Professional Documents
Culture Documents
A New Double Porosity Reservoir Model For Oil/Water Flow Problems
A New Double Porosity Reservoir Model For Oil/Water Flow Problems
Summary. This paper presents a model that takes into account the transient nature of the imbibition process and the effect of variation
in fracture saturation. Gravity effect is included in the calculation of the matrix equilibrium water saturation. This simple method re-
quires only one equation per block per component. This is attained by an analytical transfer function that depends only on the fracture
variables and by the assumption of instantaneous pressure equilibrium. The transfer function assumes that capillary pressure is the only
driving force of the process and eliminates the matrix saturation from the fracture flow equations. The assumption of instantaneous
pressure equilibrium eliminates the matrix pressure. The resulting model has the same form as standard single-porosity models. Results
of the new model are compared with those from published laboratory experiments, very fine-grid simulations of matrix/fracture transfer
for a single matrix block, and simulation of field-scale water-injection problems by standard double-porosity models. These results
demonstrate that the new model provides an economical and accurate means of predicting the performance of fractured reservoirs.
Introduction
A naturally fractured reservoir is a complex system composed of The model proposed here is based on two main considerations:
irregular matrix blocks surrounded by a network of highly perme- (1) a relationship for Swma in terms of ~t, Swf, and Pof and (2) the
able fractures. The behavior of such reservoirs has been studied assumption of instantaneous pressure equilibrium in the block. The
intensely for the last 30 years. The idealization of the real system first relationship gives the matrix saturation as a function of the
by a simpler one formed by two superimposed continua of regu- primary variables; because of the latter, pressure-dependent terms
larly shaped matrix blocks and fracture networks was proposed by will be the same in both the fracture and the matrix. Therefore,
Barenblatt et al., 1 Barenblatt and Zheltov, 2 and Warren and the following two equations are added to the system:
Root. 3 This approach is still used in most simulators for fractured
Pof=Poma=Po .................................... (2)
reservoirs.
For multiphase problems, experimental work 4 ,5 provides valu- and Swma =Swma(~t,Swf'Po)' .......................... (3)
able data for the understanding and modeling of the imbibition proc-
First, the matrix saturation function Swma(~t,Swf'Po) is devel-
ess. Numerous authors 6-22 have presented models for the
oped; then, the assumption of instantaneous pressure equilibrium
simulation of multiphase flow,
is justified.
The current trend is to use pseudofunctions to handle
matrix/fracture transfer. The standard procedure is to apply the fluid- Matrix Saturation Function. Looking at the physics of the imbi-
flow equations for each medium and to account for the flow be- bition process and considering only capillary pressure as the driv-
tween matrix and fracture with a type of transfer function. Several ing force, we can apply a material-balance equation and use Darcy's
expressions for this transfer function have been presented in the law to obtain the partial-differential equation for the problem, as
literature. They have in common a shape factor, a term dependent Beckner et al. 18,19 and Bech et al. 21 did. Consider for simplicity
on saturation, and a driving force based on the difference in pres- a ID Cartesian system where a porous matrix medium is surrounded
sure or potential between the matrix and the fracture. Coats 20 in- by water. The general hyperbolic form of the flow equation for
troduced a pseudo-capillary pressure in the driving force, and a two-phase system (see Eq. 2.91 of Ref. 23) can be simplified to
Beckner et at. 18,19 and Bech et al,21 used another approach for
the driving force that is explained later.
In standard models for fractured reservoirs, the matrix/matrix :x (D a:;n)= a:;n, ............................. (4)
flow is ignored and the matrix unknowns are generally eliminated
from the system of equations before it is solved directly or itera- where D= -(kkry./cPiJ-w)(dP)dS w ) . • . . . . . . . . . . . . . . . . . . . . (5)
tively. Such models normally cannot handle the transient nature
of matrix/fracture transfer. In this paper, we explore some ideas has the units of area per unit time. Thus, it can be called capillary
for eliminating the matrix unknowns from the flow equations at diffusivity. The variable S!n is a normalized saturation defined by
the outset and concurrently modeling the transient nature of S!n=Swma -Swc/Swe -Swc' .......................... (6)
matrix/fracture transfer,
where Swe is the equilibrium water saturation or the matrix water
New Model saturation after imbibition is complete. It can vary with Swf as a
result of gravity effects (discussed later).
The standard difference equations for water and oil components Initial and boundary conditions are S!n=O at t=O, O<x<L, and
for each gridblock can be written as S~n=l at the boundaries.
Because Eq. 4 is nonlinear, it can be solved only by numerical
~Tw~cpw = -
v ~t (cPfSwf
--+
cPmaSwma )
+qw ... , ....... (la) methods. A plot or table ofthe average matrix saturation vs. time
~t Bwf Bwma can be obtained from such a solution. In this solution, time refers
to the elapsed time during which the matrix was subjected to the
V (¢fSOf sudden change in saturation at the boundaries. Beckner et al. 18,19
and ~To~cpo=-~t - - + cPmaSoma) +qo' ......... (lb) and Bech et az.2 1 used this approach to calculate the imbibition
~t BOf Boma driving force.
where subscripts f and rna refer to fracture and matrix systems, To account for the partial immersion of the matrix block, we must
respectively, and cP is the potential for the phases. All flow terms make a superposition in time. For example, consider the case where
on the left side are for the fracture system. To solve the system the water level in the fracture reaches the bottom of the matrix block
of equations for the primary variables PDf and Swf' we must ex- at time zero and we need to calculate the imbibition at time tn'
press Poma and Swma in terms of these variables. Of course, the Clearly, the first interval ~z 1 (at the bottom of the block) was ex-
normal constraint equations must also be satisfied. posed to water for time tno However, for an interval above the first
and below the fracture water level, say ~Zi' the exposure time is
Copyright 1992 Society of Petroleum Engineers tn -ti' where ti is the time when the water level in the fracture
SPE Reservoir Engineering, November 1992 419
fits the experiment. When this is not possible, a fine-grid simula-
1.0 r--------=='"""---------..., tion of a single matrix block or even the numerical solution of Eq.
4 can be used to estimate D.
The infinite series (Eq. 9) can be replaced by a finite exponen-
0.8
tial series that gives the exact solution at some chosen points and
c::
agrees well in the rest of the domain. Using a two-term series in
~ Eq. 9 yields
en
,;
f(tD) = lh(e-~lID +e-bID) . ........................ (11)
~ 0.6
~
en
The parameters ~I and ~2 are calculated by requiring Eq. 11 to
-g be exact at two points. Fig. 1 shows the solution for the ID case.
Infinite terms
=i Two terms Here we used ~I =8.0405 and ~2 =22.611. The two points where.
E
0 0.4 the match is exact are shown by arrows. There is good agreement
Z
Q)
between the infinite series and this specific two-term solution, and
~ for tDx > 0.2, the agreement is excellent. In the actual implemen-
Q)
> tation of this imbibition model in the simulator, we use the 2D so-
<
0.2 lution with two terms in each direction. So Eq. 10 becomes
4
Swn = 1- 1,4 E rAil . ............................ (12)
i=1
0.0 ':'-'-'....L.L.L"-~':'-'-.l...L.1...l...L'-':-'-c:'-.L.L.L1..LLJ_'::':.l...L.1...l...Ll..J....l.""':_':_I..J...1..J...J...J...1...L.U
M ~ U ~ ~ 1~
Here, Ai is obtained by combining ~ 10 ~2' and the dimension-
Dimensionless time. to
less time for the two directions:
Fig. 1-1 D analytical solution for average matrix saturation.
Ai=D(:~ + ~~ ),j=I,2; k=I,2 . .................. (13)
x y
reaches the level of the interval ~Zi. Obviously, ~Zi and ti would
have to be stored for each block and each timestep, which is a limi- The partial immersion of the block is handled with a fast-
tation ofthis kind of model. But if D can be considered a constant, convolution integral (see Refs. 25 and 26):
Eq. 4 becomes linear and can be solved analytically (see Eq. 4 of
Ref. 24). For ID problems, the solution for the average water satu- r In+1 dSIIif
Swn (tn + 1)= J Swn(t n + 1 -O)--dO, ............... (14)
ration in the matrix is given by
o dO
Swn=l-f(tDx), ..................................... (7)
which has the solution
where tDx is a dimensionless time defined by 4
tDx =DtILl· ...................................... (8) Swn(tn+ I) = SIIif(t n +1)- 'A E Bi(tn+I), ............. (15)
i=1
Lx is the length of the matrix block and f(tD) is the infinite series
8 e -,..2(2n+ 1)21D
+~SIIif( l-e-AjtJ.I) ....... (16)
00
For 2D problems, the solution is As a result, the normalized Swn for time level n + I is a function
of SIIif and ~t only. Therefore, only Bi (t n ) must be stored at the
Swn = l-f(tDx)f(tDy). . ............................ (10)
previous timestep and kept updated in the current step.
While D can vary considerably with saturation, it may be possi- The equilibrium matrix water saturation, Swe' is adjusted to ac-
ble to find a value for D that yields a reasonable match of the ana- count for the effect of gravity. A relation between Swe and Pc is
lytical solution with the actual imbibition process. The most accurate obtained from the capillary pressure curve. For a given SIIif' there
way to do this probably is to perform a laboratory experiment with is an additional pressure caused by the gravity force that results
actual reservoir rock and fluids and to find the best value of D that in a specific Swe'
Pc
Fig. 2-Matrlx block with partial water Immersion and capillary pressure curve.
700
...::::::::!! 700 ,...............................
.... ......... 600
.......
.. ,
W
.!:!.500
.•...
.... I!!
8
~400 Kleppe & Mo"",
0_ 13.98 ceJmin
.' "8 ...... g. . o. 1.398 CClmln
:::J ......... 0 _ 0.140 ceJmln
~300
Kleppe & Morse 0..
• ... 0 _ 1.398 cm2lmin (5
.. -0. 0 _ 0.350 cm2lmin
o _0.2SO cm2lmin 200
0_ 0.140 cm2lmin
100
0
1000 1500 2000 2500 3000 0 soo 2500 3000
Waler Injected (cm3)
Fig. 3-Comparlson of the proposed model with Kleppe and Fig. 4-Comparlson of the proposed model with Kleppe and
Morse's5 high-rate experiment. Morse's5 low-rate experiment.
Swma = (Swe - Swc)Swn + Swc' ........................ (17) where f(tD) for each of the three directions is defined by Eq. 9.
For a cubic block, PD is more than 90% of its stabilized value for
Instantaneous Pressure Equilibrium. Consider the standard partial- tDx greater than 0.1. With typical values for the parameters
differential equation for single-phase ID flow in a porous medium: (Lx = 10 ft, c= 10 -5 psi -1, p, = 1.0 cp, cp =0.30, and k= 10 md),
I 0.6 il// .2
e 0.6
~ ,.,,:..../
~
!
CD
Oi
3:
0.4 // ~ 0.4
I6
Z
:1
'/
Coats' Fino Grid
Coats' Double Porosity Model
Now Modol, 0 - 0.035 ft2ID
J
Coats' Fino Grid
Coats' Double Poroslly Model
Now Modal, 0 _ 0.019 ft2ID
0.2 0.2
Fig. 5A-Match of the proposed 20 solution with Coats'20 Fig. 58-Match of the proposed 3D solution with Coats'20
fine-grid simulation example for a cubic block with high Pc. fine-grid simulation example for a cubic block with high Pc.
~ ~
.~ .~
e 0.6 e 0.6
~ ~
~
~ 0.4 I
Coats' Fine Grid ~ 0.4
Coats' Double Porosity Model Coats' Fine Grid
New Model. 0 _ 0.0095 ft2lD
) )
Coats' Double Porosity Model
New Model. 0 - 0.0100 ft2ID
0.2
0.2
0.0 ~L.L.L..L.1..LL:-=:..L.1..LLLJ...j~:-'-L.L.L..L.1._'_:_'__=_'..L.1..LLl...L...:'~.LLL.L.L..L.1..J..J
0.00 0.05 0.10 0.15 0.20 0.25 0.0 ~L.L.L..L.L-'-'-::-=:..L.L.J...LL.L.L-.J..J-:-'-L.L.L..L.L."7~.l...J...J...J...~...LLJ...J...u...J..I...J...:'-'.
0.00 0.25
Dimensionless Time. 10
the time required for pressure equilibrium, computed from Eq. 19, block. Water was injected into the bottom at a constant rate, and
is 0.0005 days. This can be considered instantaneous for most simu- fluids were produced from the top. They reported results for injec-
lation purposes. tion rates of 3.3 (low rate) and 35.0 cm 3 /min (high rate). Addi-
tional data on these experiments can be found elsewhere. 5,19
Results This example is very interesting because the effect of the varia-
tion of water saturation in the fracture on the imbibition can be test-
We tested the new model in several examples, ranging from single
ed. Beckner et at. 19 tried to match these experiments using several
matrix blocks to full-field-scale simulation. These include the lab-
published transfer functions and a model of their own. While they
oratory experiments made by Kleppe and Morse,5 some numeri- obtained good agreement with their own model, other models
cal examples reported by Coats,20 field-scale examples presented showed a poor match mainly at early times. To apply our model,
by Kazemi et at. 6 and Thomas et at., 11 and the SPE cross- we used an equivalent Cartesian system with the same volume and
sectional problem presented by Firoozabadi and Thomas. 27 same area open to flow because our analytical solution is for Carte-
sian coordinates.
Kleppe and Morse's Experiment. Kleppe and Morse 5 performed Fig. 3 shows oil recovery from the core vs. water injected for
a laboratory experiment on the waterflooding of an oil-saturated the high-rate case. The solid line refers to Kleppe and Morse's data.
core. They placed the core in a cylinder with a small space between Dashed lines correspond to different values of D used in our model
the two to simulate a fracture system surrounding a single matrix to reproduce the experiment. The match is good with D between
0.8
Thomas et al:. model
Kazemi et at's model
~ New Model. 0 _ 0.045 ft2ID
.~'" 0.6
~
!
i 0.4
Coats' Fine Grid
N
Coats' Double Porosity Model
J 0.2 ':
New Model. 0 - 0.0060 ft2lD
0.8 0.8
Thomas 01 al.
Coals
Dew _ O.035l121d
c:
~ 0.6
!
;3
! 0.4
Fig. 10-Comparlson of Thomas ef a/.'s11 IInedrlve example Fig. 11-Sensltlvlty to D for the SPE cross-sectional
results with the proposed model and Coats' 20 model. example. 27
0.25 and 0.35 cm 2 /min. The critical rate 4 ,28 for this case is esti- matched the results with D=0.045 ft21D. Fig. 9 shows WOR pre-
mated to be about 26 cm 3 /min, and the experiment was indeed dictions for more than 3 years. Our model shows a higher WOR
above this value. at early times and a lower WOR after 3 years. Increasing D will
Fig. 4 shows the same plot for the low-rate case. Again, the solid narrow this difference, indicating that Thomas et af. ' s II and Kaze-
line is for Kleppe and Morse's data. Clearly, the agreement is good mi et af.'s models predict a shorter transient imbibition than our
for D between 0.14 and 13.98 cm 2 /min. The results are not very model for this example.
sensitive to the value of D in this case because the injection was
below the so-called critical rate. 4,28 Thomas et al. 's 3D Example. Here the new model was used in
a 3D field-scale numerical example presented by Thomas et af. II
Coats' Examples. Coats 20 ran several numerical examples of sin- They simulated a linedrive water injection in a 2,OOO-ft square reser-
gle matrix blocks subjected to instantaneous immersion in water. voir with a thickness of 250 ft. Initial injection rate was 7,000
These examples can test the transient nature of the imbibition proc- STBID, constrained by a maximum bottomhole flowing pressure
ess. He used two types of geometric systems to emphasize gravity (BHFP) of 7,400 psig. Total liquid production was 4,000 STB/D.
effects: a lO-ft cubic block and a 30-ft-high parallelepiped. He also Other data can be found in Ref. 11. The value of D was obtained
used two capillary pressure curves (a high- and low-capillary- with the fine-grid simulation of a single matrix block subjected to
pressure curve) to emphasize different strengths of capillary imbi- instantaneous immersion, as reported in Coats' examples. The fluid
bition. The new function proposed in this paper was used to match and rock data here correspond to the cubic matrix block and high
his fine-grid results for these four cases. Because the problem in- Pc in Coats' problem, where we matched the results with
volves instantaneous immersion, it was not possible to account for D=0.035 ft21D. Fig. 10 shows water cut vs. time for a 10-year
the gravity effects properly in the new model. A constant value of period with this value of D. Our method gives practically the same
Swe, depending on the Pc curve and the matrix block size, was results as Coats' method and only slightly different results from
used for each case. Ref. 20 gives data for these examples. Thomas et af.'s method.
Fig. 5A shows the high-Pc results for the cubic block in terms
of the average normalized water saturation. For the 2D solution, SPE Cross-Sectional Problem. This section shows the results for
a good match is obtained with D=0.035 ft 2 /D. If we use the 3D a problem derived from the cross-sectional water injection study
solution instead, the match is almost perfect, as shown by Fig. 5B presented by Firoozabadi and Thomas. 27 The 2,000 x 1,000 x 250-
for D=0.019 ft21D. As Fig. 6 shows, a perfect match cannot be ft reservoir was discretized in a 10 x 5 grid system with all blocks
obtained for the parallelepiped with high Pc because gravity ef- measuring 200 x 1,OOOX50 ft. Water was injected at thex=Obound-
fects are not handled accurately in this case. Although we fit the ary at a maximum rate of 1,750 STB/D, constrained by a maxi-
data, we needed a different value of D from that used for the cubic mum BHFP of 6, 100 psig. Total liquid production was 1,000 STB/D
block. at the opposite boundary (x=2,000 ft). Refs. 11 and 27 give rock
For the low-Pc cases, the imbibition is much less dominating. and fluid data. The difference between this problem and that in Ref.
Fig. 7 shows the results for the cubic block, and Fig. 8 shows the 2 is that all matrix blocks have the same size here but were differ-
results for the parallelepiped. The match here is not as good as in ent in the original paper. An estimated value for D=O.OIO ft21D
the previous cases, probably because gravity effects are based on was used, the value that matched the fine-grid simulation of the
100% water saturation in the fracture, so the transient of the gravi- cubic matrix block shown in Fig. 7. Results in terms of water cut
ty effect is ignored. are shown in Fig. 11 for the lO-ft cubic matrix blocks. In this case,
a small sensitivity to D can be noticed for values >0.010 ft21D.
Kazemi et al. 's Example. Kazemi et af. 5 simulated the injection In Fig. 12, capillary diffusivity was kept constant and matrix block
of water in a five-spot quadrant. They used an 8 x 8 grid with blocks size was changed. Also, Swe was changed according to the rela-
of 75 ft on each side and 30 ft high. The matrix blocks are 10 ft tionship developed earlier, giving higher values for larger blocks.
on the sides and 30 ft high, so no' horizontal fractures exist. Also, Fig. 12 shows that smaller matrix blocks give lower water cut ini-
with the given Pc table, gravity is not important and we used a con- tially owing to faster imbibition. Fig. 13 shows oil recovery as a
stant value of Swe=0.70. To obtain the appropriate value of D, we fraction of the oil in place for the same runs. Clearly, lO-ft cubic
solved Eq. 4 numerically using given fluid and rock properties and matrix blocks gave higher recoveries for the period investigated
..............
#-......................... .. 0.25
0.8
/. ..............................................
........
....... 0.20
! . . .. ¥
::::.
...•.F ~
........ !
, '"~ 0.15
/;/
",
~
a:
5ft
0
' 10ft
25ft 0.'0
0.2 5 ft
10ft
0.05 25ft
,
........
......
0.0 0.u...u.c::<O:L:i...L..L.J....L.L..L...L...L-!-,-...L...L.J...J....L...L..LJ....L.J..J'4....L.JLL..LLLL,LLLW
16 S 20 6 S 10 12 '4 16 18 20
Time (years)
Fig. 12-Sensltivlty of water cut to block size for the SPE Fig. 13-Sensitivlty of oil recovery to block size for the SPE
cross-sectional example. 27 cross-sectional example. 27
for this example. The largest-block case eventually will give a higher 4> = porosity
recovery because of the lower residual oil saturation in the matrix <I> = potential, psi, m/Lt2
system, but if we have capillary continuity, then the block size will
not affect the ultimate recovery. Subscripts
c = connate
Discussion and Conclusions D = dimensionless
The new model presented in this paper has a number of advantages e = equilibrium
over previous approaches. f = fracture, final
1. The model takes into account the transient·nature of the imbi- i = initial
bition process and the effect of variation in fracture saturation. rna = matrix
2. Gravity effect is included in the calculation of Swe' n = normalized
3. The method is simple, and for the oil/water system, it requires
0= oil
only two equations per block, making it easy to implement in ex-
isting standard black-oil single-porosity simulators. p = pore
4. Results for single matrix block and field-scale problems show r = relative
good agreement with fine-grid simulations, previous models, and/or w = water
laboratory experiments. x, y, Z = x, y, z directions
While the approach presented is quite general, the current ver-
sion of the model is for two-phase oil and water flow. Further test- Superscript
ing of this approach with additional laboratory data and field-scale - = average
problems is needed. Extending this approach to handle steam in-
jection and tracer flow in fractured reservoirs may also be possible. Acknowledgment
We thank Petrobnis for the support given to this work.
Nomenclature
B = FVF, RBISTB, or convolution parameter References
c = compressibility, psi -\, Lt2/m
1. Barenblatt, G.!. etal.: "Basic Concept of the Single Phase Flow Through
D = capillary diffusivity, cm 2 /min, ft 2 /D, L2/t Fractured Porous Media, " Prikladnaia Matematika i Mechanika, Acade-
H = water/oil contact level in fracture, L mia Nauk, SSSR (1960) 24, No.5, 852-64.
k = permeability, md, L2 2. Barenblatt, G.!. and Zheltov, Ju. P.: "On the Basic Equations of the
L = dimension of matrix block, ft, L Single Phase Flow of Fluids Through Fractured Porous Media," Dok-
p = pressure, psi, m/Lt2 ladi Akademii Nauk, SSSR (1960) 123, No.3, 545-48.
Pc = capillary pressure, psi, m/Lt 2 3. Warren, J.E. and Root, PJ.: "The Behavior of Naturally Fractured
q = flow rate, L3 It Reservoirs," SPEl (Sept. 1963) 245-55; Trans., AIME, 228.
4. Mattax, C.C. and Kyte, J.R.: "Imbibition Oil Recovery From Frac-
S = saturation, fraction tured Water Drive Reservoirs," SPEl (June 1962) 177-84; Trans.,
Swn = average normalized saturation in matrix block AIME,225.
S.t" = normalized saturation distribution in matrix block 5. Kleppe, J. and Morse, R.A.: "Oil Production From Fractured Reser-
(solution of Eq. 4) voirs by Water Displacement," paper SPE 5084 presented at the 1974
t = time, t SPE Annual Meeting, Houston, Oct. 6-9.
T = transmissibility, STBID-psi, m 2 /Lt 6. Kazemi, H. etal.: "Numerical Simulation of Water-Oil Flow in Natu-
V = volume, ft3, L3 rally Fractured Reservoirs," SPEl (Dec. 1976) 317-26; Trans., AIME,
261.
'Y = specific weight, psi/ft, mL2/t 2 7. Kazemi, H. and Merrill, 1.S.: "Numerical Simulation of Water Imbi-
8 = dummy variable bition in Fractured Cores," SPEl (June 1979) 175-82.
~, h = model parameters 8. Gilman, J.R. and Kazerni, H.: "Improvements in Simulation of Natu-
p. = viscosity, cp, mILt rally Fractured Reservoirs," SPEl (Aug. 1983) 695-707.