Sunteți pe pagina 1din 12

Analysis and Interpretation of Water/Oil

Ratio in Waterfloods
Y.C. Yortsos, SPE, Youngmin Choi,* and Zhengming Yang,** SPE, U. of Southern California
and P.C. Shah,† SPE, Dowell

Summary In a more recent study, Chan6 used numerical simulation to


An important problem in water control is the identification of the examine the sensitivity of the curve of WOR produced vs. pro-
dominant reservoir or production mechanisms. Recently, Chan duction time on various reservoir and production factors. He con-
共Chan, K.S.: ‘‘Water Control Diagnostic Plots,’’ paper SPE 30775 jectured that a log-log plot of this curve can be used to diagnose
presented at the 1995 SPE Annual Technical Conference and Ex- the origin of the water production, and specifically to determine
hibition, Dallas, 22–25 October兲 postulated that a log-log plot of whether it is due to channeling 共heterogeneity兲 or to coning. An
the water/oil ratio 共WOR兲 produced vs. production time may be essential part of his conjecture is that a log-log plot of WOR vs.
used to diagnose these factors. In this paper we provide analytical production time contains linear segments, the slopes of which are
and numerical results for a variety of waterflood conditions to different in the cases of channeling or coning, hence they can be
explore this possibility. used for diagnostic purposes. No expressions for these slopes
We show analytically that the late-time slope of the log-log plot were postulated or derived, however. Chan’s conjecture is pre-
can be related to the well pattern and the relative permeability sumed to have strong support from numerical simulations and
characteristics, specifically the power-law exponent in the field data.
k ro ⫺S w relationship or the reservoir heterogeneity. Analytical re- Motivated by Chan’s work we provide in this paper a funda-
sults are provided for the behavior immediately following water mental investigation by conducting analytical and numerical stud-
breakthrough. In certain simple cases 共mobility ratio equal to 1, ies of waterflooding under a variety of conditions. The key objec-
layered systems兲, analytical type curves can be derived. In the tive is to analyze the behavior of the WOR vs. time curve in
general case, we use numerical simulation to provide numerical various time domains 共for example, following breakthrough or at
type curves. late times兲 and to develop a methodology for interpreting the be-
The results are summarized in terms of the various power-law havior observed. There are two important differences from the
scalings in different time regimes and by a catalog of numerical previous works: We extend the 1D analysis by considering the
type curves. They confirm the potential of WOR-time plots as more realistic power-law dependence of the relative permeability
diagnostic tools for reservoir analysis and characterization. ratio at high water saturations, and we also analyze the behavior
of multidimensional patterns. In certain cases, analytical expres-
sions are derived under simplifying assumptions, which are sub-
sequently supported by numerical studies. In this paper, the main
focus is on waterflooding.
Introduction In order of increasing complexity the analysis proceeds by in-
An important problem in water control is the identification of vestigating the following:
dominant reservoir or production factors 共for example, channeling 䊉 one-dimensional 共single-layer or homogeneous兲 displace-
or water coning兲 from produced water/oil data. Water production ment;
can be the result of a natural waterdrive, edge or bottom water- 䊉 two-dimensional homogeneous displacement in various pat-

drive or a waterflood. Despite its relevance, however, the problem terns;


has received rather scant attention in the literature so far. Aspects 䊉 two-dimensional heterogeneous displacement 共channeling兲;

of waterflooding under various conditions have been extensively 䊉 three-dimensional displacement 共including layering兲.

described in the literature 共see, e.g., the SPE monographs of


Craig1 and Willhite2兲. Limited, however, are studies on the depen- We show that the X-plot approach is a special case of the 1D
dence of the water/oil ratio 共WOR兲 on production time. Based on displacement at intermediate times. At later times, the log-log plot
the solution of the one-dimensional 共1D兲 Buckley–Leverett equa- of the WOR vs. time is asymptotically a straight line, with a slope
tion and under a certain assumption on the functional relation of that reflects relative permeability and fractional flow effects in the
the relative permeability ratio, Ershaghi and Omoregie3 共see also 1D case and the particular pattern geometries in the 2D and 3D
Ershaghi and Abdassah4兲 proposed the so-called X plot to inter- cases. Ultimately, and at sufficiently large times, all patterns re-
pret and extrapolate water/oil production. In this approach, a com- flect 1D displacement behavior. Various properties of log-log
bination of the fractional flow function was found to vary linearly plots of the WOR vs. time are also established for heterogeneous
with the cumulative oil production in an appropriate semilog plot. systems represented as a permeability streak and layers, respec-
The X-plot method successfully matched various field data. tively. In the latter case, effects of cross flow and communication
Lo et al.5 explored the applicability of the X plot further by between layers are also discussed. Finally we provide an interpre-
conducting numerical simulations in two-dimensional 共2D兲 and tation methodology for the analysis of the WOR vs. time re-
three-dimensional 共3D兲 systems and by investigating various ef- sponse. Numerical simulations confirm these findings in the ap-
fects. They concluded that a linear relationship between the propriate geometries.
ln共WOR兲 and the cumulative oil production adequately fit many Before we proceed we note that the analysis below is in terms
of their results. Like Ershaghi and Omoregie,3 they interpreted of the dimensionless time 共fraction of pore volumes injected兲, in
their findings using the 1D Buckley–Leverett equation using the order to compare our results with Chan.6 Extending the results to
previously assumed dependence on the relative permeability ratio. WOR vs. cumulative oil recovery plots as in Lo et al.5 is straight-
forward and discussed in Appendix A. Reference to these plots
will be made where appropriate.
*Now with Jason Natural Products.
**Now with MSC Software Corp.

Now with Landmark Graphics Corp. Analysis
Copyright © 1999 Society of Petroleum Engineers The dependence of the WOR on time 共or oil produced兲 results
This paper (SPE 59477) was revised for publication from paper SPE 38869, presented at from the interaction of two effects: the flow rate partition in dif-
the 1997 SPE Annual Technical Conference and Exhibition held in San Antonio, Texas,
5–8 October. Original manuscript received for review 18 December 1997. Revised manu-
ferent streamtubes, due to the flow pattern, the viscosity ratio,
script received 3 December 1998. Manuscript peer approved 13 September 1999. layering and heterogeneity, and the displacement in a given

SPE Journal 4 共4兲, December 1999 1086-055X/99/4共4兲/413/12/$3.50⫹0.15 413


streamtube, which is affected by the viscosity ratio M and frac-
tional flow. For the case of a passive tracer 共M ⫽1, no fractional
flow effects兲, the streamlines are fixed, dictated by the well pattern
and the system heterogeneity, and the displacement within a
streamtube is one dimensional. For a waterflood with variable M
and fractional flow effects, these assumptions will fail, in prin-
ciple. However, when the mobility ratio is close to unity, the
resulting error is expected to be small, and an approach based on
decoupling areal heterogeneity and fractional flow effects appears
to be reasonable, at least as a first approximation. The problem
then reduces to 共i兲 understanding the features of 1D displace-
ments, and 共ii兲 obtaining the probability distribution, p(t), of
breakthrough 共arrival兲 times of various multidimensional prob-
lems. This simplifying approach is adopted for analytical purposes
in this paper, where the above two topics are discussed separately.
In the first, we extend the Buckley–Leverett analysis beyond the
range considered by Ershaghi and Omoregie.3 The second topic,
involving multidimensional effects, has not been addressed ana-
lytically in the literature so far. Throughout this paper an immo-
bile initial water saturation is assumed, so that before water break-
through the WOR is negligibly small.

1D Displacements. Consider a 1D immiscible displacement 共lin-


ear or radial兲, which in the absence of capillarity is described by
the traditional Buckley–Leverett equation. Although relatively
simple, the behavior of 1D displacements is important, since it
emphasizes petrophysical 共relative permeability and fractional
flow兲 effects. As shown below, it also controls the late-time be-
havior of more general problems 共for example, that of a layered
system after water breaks through in all layers兲. In appropriate
dimensionless variables, with x expressing the fractional distance
between the injector and the producer and t the number of pore
volumes injected, the standard conservation equation applies,
⳵Sw ⳵ f w
⫹ ⫽0, 共1兲
⳵t ⳵x Fig. 1–Relative permeability ratio ␬ „ Sw … for two different expo-
where f w is the fractional flow function, nents b plotted vs. „a… the X-plot coordinates or „b… the power-
law coordinates „here S or ⴝ0.25….
1
f w⫽ , 共2兲
␬共 Sw兲
1⫹
M
where A,B⬎0 are appropriate constants. This is the essence in the
expressed in terms of the viscosity ratio, M ⫽ ␮ o / ␮ w , and the Ershaghi and Omoregie3 X-plot method. For the hypothetical case
relative permeability ratio, ␬ ⫽ k ro /k rw . Note that Eq. 1 applies in Fig. 1共a兲, this regime applies up to S w* ⫽0.5. Use of Eq. 6 in
also for variable injection rates. The solution of Eq. 1 is well Eqs. 4 and 5 then leads to
known,
共 1⫹W 兲 2
x ⫽Bt. 共7兲
⫽ f w⬘ 共 S w 兲 . 共3兲 W
t
Note that at large W and under the condition that the exponential
Then the saturation, S w* , at the producer (x⫽1) is implicitly behavior is still obeyed, Eq. 7 suggests a linear log W-log t plot
given as a function of time from the solution of with slope 1.
䊉 On the other hand, for values of S * relatively close to 1
␬ ⬘ 共 S w* 兲 M w

冉 冊
⫽⫺ , 共4兲 ⫺S or , ␬ can be approximated by a power law,
␬ 共 S w* 兲 2
t
1⫹ ␬ ⫽a 共 1⫺S w ⫺S or 兲 b , 共8兲
M
where a⬎0 is a constant, and b⬎1 is the exponent of the relative
obtained by substituting Eq. 2 in Eq. 3. Denoting by W the WOR
permeability to oil. In deriving Eq. 8 it was assumed that k ro
at the producing well, it is then easily shown that
varies in this region as
M
W⫽ . 共5兲 k ro ⬃ 共 1⫺S w ⫺S or 兲 b , 共9兲
␬ 共 S w* 兲
as can be verified by percolation theory 共for example, see Ref. 7兲.
Hence, to obtain the relation between W and t, one needs to first For the hypothetical case of Fig. 1共b兲, where b⫽2.1, this regime
solve Eq. 4 for S w* and subsequently substitute the result in Eq. 5. applies after S w* ⫽0.55. Use of Eq. 8 in Eqs. 4 and 5 leads to
Clearly, this relation will be directly influenced by the assumed
dependence of the ratio ␬ on the saturation. 共 W⫹1 兲 2
An inspection of a typical plot of ␬ vs. S w 共Fig. 1兲 shows the ⫽ba 1/b M ⫺ 共 1/b 兲 t, 共10兲
W 1⫹ 共 1/b 兲
following two different regimes.
䊉 At intermediate values of S * , ␬ can be roughly approximated
which at large W suggests a linear log W-log t plot, but now with
w
by the exponential slope b/(b⫺1).
We will use these two different limiting regimes to analyze the
␬ ⫽A exp共 ⫺BS w 兲 , 共6兲 properties of 1D displacements. We will focus on two different

414 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999
Fig. 4–Log-log plot of the WOR vs. time for a 1D displacement
Fig. 2–Behavior following breakthrough using X-plot coordi- and M ⴝ0.5, 1, 2, 5, and 10. The late-time slope is equal to 1.5.
nates †X ⴝ1/t B ⴚ1/t , Y ⴝ W B /„1ⴙ W B … 2 ⴚ W /„1ⴙ W … 2 ‡, for M
ⴝ0.1, 1, 2, 5, and 10.
expected, Eq. 12 is satisfied better under conditions of more fa-
vorable mobility.
time regimes, one following breakthrough and a late-time regime. Either Eq. 11 or 12 can be linearized near t B to yield the scal-
Equivalent results for the relationship between W and the dimen- ing,
sionless cumulative oil production, Q o , are given in Appendix A.
W⫺W B ⬃t⫺t B , 共13兲
Behavior Following Breakthrough. Of most interest in practi-
cal applications is the early part of the WOR curve that immedi- which indicates a straight line of slope 1 in a log-log plot of
ately follows breakthrough. For unfavorable mobility displace- increments of the WOR vs. increments of time. The numerical
ments, where S w* is relatively small, hence where X-plot results of Figs. 2 and 3 confirm this scaling at early times. In
conditions prevail, we can use the Buckley–Leverett theory above principle, Eqs. 11 through 13 can be used for diagnostic purposes.
to obtain the following equation after breakthrough: However, such tests will be sensitive to the particular choice of

冋 册
W B , the value of which needs to be estimated by other means.
1 1 W WB Late-Time Behavior. Following breakthrough, the saturation at
⫺ ⫽B 2⫺ ; W B ⬎1. 共11兲
t tB 共 W⫹1 兲 共 W B ⫹1 兲 2 the producing end is relatively high. Again, the WOR behavior
depends on the particular regime of the relative permeability de-
Here t B and W B are, respectively 共dimensionless兲 breakthrough pendence. For sufficiently large M , such that the water saturation
time and the WOR at breakthrough 共which in the Buckley– at breakthrough probes the X-plot region 共where Eq. 6 is appro-
Leverett theory is nonzero兲. In an appropriate log-log plot, there- priate兲, Eq. 7 can be rearranged to yield
fore, this equation represents a straight line of slope 1. On the
other hand, for more favorable mobility displacements, where the 共 1⫹W 兲 2
breakthrough saturation is relatively high and where the power- ⫽B. 共14兲
Wt
law regime is obeyed, the equation near breakthrough is

冋 册
Under such conditions, the ratio 关 (1⫹W) 2 /Wt 兴 is independent of
1 1 ba 1/b W b⫹1/b W Bb⫹1/b b⫹1 production time. Furthermore, at large W 共assuming that exponen-
⫺ ⫽ 1/b ⫺ ; W B⬎ . 共12兲
t tB M 共 W⫹1 兲 2 共 W B ⫹1 兲 2 b⫺1 tial behavior is still obeyed兲, Eq. 14 can be approximated by
This equation is also a straight line with a unit slope in the appro- log W⬇log t⫹C, 共15兲
priate log-log plot.
where C is a constant, and which shows that a log W-log t plot is
The validity of Eqs. 11 and 12 was tested numerically using a
linear with slope 1.
commercial finite-difference simulator. Fig. 2 shows that, for
For sufficiently low M , on the other hand, or at sufficiently
rather large M and at early times following breakthrough, Eq. 11
large times and for any M , the power-law regime, Eq. 8, applies.
is well obeyed. Fig. 3 shows the corresponding test for Eq. 12. As
In contrast to the X-plot prediction, Eq. 14, the ratio 关 (1
⫹W) 2 /Wt 兴 is not constant but varies with time following Eq. 10,
which can be rearranged to
共 1⫹W 兲 2
⬃W 1/b . 共16兲
Wt
As W increases, this can be further simplified to
W 1⫺ 共 1/b 兲 ⬇ba 1/b M ⫺ 1/b t 共17兲
or
b
log W⬃ log t⫹H, 共18兲
b⫺1
where H is a constant. Under these conditions, therefore, a plot of
log W vs. log t is asymptotically a straight line, but now with slope
b/(b⫺1). This slope provides information on the exponent of the
power-law dependence of the oil’s relative permeability on satu-
ration. As will be shown below using numerical simulations, this
asymptotic behavior is eventually common to all systems tested.
Fig. 3–Behavior following breakthrough using power-law coor- Using the numerical simulator we proceeded to test numeri-
dinates †X ⴝ1/t B ⴚ1/t , Y ⴝ W Bbⴙ1/ b /„1ⴙ W B … 2 ⴚ W b ⴙ1/ b /„1ⴙ W … 2 ‡, cally the validity of these two regimes for practically relevant
for M ⴝ0.1, 1, 2, 5, and 10. values of the WOR. Figs. 4 and 5 show log-log plots of the WOR

Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999 415
Fig. 7–Well-doublet geometry.

The interpretation of these results in terms of the 共normalized兲


cumulative oil production, Q o , is straightforward as is shown in
Appendix A. Under X-plot conditions, the behavior is linear in a
semilog plot,
ln W⬇BQ o ⫹C ⬘ , 共19兲
Fig. 5–Log-log plot of the WOR vs. time for a 1D displacement
and M ⴝ0.5, 1, 2, 5, and 10. The late-time slope is equal to 2. where C ⬘ is a constant. At later times, however, the behavior is a
power law,
ln W⬇⫺b ln共 1⫺Q o 兲 ⫹H ⬘ , 共20兲
vs. dimensionless time 共expressed in pore volumes兲 from 1D where H ⬘ is another constant. Eq. 20 is physically more appropri-
simulations for different values of M and for values of b equal to ate than Eq. 19 in that it predicts an infinitely increasing WOR as
3 and 2, respectively. Based on the latter, the theoretically ex- Q o approaches the maximum value of 1, in constrast to Eq. 19,
pected late-time slope in Figs. 4 and 5 is 1.5 and 2, respectively. which predicts a finite WOR value at complete oil recovery. We
The numerical results show that for favorable viscosity ratios 共M can approximate Eq. 20 to read like Eq. 19 if Q o is not large, in
of the order of 1兲, the asymptotic slope agrees quite well with the which case we may make the approximation ln(1⫺Qo)⬇⫺Qo ,
theoretical predictions. For unfavorable mobilities, the slope at hence
early times is smaller than that predicted by Eq. 18, although it
eventually approaches the theoretically expected value. This is ln W⬇bQ o ⫹H ⬘ . 共21兲
also consistent with the theory: at large M and small b, the water This is similar to Eq. 19, but now with b in place of B. This
saturation at breakthrough is relatively small, hence X-plot condi- approximation will not be valid as Q o approaches 1, however.
tions are applicable. According to Eq. 15 the corresponding slope
would be approximately equal to 1, which is indeed observed in 2D Displacements in Homogeneous Systems. Consider next 2D
the respective regions of Figs. 4 and 5. Another way of testing displacements. Here, the fluid production and the WOR history
whether the system probes the power law or the X-plot regime is will also be affected by the particular well pattern. The problem
to plot the ratio 关 (1⫹W) 2 /Wt 兴 , which should remain constant if can be quite complex, particularly when heterogeneity and vary-
X-plot conditions apply. Fig. 6 shows that for larger M 共e.g., M ing viscosity are involved. We will proceed by using the approxi-
⫽10 in Fig. 6兲, the ratio remains approximately constant for a mation that the WOR history is the contribution of two effects:
substantial range of values of W, indicating the validity of the X one due to the displacement pattern 共streamtubes兲 in the absence
plot in such cases. On the other hand, for smaller values of M , or of relative permeability or viscosity ratio effects, and another due
at later times, this ratio varies and obeys instead the power-law to mobility and relative permeability effects, as in 1D above. As
behavior, Eq. 16. long as the mobility ratio is not very different from unity, this
In summary, in 1D displacements, the early behavior following approximation is expected to be reasonably valid. In this section
breakthrough is characterized by increments in W being propor- we will consider homogeneous and symmetric patterns. Effects of
tional to increments in t 共the more extended Eqs. 11 or 12, respec- layering are discussed later.
tively, could also be used兲; the late-time behavior exhibits power- Well Doublet in an Unbounded Reservoir. Consider first a
law scaling with slope b/(b⫺1), preceded in the case of well-doublet geometry in an unbounded reservoir 共Fig. 7兲. Within
unfavorable mobilities by the X-plot regime, which may also be each streamtube, the displacement will be approximated as piston
approximated as a power law with slope 1. Since b⬎b⫺1, both like, hence contrary to the previous section, where the WOR ex-
these slopes are larger than or equal to 1. This observation should presses relative permeability effects, here it reflects solely areal
be important for diagnostic purposes in higher dimensions 共see displacement, since different streamtubes will break through at
below兲. different times. For a homogeneous system, the solution to this
problem can be found analytically.
Denote by ␣ the angle made by an arbitrary streamline emanat-
ing from the injector and the straight line connecting the injector
and the producer 共Fig. 7兲. We will use this angle to parametrize
the problem. Water breakthrough occurs when the front in the
streamline with angle 0° 共the straight line connecting the injector
and the producer兲 first breaks through. At any given time, the
front in a different streamtube, with angle ␣, breaks through. Then
the instantaneous fractional flow at the production well is

f w⫽ , 共22兲

hence, the WOR can also be expressed in terms of ␣,

W⫽ . 共23兲
␲⫺␣
To relate W to time, one needs to evaluate the time of water
Fig. 6–Characteristic plot showing that X-plot conditions apply breakthrough 共also denoted as the arrival time兲 of a streamtube
for relatively large M „ b ⴝ2.0…. with arbitrary angle ␣. In well-doublet geometry, this can be ob-

416 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999
共note again that the analytical solution corresponds to an infinite
reservoir, while the rescaling of time in the time axis reflects the
assumed pore volume of the finite numerical reservoir兲. For the
varying M curves, a qualitatively similar behavior is shown that
lasts for a period of time, during which the slope is approximately
the same as the analytical. Following this period the curves do not
reproduce the analytical predictions for several reasons: The
curves start reflecting the finite-size domain of the numerical cal-
culations, and possibly the lack of grid refinement at the wells;
capillary effects start to become important 共the latter effect was
also present in all our simulations at a sufficiently high WOR兲;
finally, variable mobility effects also become important, and may
lead to viscous instabilities, etc. These were not analyzed in this
paper.
The results of Fig. 8 show that the well pattern and 2D geo-
metrical features dominate over fractional flow effects. Indeed, the
Fig. 8–Log-log plot of the WOR vs. time for a well doublet and
1D model predictions would have suggested an asymptotic slope
M ⴝ0.1, 0.5, 1, 2, 5, and 10 „ b ⴝ1.5…. The solid line is the ana-
lytical prediction. The late-time slope is 1/3. of 3, which is clearly not observed in Fig. 8. We note that this is
a consequence of the unbounded nature of the reservoir in the
problem considered here. When the reservoir is bounded, the
curve ultimately approaches 1D asymptotic behavior, as will be
tained rather easily, since all streamlines are circular.8 共This was demonstrated below. Similar results were found for other values
also discussed by Brigham in 1995 unpublished notes.兲 Then one of b studied.
can show the exact result, In essence, the well doublet in an unbounded reservoir is dia-

冉 冊
sin ␣ ⫺ ␣ cos ␣ metrically opposite to that of the 1D problem. While the former
t⫽ , 共24兲 emphasizes relative permeability effects, the latter is dominated
sin3 ␣ by areal displacement features. The analysis of this behavior is
where time was dimensionalized with the characteristic time novel. As will be shown below, however, most realistic 2D prob-
4 ␲ L 2 ␾ h/ v 共where ␾ is porosity, h the reservoir thickness, 2L the lems share common features with both these limits.
distance between the two wells, and v the volumetric injection General Patterns. Consider next more general patterns. As be-
rate兲. Eqs. 23 and 24 implicitly provide the relationship between fore, to analyze these problems we will decouple geometry from
W and t. As before, we will analyze the displacement behavior in fractional flow effects. To evaluate the effect of the pattern, we
the two different time domains: following breakthrough and at will make use of the findings of Koplik et al.,9,10 who considered
late times. the problem of tracer dispersion in 2D multipole and bounded
Behavior Following Breakthrough. To obtain the behavior potential flows. These investigators analyzed the probability dis-
following breakthrough we expand Eqs. 23 and 24 at small ␣. tribution, p(t), of the arrival times, namely, the time t it takes for
Then we can readily show the power-law scaling, a tracer particle released at a source to reach the sink well. In our
context, the arrival time also represents the time for water break-
2 ␲ 2t B 2 through from a given streamtube in a waterflood with M ⫽1 and
t⫺t B ⬇ W , 共25兲 in the absence of fractional flow effects. Therefore, results on p(t)
5
can be potentially useful to our problem. The well-doublet prob-
which can be further rearranged to lem discussed above is a particular case of the more general mul-

W⬇ 0.25 冉 t⫺t B
tB 冊 1/2
. 共26兲
tipole problem treated here.
To obtain an expression for the WOR behavior, it is necessary
to relate W to p, which is obtained as follows. Consider first the
In the appropriate log-log plot, Eq. 26 indicates a straight line case where fractional flow effects are not included, such that the
with slope 1/2. Note that in contrast to 1D displacements 关e.g., displacement is piston like within a streamtube. Following water
Eqs. 11 and 12兴, here the value of the WOR at the time of water breakthrough, only water will be produced from that streamtube.
breakthrough is zero (W B ⫽0). Also note that the scaling expo- Under this assumption, it is easy to show that
nents are different in the two expressions, Eqs. 13 and 26.
t
Late-Time Behavior. Late-time behavior can be obtained from 兰 t * p 共 t B 兲 dt B 1
an asymptotic analysis of Eqs. 23 and 24 in the limit ␣ → ␲ . Then W共 t 兲⫽
B
⫽ ⫺1, 共28兲
it is straightforward to show the power-law scaling, 兰 t⬁ p 共 t B 兲 dt B 兰 t⬁ p 共 t B 兲 dt B
W⬃t 1/3, 共27兲
where we made use of the normalization condition,
which in a log-log plot represents a straight line of slope 1/3. This
slope is smaller than unity. This feature is generic to multidimen-
sional displacements, as will be shown below, and in contrast to
Eq. 18 reflects areal displacement, not relative permeability, ef-
冕 t*
B

p 共 t 兲 dt⫽1. 共29兲

fects. To test the applicability of the analytical result, Eqs. 23 and


24, we conducted a numerical simulation of the general immis- When relative permeability and fractional flow effects are in-
cible displacement problem in this geometry. Fig. 8 shows log-log cluded, Eq. 28 must be modified. Let us denote by W s the WOR
plots of the WOR vs. time obtained for the well-doublet geometry corresponding to 1D displacement in a single streamtube 关e.g.,
in a large reservoir for different values of M and for b⫽1.5. The Eqs. 4 and 5 for the 1D displacement discussed previously兴. Be-
analytical predictions are depicted by a solid line. We note the cause different streamtubes have different lengths and different
following. breakthrough times, t B , the WOR of a given streamtube at a given
The analytical solution agrees quite well with the numerical for time t must be evaluated based on the normalized time t/t B .
M ⫽1. Given the approximations inherent to the numerical solu- Thus, in a given streamtube, the volume rate fraction of water will
tion, particularly the fact that the numerical domain is bounded, in be given by 关 W s (t/t B )/(1⫹W s (t/t B )) 兴 , while that of oil will be
contrast to the analytical, and that no grid refinement was done 关 1/(1⫹W s (t/t B )) 兴 . The overall WOR will consist of the sum of
near the wells, where the flow is radial, the agreement is actually contributions from those streamtubes where water has broken
very good. The late-time slope sets at about t⫽0.005 in Fig. 8 through and from those where it has not. Then,

Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999 417

t*
B
t W s 共 t/t B 兲
1⫹W s 共 t/t B 兲
p 共 t B 兲 dt B
W共 t 兲⫽ 共30兲
冕 冕
,
t p 共 t B 兲 dt B ⬁
⫹ p 共 t B 兲 dt B
*
tB 1⫹W s 共 t/t B 兲 t

which can also be written as


1
W共 t 兲⫽ ⫺1. 共31兲
冕 t

*
tB
p 共 t B 兲 dt B
1⫹W s 共 t/t B 兲
⫹ 冕t

p 共 t B 兲 dt B

To evaluate Eq. 30 or 31 we need an expression for the 1D WOR


function, W s , and of the breakthrough time distribution, p(t). Fig. 9–Log-log plot of the WOR vs. time for a quadrupole and
Again we will consider two different time regimes, near break- M ⴝ0.1 „ b ⴝ3.0…. The late-time slope is 0.5.
through and late times.
Behavior Following Breakthrough. Under the assumption that
the production geometry is symmetric 共e.g., a symmetric five spot, N/(N⫹2), and in the appropriate log-log plot a straight line with
etc.兲, the argument used for the well doublet applies here as well. slope N/(N⫹2). This slope is always less than 1, in contrast to the
We expect that the time of breakthrough is an even function of the 1D case where the slope is always greater than 1. This difference
streamfunction, ␺, of the streamtube that has broken through, could be quite useful for diagnostic purposes. As expected, for the
namely, that case of a doublet (N⫽1), Eq. 36 coincides with Eq. 27, where the
t⫺t * ⬃ ␺ 2 . 共32兲 exponent 1/3 was predicted. The quadrupolar flow case (N⫽2)
gives a slope of 1/2. An isolated five spot (N⫽4) has a late-time
Right after breakthrough, we have the scaling W⬃ ␺ . Combining power law scaling with exponent 2/3, etc. The corresponding plot
it with Eq. 32 gives in WOR-oil recovery coordinates is
W⬃ 共 t⫺t * 兲 1/2 共33兲 N
ln W⬃ ln Q o , 共38兲
which suggests a straight line in the appropriate log-log plot with 2
slope 1/2. This result is identical to the well-doublet problem.
as will be seen in Appendix A. In contrast to Eq. 19, this is a
Various numerical results indicated that such behavior is reason-
power law in the present case of an unbounded reservoir.
ably well obeyed in a small interval following breakthrough for
In bounded geometries and in the absence of fractional flow
favorable values of M .
effects, Eq. 37 suggests an exponential dependence of the WOR
Late-Time Behavior. To evaluate the late-time behavior for a
on time. However, it will be shown shortly that, in this case, the
general pattern, the properties of the arrival time distribution p(t)
1D power-law scaling associated with fractional flow effects ac-
are needed. For the general multipole flow problem in unbounded
tually overwhelms this exponential behavior and leads to the pre-
geometries, Koplik et al.9,10 derived the following asymptotic be-
vious 1D power-law scaling.
havior: We consider Eq. 31 with fractional flow effects. To uncover its
p 共 t 兲 ⬃t ⫺ 共 2N⫹2 兲 / 共 N⫹2 兲 , 共34兲 asymptotic behavior at late times, we must make use of the
asymptotic properties of W s (t), which for the present purposes we
where N indicates the polarity of the pattern: N⫽1 corresponds to will take to scale as W s ⬃t m , where m⫽b/(b⫺1), as described in
the well-doublet problem, N⫽2 is the quadrupolar flow case, Eq. 18. The analysis is somewhat elaborate and is presented in
namely, two injectors on a line symmetrically opposite to a pro- Appendix B, where we show the following.
ducer with rates q, q and ⫺2q, respectively, while N⫽4 corre- 䊉 When p(t) decays following a power law, as in the case of
sponds to an isolated five spot, namely, one source and four sinks multipole flows, the WOR scaling is the same as in the case with-
symmetrically placed. These problems correspond to unbounded out fractional flow effects, Eq. 36, provided that m⫽ b/(b⫺1)
geometries 共but also to bounded geometries, when the effect of ⬎N/(N⫹2). The latter is obviously satisfied for all values of b
the boundaries has not yet been felt兲. and N since 2b⫹N⬎0. One concludes that for typical cases, and
Patterns in bounded geometries, as is the case of a five-spot before boundary effects are felt in the WOR, the late-time scaling
pattern, produce stagnation points at corners, and retard flow reflects the particular pattern geometry only, even if fractional
along neighboring streamlines. Then the breakthrough-time distri- flow effects are considered. From the corresponding slope N/(N
bution becomes exponentially slow,9,10 ⫹2), the value of N, and hence that of the flow polarity, could be
p 共 t 兲 ⬃exp共 ⫺Gt 兲 , 共35兲 identified.
䊉 When boundary effects are felt, however, where p(t) follows

where G⬎0 measures the rate of stretching of the streamlines the exponential decay, the asymptotic scaling of W s (t) dominates
near stagnation points. The difference between the power law, Eq. the asymptotic behavior, hence the WOR scaling would follow
34, and the exponential decay, Eq. 35, when the geometry be- that of 1D displacement, which predicts a slope b/(b⫺1). Thus,
comes bounded must be emphasized. the 1D slope will eventually emerge at sufficiently late times in
Consider then the application of these scalings to the WOR any problem. The corresponding scaling of the WOR-oil recovery
expressions, Eq. 28 or 31. If we neglect fractional flow effects, relationship is now
Eq. 28 applies, and from it we obtain the late-time scaling,
ln W⬇⫺b ln共 1⫺Q o 兲 ⫹H ⬘ , 共39兲
W⬃t N/N⫹2 , 共36兲
as in Eq. 20, which as we noted above can also be approximated
for the case of multipole flows in unbounded geometries, followed by the semilog relationship, Eq. 21.
by the scaling, We tested the above theoretical predictions by numerical simu-
lation. Fig. 9 shows results for the case of a quadrupole (N⫽2) in
W⬃exp共 Gt 兲 , 共37兲
an unbounded reservoir. We observe that the theoretical slope of
in bounded geometries. Eq. 36 predicts that in unbounded geom- 1/2 is well satisfied. Numerical simulations were subsequently
etries 共or at relatively early times, when the effect of boundaries is conducted to test the late-time behavior of displacements in
not yet felt兲 the late-time behavior is a power law with exponent bounded geometries. Fig. 10 shows results obtained for a well

418 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999
Fig. 10–Log-log plot of the WOR vs. time for a bounded well Fig. 12–Log-log plot of the WOR vs. time for a bounded well
doublet with aspect ratios of B / A ⴝ1, 2, and 3 „M ⴝ1, b ⴝ3.0…. doublet with aspect ratios of B / A ⴝ1, 2, and 3 and of a doublet
The late-time slope is 1.5. with a heterogeneity channel „M ⴝ1, b ⴝ3…. The late-time slope
is 1.5.

doublet in a bounded geometry, with aspect ratios varying in the


range of 1 to 3 for M ⫽1 and b⫽3. The asymptotic limit is a
power law with a slope nearly equal to 3/2, which corresponds to etry. A vertical cross section of a multilayered system was taken
the 1D theoretical predictions with b⫽3 共rather than a slope of as a model of a layered reservoir. The latter will be discussed in
1/3, as predicted in the unbounded case兲. Similar results were also the section on 3D displacements. In both of these problems we
found for b⫽2, where the theoretical slope of 2 was confirmed. A adopted the same approach to approximate the behavior of the
more common bounded geometry is that of a five spot. Simula- WOR as above, namely, we combined the statistics of the arrival
tions for this geometry were conducted for various values of M times, p(t) 共obtained by an analysis of the M ⫽1 problem in the
and different values of b. Fig. 11 shows results for b⫽2. When absence of fractional flow effects兲 with the dependence in time of
M ⫽1, one can identify a late-time behavior approaching a power- the WOR for a single streamtube 共obtained by an analysis of a 1D
law scaling with slope 2, which is consistent with the theoretical displacement that includes fractional flow effects兲. As pointed out
value b/(b⫺1). Similar agreement was found for other values of above, this approximation, where one neglects cross flow between
the parameter b tested. In all cases, however, the agreement be- adjacent streamtubes due to the mobility contrast, becomes less
comes less satisfactory as the mobility ratio deviates from unity valid as the mobility ratio deviates from unity. The effect of cross
and b becomes smaller. The origin of this discrepancy is due to flow was studied numerically.
the changing mobility during the flow, as well as to the fact that In the presence of a high-permeability channel between two
X-plot conditions are in effect for a longer period of time. wells, the most significant effect is felt immediately following
We conclude that in 2D patterns, the early behavior following breakthrough. Mathematically, the presence of a channel can be
breakthrough in symmetric homogeneous patterns follows scaling, represented by a sharp peak in the probability distribution func-
Eq. 33, while the late-time behavior is characteristic of the polar- tion, p(t), near the time of water breakthrough in the channel, t B* .
ity of the system, Eq. 36, in the case of an infinite reservoir or Hence, at least in the neighborhood of this time we should have
before the boundaries are felt, but it reduces to the 1D scaling, Eq. p 共 t 兲 ⬇d ␦ 共 t⫺t * 共40兲
B 兲,
18, in the case of a bounded reservoir. In the latter case, the
late-time scaling will be dominated by the 1D behavior provided where ␦ denotes the delta function and parameter d (0⬍d⬍1)
that the time is sufficiently large and that the mobility ratio re- expresses the fraction of the flow rate allocated to this channel
mains close to unity. 共and which is roughly equal to the ratio kh/ 兺 kh, where kh is the
permeability-thickness product of the channel兲. Substitution of
2D Displacements in Heterogeneous Systems: A Permeability Eq. 40 into Eq. 35 gives the following expression for W near
Streak. The next problem we considered involves model hetero- breakthrough:
geneous systems. We approximated channeling by considering a dW s 共 t/t B* 兲
high permeability streak between two wells in well-doublet geom- W⬇ . 共41兲
1⫹ 共 1⫺d 兲 W s 共 t/t B* 兲
Based on Eq. 41 we expect a sharp increase of W at breakthrough,
followed by a continuously decreasing slope as the asymptotic
plateau d/(1⫺d) is approached. The latter expresses the kh ratio
between the permeability streak and the rest of the formation. Fig.
12 shows numerical results for a bounded well doublet in a rect-
angular reservoir with a permeability streak 共with k 1 /k 2 ⫽100兲
joining the two wells. For comparison, the behavior of the same
system, but in the absence of heterogeneity, is also shown. We
note that Eq. 41 is obeyed well and for some time after break-
through it tends towards an asymptotic limit, which in this case
appears to be close to 5. The latter is a good estimate for the ratio
d/(1⫺d), if one also notes that in this simulation the permeability
channel has a relative thickness of 1/21. As time increases, other
streamlines break through, and eventually the relatively rapid rise,
characteristic of the behavior of a homogeneous system, sets in as
shown in Fig. 12. At sufficiently large times, the system behaves
as a homogeneous composite, where the previous scalings apply.
Fig. 11–Log-log plot of the WOR vs. time for a five spot and We also note that in the time period between t B* and the time of
M ⴝ1, 5, and 10 „ b ⴝ2.0…. The late-time slope is 2. breakthrough of the rest of the system, Eq. 41 can be rearranged to

Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999 419
Fig. 13–Log-log plot of the WOR vs. time for a bounded well
doublet with a permeability streak and for different cross flows: Fig. 14–Log-log plot of the WOR vs. time for a vertical cross
m y ⴝ0.001, 0.1, 10, and 100 „from top to bottom… „M ⴝ1, b section with 2, 5, and 10 noncommunicating layers „M ⴝ1, b
ⴝ3.0, m x ⴝ100…. The late-time slope is 1.5. ⴝ2.0….

d
1⫺d
⫺W⬃
d
共 1⫺d 兲 2
t ⫺a , 共42兲
W⫽
冋 冕
0

Wp

1⫹W p
k
km 冉 冊
t

k
km冉 冊
t
k f 共 k 兲 dk
册 冒冋冕 0
⬁ k f 共 k 兲 dk

1⫹W p 冉 k
km
t 冊册
.

共43兲
Some special cases are of interest: For a single layer, we have
which shows that a log-log plot of d/(1⫺d) ⫺W vs. t would be a f (k)⫽ ␦ (k⫺k m ), and Eq. 43 gives the single-layer expression
straight line with the 1D slope m⫽b/(b⫺1). This equation could W⫽W p (t) as expected. For the case of two layers of equal thick-
be used for diagnostic purposes. ness, denoted by subscripts 1 and 2, respectively, we have f (k)
The previous analysis was based on the absence of cross flow ⫽1/2关 ␦ (k⫺k 1 )⫹ ␦ (k⫺k 2 ) 兴 , which when substituted into Eq. 43

冋 册
between the streak and the rest of the reservoir. Consider next the leads to

冉 冊 冉 冊
case of cross flow. The parameters affecting the behavior of this
system are the two ratios, m x ⬅k x,s /k x,r and m y ⬅k y,s /k y,r , where k1 k2
k 1W p t k 2W p t
k is permeability, subscripts x and y refer to the streamwise and km km

冉 冊 冉 冊
transverse flow directions, respectively, and s and r refer to the ⫹
streak and the reservoir, respectively. Fig. 13 shows simulations k1 k2
in which the effect of cross flow is explored in the geometry of a 1⫹W p t 1⫹W p t

冋 冉 冊册
km km
well doublet in terms of the parameter m y that varies from 0.01 W⫽ . 共44兲
共negligible cross flow兲 to 100 共high cross flow兲 and with m x k1 k2

冉 冊
⫽100. The presence of the streak is evident in the case of small ⫹
k1 k1
cross flow, where the WOR increases fast and appears to stabilize 1⫹W p t 1⫹W p t
to a constant value. However, as the cross flow increases, the km km
response is mitigated, and at large values the system behavior
Of particular interest in the latter case is the behavior following
approaches that of a homogeneous system. We recall that for the
the breakthrough of the first layer 兵 W p 关 (k 1 /k m ) t 兴 ⬎0 其 and before
well doublet described here the slope of the WOR curve in a
the breakthrough of the second layer 兵 W p 关 (k 2 /k m ) t B2 兴 ⫽0 其 .
log-log plot is 1/3 at intermediate times and approaches b/(b

冋 册
Then, the above becomes
⫺1) at late times 共if the system is bounded兲.

冉 冊 冒
冉 冊 冋 册
k1
3D Displacements. To conclude our analysis we considered next k 1W pt
km k1

冉 冊
3D displacements in layered reservoirs in various patterns, such as W⫽ ⫹k 2 , 共45兲
a well doublet or a five spot. k1 k1
1⫹W p t 1⫹W p t
Consider first the case of noncommunicating layers. This is a km km
special case of the general layer problem but with M ⫽1 and in
the absence of fractional flow effects. We follow the previous which is essentially identical to Eq. 41 involving a heterogeneity
approach and combine the WOR behavior of each layer, to be streak. For sufficiently large times, W approaches k 1 /k 2 , which
denoted by W p , and which is influenced by the particular 2D flow for the case of equal thickness layers treated here represents the
pattern, with the heterogeneity of the layers. For this purpose, we kh ratio for the particular layered system. As in the heterogeneity
need the probability density function 共pdf兲, f (k), characterizing case, the quantity k 1 /k 2 ⫺W plotted vs. time in a log-log plot
the heterogeneity of the layered system. In such a system, the must display a power-law scaling 共with an exponent characteristic
breakthrough times of the various layers follow the statistics of of the associated areal displacement兲, namely,
f (k), with the breakthrough time being inversely proportional to k1
k. Therefore, the WOR of each layer can be expressed as a func- ⫺W⬃t ⫺␭ , 共46兲
k2
tion of its permeability, W p 关 (k/k m ) t 兴 , where, for simplicity, we
normalized time with the breakthrough time of the layer with where ␭ is the corresponding exponent.
permeability k m , which denotes the arithmetic mean. In the ab- Fig. 14 shows simulation results for a vertical cross section
sence of communication between the layers, the fraction of the with 2, 5 and 10 layers with permeability contrasts of 5:1,
flow partitioned to a layer is proportional to its kh value or, 5:4:3:2:1, and 10:9:8, . . . :1, respectively. The distinction between
equivalently, to k f (k)dk. Based on these remarks, we can derive layers is clear in the first case, as expected. As the number of
the following result for the overall WOR: layers increases, however, the distinction becomes less clear and

420 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999
Fig. 15–Log-log plot of the WOR vs. time for a well doublet with
a vertical cross section with 2, 5, and 10 communicating layers
„M ⴝ1, b ⴝ2.0….

the system approaches the behavior of a homogeneous, single


layer. The distinction becomes less pronounced as the permeabil-
ity contrast and the viscosity ratio M decrease. In all cases, the
straight line with slope b/(b⫺1) emerges as the final asymptote,
as expected, although in Fig. 14 it may take considerably large
values of W 共or t兲.
The behavior near breakthrough can be approximated as dis-
cussed previously. For example, for the two-layer case and assum-
ing that only one layer has broken through, the behavior following
breakthrough can be obtained from a combination of Eq. 45 and
the 1D analysis for the behavior following breakthrough to yield

1 1
⫺ ⫽F
t tB
冋 冉
W 1⫺
k2
k1
W

共 W⫹1 兲 2
冊 冉

W B 1⫺ W B
k2
k1
共 W B ⫹1 兲 2
冊册 , 共47兲

where F is a constant and subscript 1 denotes the largest perme-


ability layer. In the appropriate log-log plot, Eq. 47 indicates a Fig. 16–Log-log plot of the WOR vs. time for a well doublet with
straight line with slope 1. The validity of this scaling was tested a vertical cross section with infinitely many layers following a
numerically and found to be good. log-normal distribution in k : ␴ ⴝ„a… 0.001, „b… 0.01, and „c… 0.07
If cross flow is allowed between the various layers, the layered „M ⴝ1, b ⴝ2.0….
structure becomes less distinct. Fig. 15 shows the results of a
simulation similar to that in Fig. 14, except that now the trans-
verse permeability is not equal to zero but is set equal to the
theoretical result is more rapidly approached. The trends shown in
smallest streamwise permeability in the system. The overall effect
Figs. 16 and 17 should be useful in the analysis of multilayer
is to smoothen the transition between layers, and to eventually
systems.
approach the behavior of a single-homogeneous layer.
Multiple Layers. It would be interesting to generalize the re- A five spot. Finally, we examined the response of a quarter of a
sults for the noncommunicating layers obtained above to an arbi- five spot. Fig. 18 shows results for two-layer and five-layer sys-
trary distribution. To do so, we will make use of Eq. 43 and tems with permeability contrasts of 10:1 and 10:7:5:3:1, respec-
consider a log-normal distribution for f (k). Now the main param- tively, in the absence of cross flow. As in the case of the vertical
eters are M and the variance of the log-normal distribution, ␴. cross section and a line drive 共not shown here兲, the two-layer
Shown in Figs. 16 and 17 are analytical results obtained for dif- structure is quite apparent in the WOR response. Also clear, as in
ferent values of M and ␴. We observe the following. the case of a vertical cross section, is the structure of the five
For a fixed M , the WOR response starts earlier, as the variance layers. As anticipated, the responses in both cases approach that of
increases, and at large variance displays a tail, which is absent the homogeneous. Fig. 19 shows the effect of cross flow for the
from the single-layer problem. In fact, for the variance of Fig. same system. It is similar to the previous case of a vertical cross
17c, the WOR stays at a nonzero value plateau for a period of section 共Figs. 14 and 15兲.
time, even though the initial water saturation here is immobile.
This early tail reflects the high permeability streaks that exist in
the layered system at large values of the variance of the log- Interpretation Methodology
normal distribution, and which break through sufficiently early. The analysis presented above identified the various features of the
The implication is that WOR values other than zero are possible response of the WOR vs. time for a variety of well geometries and
for systems with immobile S wi , provided that the heterogeneity reservoir characteristics in the case of a waterflood. The dominant
共or the Dykstra–Parsons兲 coefficient is sufficiently large. Eventu- features of the response depend on the time regime of interest.
ally, however, the response rises asymptotically to the same curve Our analysis has shown that we can distinguish four such regimes:
of the single layer with the theoretically expected power law. The 共i兲 a regime before any significant water breakthrough; 共ii兲 a re-
effect of M is also shown in Figs. 16 and 17. Compared to M gime immediately following breakthrough; 共iii兲 an intermediate
⫽1 共Fig. 16兲, the results for M ⫽5 共Fig. 17兲 have similarities in regime, in the case of a high-permeability streak or a two-layer
the existence of early breakthroughs and of a plateau region of system with high-permeability contrast; and 共iv兲 a late-time re-
nonvanishing WOR, but also some differences, in that the re- gime. A schematic of the response is shown in Fig. 20 for the two
sponse of the system is more spread out and that the asymptotic cases 共with or without a permeability streak兲, respectively.

Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999 421
Fig. 19–Log-log plot of the WOR vs. time for a five spot with
two and five communicating layers „M ⴝ1, b ⴝ3.0….

Regime 共ii兲 describes the behavior following breakthrough.


Conditions for the validity of the relative permeability, Eq. 6, may
be applicable in this regime. The interpretation is not in terms of
a straight line in a log W-log t plot. Use of the latter would result
in slopes close to 1, which cannot be interpreted as equal to the
exponent of the relative permeability dependence. Hence, such an
approach should be done with appropriate caution. When a
multilayer system is involved, the early time behavior may also
involve a smoother shape that reflects the heterogeneity of the
system in the case of unfavorable M , or a more abrupt increase, in
the case of piston-like displacement. In this regime, a straight line
in a log-log plot interpretation is suggested.
Regime 共iii兲 is intermediate between 共ii兲 and the final
asymptotic regime. When a high-permeability streak is not in-
volved, this regime primarily reflects patterns that are unbounded
or that behave as such before the effect of boundary is felt, such as
the various multipole flows discussed earlier. Here, we have the
behavior,
W⬃t N/ 共 N⫹2 兲 , 共48兲

Fig. 17–Log-log plot of the WOR vs. time for a well doublet with
a vertical cross section with infinitely many layers following a
log-normal distribution in k : ␴ ⴝ„a… 0.001, „b… 0.01, and „c… 0.07
„M ⴝ5, b ⴝ2.0….

In regime 共i兲 there may be non-negligible WOR values, de-


pending on whether or not S wi is mobile and/or there is substantial
layering. Our results in Figs. 16 and 17 have demonstrated the
validity of the latter in the case of large heterogeneity. Here, the
WOR increases rather slowly and almost appears to be in a
plateau-like region.

Fig. 20–Schematic response of the WOR vs. time for a general


Fig. 18–Log-log plot of the WOR vs. time for a five spot with multilayer system in the „a… absence and „b… presence of a
two and five noncommunicating layers „M ⴝ1, b ⴝ3.0…. high-permeability streak.

422 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999
which indicates a straight line with slope N/(N⫹2), namely, a M ⫽ viscosity ratio, dimensionless
slope always less than 1, where N is characteristic of the well m ⫽ exponent, dimensionless
pattern 共polarity兲. The corresponding scaling between W and Q o N ⫽ pattern polarity index, dimensionless
is also a power law, p ⫽ probability of arrival times, dimensionless
Q ⫽ cumulative production, dimensionless
W⬃Q N/2
o . 共49兲 q ⫽ injection rate, L3T⫺1
When, however, a high-permeability streak is involved or the sys- S ⫽ saturation, dimensionless
tem consists of only a few discrete layers with a sharp permeabil- t ⫽ injection time, dimensionless
ity contrast, this regime mostly reflects the partition of the flow in v ⫽ volumetric injection rate, L3T⫺1
these two layers. Then, the equations pertaining to post- W ⫽ water/oil ratio, dimensionless
breakthrough and intermediate behavior, respectively, should be x ⫽ distance, dimensionless
used. ␣ ⫽ angle, dimensionless
Regime 共iv兲 is the final asymptotic state in all bounded pat- ⌫ ⫽ gamma function, dimensionless
terns, and its behavior is ␬ ⫽ relative permeability ratio, dimensionless
␭ ⫽ exponent, dimensionless
b ␮ ⫽ exponent, dimensionless
log W⬃ log t⫹H, 共50兲
b⫺1 ␾ ⫽ porosity, dimensionless
indicating that a plot of log W-log t is asymptotically a straight ␺ ⫽ streamfunction, dimensionless
line with slope b/(b⫺1). This slope provides information on the ␶ ⫽ time, dimensionless
exponent b of the power law that describes the relative oil perme-
ability dependence on saturation near the residual saturation, and Subscripts
it is always greater than 1. Alternatively, the WOR-oil recovery B ⫽ breakthrough
ratio is expressed by o ⫽ oil
ln W⬇⫺b ln共 1⫺Q o 兲 ⫹H ⬘ , 共51兲 or ⫽ residual oil
r ⫽ relative
or its further approximation at small Q o , s ⫽ 1D streamtube
ln W⬇bQ o ⫹H ⬘ . 共52兲 w ⫽ water
A semilog relation of the type of Eq. 52 was found to describe
many of the simulations of Lo et al.5 well. Acknowledgments
This work was partly supported by Dowell–Schlumberger. The
Conclusions research of one of the authors 共Y.C.Y.兲 was also supported by
In this article we reported on the progress made in the identifica- DOE Contract No. DE-FG22-96BC14994/SUB. Both these con-
tion of the relationship between the WOR and time during water tributions are gratefully acknowledged. We would also like to
production. This relationship contains two effects, one due to the thank K. Chan and M. Mack of Dowell–Schlumberger and Prof.
relative permeability and mobility and another due to the produc- Iraj Ershaghi of USC for useful discussions.
tion geometry. The first effect was uncovered by conducting a 1D
analysis, in which it was found that the late-time behavior of the References
WOR vs. time in a log-log plot is a straight line of slope b/(b 1. Craig, F.F. Jr.: Reservoir Engineering Aspects of Waterflooding,
⫺1), where b is the exponent in the dependence of relative oil Monograph Series, SPE, Richardson, Texas 共1971兲 3.
permeability on saturation. This information reflects petrophysical 2. Willhite, G.P.: Waterflooding, SPE Textbook Series, Richardson,
aspects of the formation. The geometrical effect was explored by Texas 共1986兲.
analyzing various patterns. The approach used decouples global 3. Ershaghi, I. and Omoregie, O.: ‘‘A Method for Extrapolation of Cut
flow patterns and fractional flow effects. It is essentially one of vs. Recovery Curves,’’ JPT 共1978兲 203.
noncommunicating streamtubes or layers, and is expected to be 4. Ershaghi, I. and Abdassah, D.: ‘‘A Prediction Technique for Immis-
cible Processes Using Field Performance Data,’’ JPT 共1984兲 664.
valid for values of the mobility ratio that are close to unity. Frac- 5. Lo, K.K., Warner, H.R. Jr., and Johnson, J.B.: ‘‘A Study of the Post-
tional flow effects in the displacement in individual streamtubes Breakthrough Characteristics of Waterfloods,’’ paper SPE 20064 pre-
were accounted for using Buckley–Leverett theory. Our approach sented at the 1990 SPE California Regional Meeting, Ventura, Cali-
predicts a variety of power-law scalings for different time re- fornia, 4–6 April.
gimes, from breakthrough to late-time behavior, which confirm 6. Chan, K.S.: ‘‘Water Control Diagnostic Plots,’’ paper SPE 30775
the potential of the WOR-time plot as a diagnostic tool. presented at the 1995 SPE Annual Technical Conference and Exhibi-
tion, Dallas, 22–25 October.
Nomenclature 7. Wilkinson, D.: ‘‘Percolation Effects in Immiscible Displacement,’’
Phys. Rev. A 共1986兲 34, 1380.
A ⫽ constant of the X-plot method, dimensionless 8. Bear, J.: Dynamics of Fluids in Porous Media, Elsevier, New York
a ⫽ constant of the power-law method, dimensionless City 共1972兲.
B ⫽ exponent of the X-plot method, dimensionless 9. Koplik, J., Redner, S., and Hinch, E.J.: ‘‘Tracer Dispersion in Planar
b ⫽ exponent of the power-law method, dimensionless Multipole Flows,’’ Phys. Rev. E 共1994兲 50, 4650.
C ⫽ constant, dimensionless 10. Koplik, J., Redner, S., and Hinch, E.J.: ‘‘Universal and Nonuniversal
C⬘ ⫽ constant, dimensionless First-Passage Properties of Planar Multipole Flows,’’ Phys. Rev. Lett.
共1995兲 74, 82.
D ⫽ constant, dimensionless 11. Carrier, G.F., Krook, M., and Pearson, C.E.: Functions of a Complex
d ⫽ fraction of flow, dimensionless Variable, McGraw-Hill, New York City 共1966兲.
E ⫽ constant, dimensionless
F ⫽ constant, dimensionless
f ⫽ permeability probability density function, L⫺2 Appendix A—Relationship Between Water/Oil Ratio
fw ⫽ fractional flow function, dimensionless and Cumulative Oil Recovery
G ⫽ exponent, dimensionless To obtain the relationship between W and Q o , we first note that
H ⫽ constant, dimensionless dQ w
H⬘ ⫽ constant, dimensionless ⫽W, 共A-1兲
h ⫽ reservoir thickness, L dQ o
k ⫽ permeability, L2 where Q w denotes the cumulative amount of water produced.
L ⫽ well distance, L Since we also have t⫽Q o ⫹Q w , Eq. A-1 gives

Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999 423
Q o⫽ 冕 0
tdt
W⫹1
. 共A-2兲 冕 t*
t

B
exp共 ⫺G ␶ 兲
1⫹D 共 t/ ␶ 兲 m
d␶ ⫽t 冋冕 0
1 x m exp共 ⫺Gtx 兲
D⫹x m
dx

Thus, a knowledge of the W⫺t relationship leads to expressions


for Q o and Q w .
Assuming that W is relatively large and that W⬇t ␮ , we can
⫺ 冕0
t*
B /t x m exp共 ⫺Gtx 兲
D⫹x m 册
dx . 共B-3兲

obtain the following results.


䊉 If ␮ ⬍1, then Q
o increases without bounds as t increases.
We now apply Watson’s lemma11 to evaluate the first integral,
This problem essentially corresponds to an unbounded reservoir. which to leading order becomes equal to 关 ⌫(m⫹1)/D m⫹1 兴 t ⫺m ,
Then Eq. A-2 gives a power law, where ⌫(z) denotes the Gamma function. The second integral is
t*
␮ also readily evaluated to give ( 兰 0B ␶ m e⫺G ␶ d␶ )t ⫺m , which is of the
ln W⬃ ln Q o . 共A-3兲 same order as the first integral. Combined, the two integrals decay
1⫺ ␮
䊉 If, however, ␮ ⬎1, then Q
following the power law t ⫺m , which is slower than the exponen-
o approaches a limit 共equal to 1兲 as
t increases. This problem corresponds to a bounded reservoir. tial associated with p. Hence, the power law would dominate the
Then we can manipulate Eq. A-2 to give behavior of the denominator of Eq. 31, leading to the scaling,

Q o ⫽1⫺ 冕
4
⬁ dt
W⫹1
, 共A-4兲 W⬃
DG m⫹1
⌫ 共 m⫹1,Gt B* 兲
t m, 共B-4兲

hence,
where ⌫(a,z) is the incomplete gamma function. The above is the
␮ scaling claimed in the text.
ln W⬃⫺ ln共 1⫺Q o 兲 . 共A-5兲
␮ ⫺1
Yanis C. Yortsos is Professor of Chemical Engineering and
Appendix B—Asymptotic Behavior of Water/Oil Ratio Chester Doley Professor of Petroleum Engineering at the U. of
for 2D Patterns Southern California (USC) in Los Angeles. e-mail:
To derive the asymptotic behavior of W for general 2D patterns yortsos@euclid.usc.edu. He also served as Chairman of the
which include fractional flow effects, we need to consider the Chemical Engineering Dept. during 1991–97. His research inter-
ests are in the areas of fluid flow, transport, and displacement
behavior at large t of the two terms on the right-hand side of Eq.
processes in porous media. Yortsos holds a BS degree from
31. For simplicity, we will assume that the 1D WOR has the Natl. Technical U. of Athens and MS and PhD degrees from
simple behavior, W s ( ␶ )⫽D ␶ m , where m⭓1. We shall examine California Inst. of Technology, all in chemical engineering. A
separately the two different cases corresponding to the different member of the Editorial Review Committee, he served as a
behavior of p 共unbounded vs. bounded patterns兲. member of the 1990–91 and 1996–97 Western Regional Meet-
Consider first an unbounded pattern, in which p(t) has the ing Program Committees and on Annual Meeting Technical
behavior described by Eq. 34, namely, Committees. Youngmin Choi is an R&D supervisor with Jason
Natural Products in Culver City, California. e-mail:
2N⫹2 yochoi@aol.com. He holds a BS degree from A-Joo U., South
p⫽Et ⫺␭ ; ␭⫽ ⬎1. 共B-1兲 Korea, and a MS degree from USC, both in chemical engi-
N⫹2
neering. Zhengming Yang is a software engineer with MSC
Now the second term in the denominator of Eq. 31 reads 1/(␭ Software Corp. in Costa Mesa, California. He previously
⫺1) t ⫺(␭⫺1) , while the first term can be rearranged to read worked as an engineer with China Natl. Oil Co. during 1985–91
and as a research associate at USC and Tidelands Oil Produc-

冕 t E ␶ ⫺␭
t * 1⫹D 共 t/ ␶ 兲
B
m d␶ ⫽Et
1⫺␭
冕 1 x m⫺␭
t * /t D⫹x
B
m dx. 共B-2兲
tion Co. during 1995–99. Yang holds a BS degree in chemical
engineering from Tianjin U., China; a MS degree in petroleum
engineering from RIPED Graduate School, Beijing; and a PhD
degree in petroleum engineering from USC. Piyush C. Shah is
The behavior of this integral in the limit tⰇt B* depends on the a senior engineering specialist with Landmark Graphics Corp.
magnitude of m⫺␭. If m⫺␭⬎⫺1, the integral converges, and in Houston, where he works on development of wellbore-flow
the right-hand side above scales as t ⫺(␭⫺1) , which is of the same simulation models. e-mail: pcshah@lgc.com. After postdoc-
order as the second term in the denominator of Eq. 31. However, toral work at USC, Shah held R&D positions with major oil com-
since m⭓1 and ␭⫺1⭐1, the above inequality is always satisfied. panies and also worked for Schlumberger Wireline and Dow-
Therefore, it follows that in this case we shall have W ell. His work has involved reservoir process simulation,
developing interpretation models for well and formation test-
⬃t N/ ( N⫹2 ) , as claimed in the text. ing, and conformance control. He holds a BTech degree from
For a bounded pattern, p follows exponential decay 共Eq. 35兲. Indian Inst. of Technology, Bombay, and MS and PhD degrees
Then the second term above also behaves exponentially and scales in aeronautical engineering from California Inst. of Technol-
as exp(⫺Gt). To calculate the first term, we rearrange the integral ogy. He served on the Editorial Review Committee during
as follows: 1986–89.

424 Yortsos et al.: Water/Oil Ratio in Waterfloods SPE Journal, Vol. 4, No. 4, December 1999

S-ar putea să vă placă și