Академический Документы
Профессиональный Документы
Культура Документы
冉 冊 冉 冊
value can be used to estimate the quality of the estimate.
⭸⌽ 1 K
Numerical tests of the method show that it estimates average = ⵜ⭈ ⭈ ⵜ⌽ , . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
reservoir pressure accurately, even when the reservoir is hetero- ⭸t ct
geneous or when partial-flow barriers are present. Examples
which is assumed to hold in a domain ⍀. A no-flow Neumann
with real data show that the behavior predicted by the theory is
boundary condition given by:
actually observed.
We expect the method to have value in reservoir limits testing, K
in making consistent estimates of average reservoir pressure ⭈ ⵜ⌽ ⭈ n = 0, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
from permanent downhole gauges, and in characterizing com-
plex reservoirs. which is assumed to hold at the boundary ⭸⍀. In these equations,
it is assumed that neither the permeability tensor K nor the other
Introduction properties are homogeneous. The general solution to Eq. 1 with
Several different methods of interpreting pressure-buildup data to boundary condition Eq. 2 under the assumption that the perme-
obtain average reservoir pressure have been proposed (Muskat ability tensor is symmetric positive definite is given by the expres-
1937; Horner 1967; Miller et al. 1950; Matthews et al. 1954; Dietz sion (Showalter 1994):
1965) in the past, and in recent years some new techniques have ⬁
appeared in the literature (Mead 1981; Hasan and Kabir 1983;
Kabir and Hasan 1996; Kuchuk 1999; Chacon et al. 2004). Larson
⌽共t, x, y, z兲 = ⌽0 + 兺 c 共x, y, z兲e
i=1
0
i i
−it
. . . . . . . . . . . . . . . . . . . (3)
(1963) revisited the Muskat method and put it on a firm theoretical
ground for a homogeneous cylindrical reservoir. Some of the ex- The functions i are the eigenfunctions of the differential op-
isting techniques depend on knowledge of the reservoir size and erator on the right side of Eq. 1 with the Neumann boundary
shape and assume homogeneous properties (Horner 1967; Miller condition of Eq. 2 in the domain ⍀, and the numbers i are the
et al. 1950; Matthews et al. 1954; Dietz 1965). Such methods may corresponding eigenvalues. It is known that the eigenvalues are all
result in uncertain predictions when reservoir data are unavailable positive, except for the first, 0, which is zero, and that they in-
or reservoir heterogeneity exists. The inverse time plot by Kuchuk crease without bound (Showalter 1994). Both Vega and Watten-
(1999) is essentially a modification of Horner’s method (1967) and barger (2000) and Coats et al. (1964) derived similar expressions
works well in reservoirs that can be treated as infinite during the for the general transient behavior of aquifers with heterogeneous
time of the test. The hyperbola method proposed by Mead (1981) permeability distributions. Gavalas and Seinfeld (1973) derived a
and further developed by Hasan and Kabir (1983) is an empirical similar expression for transient pressure behavior in heterogeneous
technique, not based on fundamental fluid flow principles for reservoirs of arbitrary shape. The coefficients c0i depend on the
bounded reservoirs (Kabir and Hasan 1996). Chacon et al. (2004) pressure distribution at the start of the buildup, but the eigenvalues
develop the direct synthesis technique, in which conventional do not. Hence, the eigenvalues are independent of the past pro-
theory is used to derive an average pressure directly from standard duction history of the reservoir.
log-log plots. Homogeneous properties and radial symmetry are If we consider the transient behavior of the pressure at a single
assumed. Muskat’s original derivation was a wellbore storage point, then we may combine the eigenfunction values with the
model. Larson reinterpreted Muskat’s method and derived rela- coefficient values to write Eq. 3 as:
tionships showing how Muskat’s plot could be used to estimate ⬁
⌽共t兲 = ⌽0 − 兺 ce
i=1
i
−it
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4)
⌽共t兲 ≈ ⌽0 − c1e−1t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5) Since 2>1, the transient part of the extrapolated pressure decays
more rapidly than the pressure transient itself. Hence, it can be
The Muskat plot is a plot of ln(⌽0−⌽) vs. t (Muskat 1937). Eq. 5 used to estimate the final pressure. Examples of this will be given
shows that this will be a straight line, the slope of which is –1 for in the next section.
large values of t. Unfortunately ⌽0 is unknown, being the desired
average reservoir pressure. Muskat proposed that an iterative Numerical Examples
scheme be used to find the value of ⌽0 that resulted in the This section presents some simple numerical examples of long
“straightest” line for large values of time. However, this method pressure buildups in which the actual average reservoir pressures
was difficult to apply in practice. It can be simplified by computing are known to illustrate the use of the techniques previously de-
the derivative of Eq. 4: scribed and to compare the results with some other techniques.
⬁ Table 1 lists properties for a rectangular homogeneous reser-
⭸⌽
⭸t
= 兺 ce
i=1
i i
−it
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6) voir with a well in the center.
The well is produced at 20,000 BOPD for 5 days and then shut
Eq. 6 shows that a plot of the pressure derivative vs. ln(t) will be in for 30 days for a buildup. Figs. 1 through 4 show plots of the
a straight line for large t with slope of –1. The intercept of that natural logarithm of the pressure derivative and first three residuals
line will be ln(1c1), so the first term in the expansion of Eq. 4 can against buildup time.
be determined from the data. The residual R1(t) is defined by: The derivatives were computed from a central difference of a
running 3-point average of the pressure. This smoothing was suf-
⭸⌽ ficient in the numerical cases, since the noise level in the computed
R1共t兲 = − 1c1e−1t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
⭸t pressures is low.
In each case, the late-time data form a straight line, enabling the
Eq. 7 along with Eq. 6 show that for large values of t, a plot of extraction of the characteristic decay exponents from the buildup.
R1(t) vs. ln(t) is a straight line of slope –2 with intercept ln(2c2). Note that extraction of slopes from the residual data is often more
In a similar fashion, higher residuals can be defined recursively by difficult because of noise in the data. Part of the reason for the
Rn = Rn−1 − ncne−nt. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8) noise is that these components of the pressure transient decay
much more rapidly than the principal, so rounding and numerical
In this fashion, the exponents and coefficients can be successively errors are relatively larger. Hence, error in the higher exponents is
extracted from the pressure derivative. These values may then be likely to be greater than that of the lowest. Because of this, there
used to extrapolate the pressure. The final pressure ⌽0 is obtained is a limit to the number of higher exponents that can be reliably
by computing the extrapolated pressure ⌽ext from the formula: extracted from the data. Fig. 5 shows a plot of the buildup pressure
Fig. 2—Logarithm of first residual against time. Fig. 3—Logarithm of second residual against time.
and the extrapolated pressure defined by Eq. 9. The figure shows better in this case than inappropriate extrapolation. The exponen-
that the extrapolated pressure rises to a constant value more tial extrapolation proposed here gives very good results, and the
quickly than pressure. other methods are less accurate.
Table 2 lists exponents and coefficients extracted from the The next example is a similar model in which the reservoir has
buildup data. By use of the data from this table, the individual several partially communicating compartments. Fig. 10 shows the
exponential terms in the pressure-derivative expansion may be reservoir with the locations of the partial barriers. These barriers
computed to show that over the time interval for which the linear were modeled by multiplying the grid transmissibilities along the
fit was done, only the first term is significant. barriers by 0.001.
Figs. 6 through 9 show various plots of the pressure data along The properties of this reservoir are similar to those of the ho-
with extrapolations of the data sometimes used to estimate average mogeneous model. In this case, the buildup lasts 95 days. Fig. 11
reservoir pressure. shows a plot of the logarithm of the shut-in pressure derivative
Figs. 6 and 7 show Horner plots (Horner 1967) of the data against buildup time. As before, the latter part of the curve is a
extrapolated from either early-time or late-time data. Fig. 8 shows straight line.
a plot of pressure against 1/t with an extrapolation of the late- Fig. 12 shows the buildup pressure along with the extrapolated
time data. This is loosely based on Kuchuk’s method (Kuchuk pressure from Eq. 9. Note that the extrapolated pressure in this
1999). Fig. 9 is a plot of the pressure along with a hyperbolic fit of case is considerably higher than the well pressure even at the end
the form: of the buildup period. Table 4 lists the decay parameters for this
c buildup. Note that in this case, the first exponent is considerably
P=a+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11) smaller in magnitude than the second. This seems to be a charac-
b+t teristic of compartmentalized reservoirs. Table 5 lists the extrap-
This form is based on the method of Hasan and Kabir (1983; Kabir olated average pressure estimates as before. In this case, all of the
and Hasan 1996). The value of the parameter a gives the pressure methods underpredict the average reservoir pressure, and with the
at infinite buildup time and is equal to the average reservoir pres- exception of the method of this paper, the errors are large. The
sure as estimated by the method. hyperbolic method is unable to adequately extrapolate the data,
Nonlinear regression was used to fit the data. The regression even though it achieves a reasonable fit of the data, as can be seen
was done in two ways: fitting data over a time interval from 10 to in Fig. 13. Tests of nonlinear regression using a weighted sum of
30 days of buildup and more than 25 to 30 days of buildup. Both deviations between measured and model pressures and pressure
gave similar results. Note in Fig. 9 that the model fits the data very derivative values as the objective function failed to improve the
well in the interval for which buildup data are available, but the performance of the hyperbolic model in this case.
model curve continues to increase above the simulated final res- The final numerical model example is a simulation that is char-
ervoir pressure. acteristic of heterogeneous channelized deepwater reservoirs. In
Table 3 lists the results of the various extrapolations along with this simulation, the well pressure is still changing 1 year after
the actual final average reservoir pressure. The latter was deter- shut-in. Although this is extreme, recent examples after Hurricane
mined by computing the pore-volume-weighted average pressure Katrina validate that long, slow buildups of pressure are common.
in the simulation model. As expected in this simple model, the In this synthetic example, the well flowed for 10 days, was shut in
average reservoir pressure is nearly constant. The third column for a year, and then flowed for a year followed by a second year-
lists the error in the final-pressure estimate relative to the total
pressure depletion of the reservoir following the production. This
would be a relevant measure for material-balance calculations.
Note that the P* from the Horner plot gives the worst results, not
surprisingly, since it is an appropriate estimate only for an infinite
reservoir. Surprisingly, the next-best estimate is the final buildup
pressure. That is, the use of no extrapolation at all seems to work
Fig. 14—Log-log diagnostic plots second buildup for realistic Fig. 15—Comparison of average pressure extrapolation tech-
numerical example (a); plot of natural logarithm of pressure niques for Model D (a); material balance errors for average pres-
derivative of two buildups for realistic numerical example (b). sure extrapolation techniques for Model D (b).