C1=3, C2=0 Pc=30, Sc=0.5, C1=3, C2=0

-70 0 0.1 0.2 0.3 0.4 0.5 Swd 0.6 0.7 0.8 0.9 1

Pc 0.1 16 20

Sc 0.4 0.502 0.5

Figure 3: Capillary Pressure Curves with Different C2


210 160 110 60 Pcow 10 -40 -90 -140 -190 -240 Pc=200, Sc=0.5, C1=3, C2=20 Pc=200, Sc=0.5, C1=5, C2=20 Pc=200, Sc=0.5, C1=7, C2=20

-40 0 0.1 0.2 0.3 0.4 0.5 Swd 0.6 0.7 0.8 0.9 1

0.1

0.2

0.3

0.4

0.5 Swd

0.6

0.7

0.8

0.9

Figure 1: Capillary Pressure Curves with Different Pc


50 40 30 20 10 Pcow 0 -10 -20 -30 -40 Pc=50, Sc=0.50, C1=3, C2=0 Pc=50, Sc=0.55, C1=3, C2=0 Pc=50, Sc=0.70, C1=3, C2=0

Figure 4: Capillary Pressure Curves with Different C1


40 Brooks-Corey Model (Pc=2.5, Lamda=1.5) 30 20 10 Pcow 0 -10 -20 -30 -40 Pooladi-Darvish and Firoozabadi Model (Pc=5.0) New Model (Pc=40, Sc=0.9, C1=3.0, C2=0) New Model (Pc=40, Sc=0.5, C1=3.0, C2=0)

-50 0 0.1 0.2 0.3 0.4 0.5 Swd 0.6 0.7 0.8 0.9 1

0.1

0.2

0.3

0.4

0.5 Swd

0.6

0.7

0.8

0.9

Figure 2: Capillary Pressure Curves with Different Sc

Figure 5: Comparison of Different Capillary Pressure Models

www.petroman.ir

SPE 108180

20

0
Well 3 Well 2
0.8 Initial Matched Observed

800

Water Cut

15

0.6

0.4

10

Inj 1

Well 4

1600

0.2

365

730

1095

1460

1825

2190

2555

2920

3285

Time (Days)

Well 1
5 10 15 20
0.04

Figure 9: Well 2 Water Cut Matched Result

Figure 6: Gridblock Permeability of Fracture System and Well Locations


Water Cut 100000

Initial 0.03 Matched Observed

0.02

10000 Normalized Objective Function

1000

0.01

100 0 0 10 500 1000 1500 2000 2500 3000 3500 Time (Days)

Figure 10: Well 3 Water Cut Matched Result


0.1 0 2 4 6 8 10 12 14 16 18 Iteration 0.8 1

Figure 7: Normalized Objective Function Convergence Behavior


Water Cut 0.6 1

Initial Matched Observed

0.4

0.8 Initial Matched Observed 0.2

Water Cut

0.6

0 0.4 0 365 730 1095 1460 1825 2190 2555 2920 3285 Time (Days)

0.2

Figure 11: Well 4 Water Cut Matched Result


0 365 730 1095 1460 1825 2190 2555 2920 3285

0 Time (days)

Figure 8: Well 1 Water Cut Matched Result

www.petroman.ir

SPE 108180

25 20 Initial 15 10 Capillary Pressure 5 0 -5 -10 -15 -20 -25 0 0.1 0.2 0.3 0.4 0.5 Swd 0.6 0.7 0.8 0.9 1 Estimated True

Figure 12: Estimated Capillary Pressure Curve

www.petroman.ir

Вам также может понравиться