Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Summary. A well-test interpretation method based on the analysis of the time rate of pressure change an? the. actual pressur~ response
is discussed. A differentiation algorithm is proposed, and several field examples illustrate how the method slmpitfies the analysIs process,
making interpretation of well tests easier and more accurate.
Introduction
The interpretation of pressure data recorded during a well test has ogeneous formations reveals the good definition obtained with
been used for many years to evaluate reservoir characteristics. Static derivative plots, and the distinction between currently used interpre-
reservoir pressure, measured in shut-in wells, is used to predict tation models is clearly shown.
, reserves in place through material-balance calculations. Transient-
pressure analysis provides a description of the reservoir flowing TranslentPressure Analysis Applied
behavior. Many methods have been proposed for interpretation of to Homogeneous Reservoirs
transient tests, 1 but the best known and most widely used is Horn- Conventional well-test interpretation has focused on the homogene-
er's.2 More recently, type curves, which indicate the pressure ous reservoir solution. The corresponding pressure-analysis methods
response of flowing wells under a variety of well and reservoir con- have been discussed extensively in the literature and are commonly
figurations, were introduced. 3-8 Comparison of transient-pressure used.
measurements with type curves provides the only reliable means Two complementary approaches are used for transient-pressure
for identifying that portion of the pressure data that can be analyzed analysis: (1) a global approach is used to diagnose the pressure be-
by conventional straight-line analysis methods. havior and to identify the various characteristic flow regimes, and
Recently, the quality of well-test interpretations has improved (2) specialized analyses, valid only for specific flow regimes, are
considerably because of the availability of accurate pressure data performed on selected portions of the pressure data. Results of anal-
(from electronic pressure gauges) and the development of new soft- yses with both approaches must be consistent.
ware for computer-aided analysis. An increasing number of theo- Diagnosis of pressure behavior is performed by type-curve analy-
retical interpretation models that allow a more detailed definition sis. Fig. 1 describes a well with wellbore storage and skin in a reser-
of the flow behavior in the producing formation are now in use. voir with homogeneous behavior. 5 Dimensionless pressure, PD'
Surprisingly, the commonly used analysis techniques have not is plotted on log-log scale vs. dimensionless time group, IDICD.
followed the general progress evident in hardware and in interpre- The resultant curves, characterized by the dimensionless group
tation models, making the interpretation procedure complicated and CD e 2S (Appendix A of Ref. 9), correspond to well conditions
time-consuming. Type curves are seen by various analysts as overly ranging from damaged wells to acidized and fractured wells.
simplistic or overly complex, difficult to distinguish, andlor cumber- Two flow regimes of interest can be identified in the pressure
some to use. Yet, mere identification of straight lines on a pressure- response (Fig. I). At early time, all the curves merge to an asymp-
vs.-time graph is a "ruler approach" -convenient for hand analysis tote of slope equal to unity, corresponding to pure wellbore-storage
but ignoring powerful computing facilities that are available. Fur- effect given by
thermore, the conventional straight-line analysis methods fail to use
PD =tDICD ...................................... (1)
all the data available and can result in significant errors.
We propose an interpretation method based on the analysis of Later, when all storage effect is over, the constant sandface flow
the derivative of pressure with respect to the appropriate time rate is established, and the reSUlting pressure behavior produces
function-natural logarithm of time or Homer/superposition time the usual straight line on a semilog plot:
functions. This method considers the response as a whole, from PD =0.5[ln(tDI CD ) +0.80907 + In CDe 2S ] ... (2)
very-early-time data to the last recorded point, and uses the type- This regime, called infinite-acting radial flow, does not show a char-
curve-matching technique. It provides a description of the flow be- acteristic shape on log-log scale. The locus "approximate start of
havior in the reservoir, but with the logarithmic derivative, it also the semilog straight line" therefore has been marked on the type
emphasizes the infinite radial flow regime, of prime interest in well- curve of Fig. 1. The interpretation procedure with this type curve
test interpretation. The approach is an extension of the Homer is illustrated with a 30-hour buildup (Table 1), whose detailed inter-
method to analyze the global response with improved definition. pretation was presented in Ref. 10.
Use of the derivative of pressure vs. time is mathematically satis- The first step is to plot the buildup pressure difference, p(~t) -
fying because the derivative is directly represented in one term of p(~t=O), vs. the elapsed time, ~t, since the well was closed (Fig.
the diffusivity equation, which is the governing equation for the 2). This plot is then compared with the type curves: the long unit
models of transient-pressure behavior used in well-test analysis. slope straight line at early times, indicative of wellbore storage ef-
Thus, the derivative response is more sensitive to small phenomena fect, is matched on the early-time asymptote of the type curves.
of interest that are integrated and hence diminished by the pressure- By moving along this 45 line, the best curve match is attempted.
vs.-time solutions. In this case, all curves above CD e 2S = 10 8 , in the damaged well
One limitation of the pressure derivative in analysis is the difficulty area, match the data equally well. The possible matches also show
in collecting differentiable pressure-transient data. Accurate and that the limit "approximate start of the semilog straight line" has
frequent pressure measurements are required. However, pressure been attained after about 23 hours of shut-in.
measurement and the computer processing technologies now avail- A semilog analysis is then performed on the last 7 hours of build-
able at wellsites allow pressure-derivative analysis. up; the pressure is plotted with respect to the logarithm of Homer
The pressure-derivative method is demonstrated for a homogene- time (Fig. 3). A straight line develops at the end of the plot and
ous reservoir and compared with conventional interpretation tech- is used in the conventional way to estimate khlp. (from the slope),
niques. The practical aspects of differentiation of actual pressure p* (from extrapolated pressure to infinite shut-in time), and S (from
data are discussed. Application of the derivative analysis to heter- the straight-line displacement at 1 hour).
The permeability group khlp. being fixed, the pressure match is
Now with Kappa Engineering .
.. Now a consultant.
known and it is possible to adjust the type-curve match. The final
match is made on CDe 2S = 4 X 10 9 . Results of the analysis are
Copyright t989 Society of Petroleum Engineers given in Appendix A.
iiiii"'
",,,
h, ft 107
cf> 0.25
,.,.,.
D'
10 11-, cp 2.5
rw' ft 0.29
",.,
D"
3750
~
102 f \ ...
..
......
'
i 3SOO
\ ..
...
'.
10
3250
.............. . .
10"' 10 102
3000
, 10 10' 10' 1C)4
10"'
(lp.. dt)/dt
4'
Fig. 2-A diagnostic tool: log-log plot of buildup data. Fig. 3-Horner plot.
take an infinite shut-in time to reach a IBU of the same amplitude are generated by taking the derivative of the pressure with respect
as the pressure drop at the end of the drawdown IDdCtp ). As a to the natural logarithm of time.
result, drawdown and buildup curves are not identical. In Fig. 5,
dpD/[d In(tDICD)) = (tDICD){ dpD/[d(tDICD)]} = (tDICD)pb
the dotted line corresponds to a drawdown type curve. After the
well is shut in at tp ' the resulting buildup response (thick line) devi- ..................................... (4)
ates from the drawdown type curve and flattens toward the same
The first typical regime observed on the type curve of Fig. 1 is
level as the last drawdown-pressure change before shut-in,
wellbore-storage effect. By combining Eqs. 1 and 4, we obtain
IDdCtp ). This deviation is more pronounced when the flow time
before shut-in is relatively short, as is often the case when explo- CtDICD)pb=tDICD. . .............................. (5)
ration wells are tested.
As for pressure, all the derivative behaviors are identical at early
In practice, it is not possible to ascertain a perfectly constant flow
time, and the curves merge on a single asymptote of slope equal
rate during drawdown, especiaJIy during the initial instants of flow.
to unity.
Log-log analysis considers the global response during a flow period
When the infinite-acting radial flow regime has been reached-
and therefore does not accommodate any rate variation during the
i.e., after the limit "approximate start of the semilog straight line"
period analyzed. As a result, only buildups, recorded on shut-in
(Fig. I)-the pressure behavior is described by Eq. 2. The semilog
wells, generally are suitable for type-curve matching. When the
slope is constant:
interpretation is performed on computer, the buildup type curve
is generated for the actual flow history before shut-in (multirate CtDICD)pb=O.5, .................................. (6)
curves):
and all the derivative curves merge to a second asymptote, the one-
...........
draw~
build-up
tp TIME LOG At
Fig. 4-Pressure history of a simple drawdown/bulldup test. Fig. 5-Drawdown and buildup type curves.
c"o.
Fig. 6-Derlvatlve type curve for homogeneous reservoir. 10 Fig. 7-Match of the derivative of actual data.
curve following the data at intennediate time between the two asymptotic regimes. This property of the derivative presents, in
asymptotes. some cases, interesting features for the interpreter. For example,
For buildup analysis, the same curves were found to be applicable when a well is tested before and after stimulation, if the well treat-
(Ref. 10) provided that the derivative is taken, not with respect to ment has not modified the characteristics of the producing zones,
natural logarithm of time, but with respect to natural logarithm of derivative behaviors recorded during both tests should match ex-
the Homer time, as modified by Agarwal 12 : actly when the data curves are free of any wellbore-storage effect.
dp/{d In[tpAt/(tp +At)]} =At[(tp + At)/tp](dp/dt). . ...... (7) The limited influence of the skin coefficient on derivative responses
will be of interest for the analysis of heterogeneous fonnations and
These buildup derivative responses, when plotted vs. the actual shut-
for the identification of boundary effects. As discussed later, the
in time, At, match on the type curve of Fig. 6. This behavior is
traditional flow regimes produce a characteristic shape much faster
present when the Homer method is valid-i.e., the drawdown has
than on the usual pressure curves.
to reach radial flow before shut-in.
Fig. 7 presents the slope of the example Homer plot (Fig. 3), Other applications of the derivative of pressure have been pro-
posed for observation wells 13 and fractured wells 14 that use the
plotted on a log-log scale vs. shut-in time. The matching procedure
on the type curve of Fig. 6 is as follows. derivative of pressure with respect to elapsed time. In the method
I. The constant-derivative part of the data plot is placed on the presented here, it is preferable to consider the derivative as the serni-
one-half straight line of the type curve. The pressure match is then log (or Homer/superposition) slope for the following reasons:
fixed accurately and kh/p. is known. 1. The sernilog derivative emphasizes the infinite-acting radial
2. The data plot is displaced along the one-half straight line until flow regime of prime interest in well-test interpretation.
wellbore-storage data match on the early-time unit-slope asymptote 2. When the derivative is considered as the slope of the semilog
of the type curve. The time match is now fixed, yielding the well- or the superposition plot, both the pressure change and the pressure
bore-storage constant, C. derivative are made dimensionless by use of the same group (kh/
3. A direct reading tells which CD e 2S curve provides the best 141.2qBp. in usual oilfield units), making the double match practical.
match between the two asymptotes, giving access to the skin fac- 3. The derivative with respect to the Homer/superposition func-
tor, S. tion converts buildup analysis to that of drawdown, simplifying the
Experience has shown that for practical reasons discussed later analysis process.
with the differentiation of actual data, it is convenient to match both 4. Buildup analysis reveals an additional advantage in the use of
pressure and pressure-derivative curves, even though it is redun- the Homer/superposition derivative of pressure: the resulting curves
dant. 10 With the double match, a higher degree of confidence in are neither compressed on the time axis, as for traditional Homer/
the results is obtained. To illustrate this, the fmal match of the ex- superposition analysis, nor on the pressure axis, as for buildup pres-
ample in Table I is shown in Fig. 8. sure type curves. The derivative displays the full amplitude of the
The skin coefficient is no longer present on derivative responses signal and therefore improves the sensitivity of the analysis plots.
when the infinite-acting radial flow configuration is reached (on 5. The noise apparent in the derivative data can be reduced when
the one-half line), and as a result, S can be estimated from only the superposition function is used because the slope (and the deriva-
the derivative CD e 2S match during the transition between the two tive) will not tend toward zero during the infinite-acting period.
1O.r---------------------------------------~
L
--------------
(2)
(I)
..............................,o::::------~
10-' 10 10' 103 AXI AX2
'oICo
Fig. 9-Dlfferentlatlon algorithm using three points.
Fig. 8-The combined match.
~
....
<:I
9
.-
.-.
NOISY nPE-<uIMS
I:u~
10'
....
--
_nPE-cuIMS
wellbonl
atorage I fracture
partial penetration
I
radial
homogeneous
fI_
I
__ tent
c:.:
~ ... ... fissures
multllayers
LOG .1t
closed
system
Fig. 10-Test of the differentiation algorithm on noisy data. Fig. 11-Log-log plot of a typical drawdown.
Differentiation Algorithm is naturally expanded at late times, when the pressure response is
The main concern when actual data are being differentiated is to hardly changing (thus making the noise-to-signal ratio significant).
improve the signal-to-noise ratio. Some noise will always be present Differentiation of early-time data generally poses no problem be-
because of gauge resolution, electronic circuitry, vibrations, etc. cause the amplitude of the time rate of pressure change is usually
Differentiation is difficult, if not inconclusive, for the relatively large enough to mask noise effects. In the few cases where early-time
high noise level associated with a low sampling rate. This is fre- data are particularly noisy, however, L has to be chosen longer than
quently the case with mechanical gauges, which also produce noise what is sufficient for the remaining data. A variable L can then be
on both pressure and time axes. used to avoid oversmoothing at late times.
Several approaches for differentiating data have been tried (Ap- Fig. 10 illustrates the differentiation of a generalized buildup type
pendix B). Because the correct result is not known when working curve. Both the original and the noisy curves are shown; in terms
with actual data, modified type curves were used to evaluate the of pressure, the two curves are not distinguishable. The differenti-
different methods. A random noise both proportional to and indepen- ation of the noisy curve, with a three-consecutive-points algorithm
dent of the amplitude of the PD signal was added to the type curve, (L=O), is also shown for comparison. The value L=O.I used in
and the number of points generating the PD curve was reduced by this case proved to be sufficient and does not affect the shape of
a random sampling process. the original derivative.
Examples of derivative calculations for the actual example dis-
Preferred Algorithm. The algorithm presented here is simple, and cussed earlier in this paper are presented in Table I. The derivative
is the best adapted to test interpretation needs. This differentiation of the pressure is estimated with respect to the modified Homer
algorithm reproduces the test type curve over the complete time time of Eq. 7 with L=O (no smoothing) and L=O.l.
interval better than others. It uses one point before and one point Common values for L are 0 (consecutive points) up to 0.5 in ex-
after the point of interest, i, calculates the corresponding derivatives, treme cases. Because L is expressed on different time scales ac-
and places their weighted mean at the point considered (Fig. 9). cording to the type of test and rate history, the resulting smoothing
effect (and the possible distortion of the derivative curve) depends
(dp/dX)i=[(~1/.1Xl).1X2 +(~z/.1X2).1Xd/(.1Xl +.1Xz ), on the particular case .
. . .. .. . . . .. . . . .. . . . .. ... . . .. . .. . . . .. . (8)
where I =point before i, 2 = point after, and X =time function (In ill End Effect. When late-time data are differentiated, i becomes closer
for drawdown, modified Homer, or superposition times expressed to the last recorded data point than L. Smoothing is not possible
in natural logarithm for buildups). The superposition function is on the right side. This is called the end effect. One solution consists
written as of using a "pseudo right" derivative in Eq. 8, which becomes fixed.
It is dermed between the last point and the first point before the
~~~~~~~~~~-~-~-~-~-~-~-~-~-~--~-~-=--=-=-~-~--~-~-:1~~~7~~
10
1 :;:;;;::::::~~10-1
Q
!:!Q
::
~------==-==-::;;.~-~---
_____~~~.3.::: 3.10-'
-----.,,:,-v
.;;//10:;;', ~
" (o,I(1w)
m1~~--~~-----b~________~~~--__----~~/__
3.~------__',~----_4
1()-1 102 103 104
tD/CD
When the complete recommended procedure is used, the distor- Thus, many types of flow regimes can appear before (and after)
tions produced by the differentiation algorithm presented here are the actual semilog straight line develops, and they follow a very
practically independent of the point density in the curve. The same strict chronology in the pressure response. Only a global diagnosis,
effect is expected to be produced on both data and theoretical curves, with identification of all successive regimes present, will indicate
as opposed to the algorithms that use all the points present in a given exactly when conventional analysis, like the semilog plot technique,
time interval for smoothing. is justified. Furthermore, the other characteristic regimes can be
Some of the irregularities observed in the derivative behavior were analyzed to provide much more thanjustkh, S, andp*, as illustrated
found to be part of the reservoir response. For example, oscillations below.
of pressure caused by tidal effects are emphasized by the derivative
at late time, when the signal is barely changing. DoublePoroslty Models
Another advantage is that the derivative would still give results One frequently encountered type of heterogeneous response
when the last flowing pressure is missing, as when the gauge is is double-porosity behavior, which is produced by fissured res-
run after shut-in or in some cases of changing wellbore storage. ervoirs. Two models of double-porosity behaviors have been
The p(~t=O) point is not needed to produce the derivative curve; studied 1618,21 : one assumes pseudosteady-state interporosity flow;
thus, provided that enough data are available, a unique match is the other assumes transient interporosity flow. Both models are con-
possible and the sldn is accessible. The derivative plots also tend sidered here, and the advantage of the derivative presentation in
to compensate starting-time errors encountered when shut-in time distinguishing various types of heterogeneous responses is shown.
is not accurate enough compared with the pressure-gauge sampling
Pseudosteady-State Interporosity Flow Model. The plots of the
frequency. In addition, for gas wells, the differential of the real
semilog slopes of examples are presented in Fig. 12.17 For Ex-
gas potential m(p) 15 replaces the calculation of an integral by that
ample A, w=1.0, Ae- 2S =3x 10- 4 , (CDe 2S,k+ma=IO- I ; and for
of a product.
Example B, w=O.I, Ae- 2s =IO- 7 , (CDe 2 )f+ma=10 4 . During
dm(p)/d In(~t)= {2p/[/.L(p)z(p)]}~t.1p'. . ............. (10) the homogeneous regimes, the response follows a derivative CD e 2S
curve, whereas at transition time, the flattening of the pressure be-
AppHcatlon to Heterogeneous Reservoir Bathing havior is changed into a very characteristic drop of the derivative.
Recent theoretical developments and related publications demon- The transition regime is now described by two families of curves
strate a general oil industry interest in the behavior of heterogeneous (Appendix A of Ref. 9): early transition is defined by the dimension-
formations. In fact, it is our experience, based on a very large num- less group (ACDf+ma)/[w(I-w)] and late transition by (ACDf +ma )/
ber of well tests, that in some areas, up to 30 or 40% of the wells (1 - w). If the storage effect is present at the start of the transi-
show a heterogeneous behavior. This is evident when high-accuracy tion (Example B), the response deviates from the corresponding
pressure data, high-definition analysis techniques such as plots of (ACDf+ma)/[w(I-w)] curve (1.11 X 10- 2 in this case), but storage
the derivative of pressure, and computer-aided interpretation are being over at late transition times, the match on (ACDf+ma)/(I-w)
used. The combined recent progress in data acquisition, data proc- (l.1IxlO- 3 ) is good.
essing, and computing techniques offers new prospects for the in- The double-porosity model illustrates the gain in sensitivity of
terpretation of well-test data. Much more information is pulled out the derivative approach. The flattening of the pressure response
of the well during today's tests. Interpretation should make full use during transition is generally difficult to identify on a log-log scale.
of all data available for analysis. In many cases, a semilog scale has to be used for refining the pres-
Fig. 11 presents a typical drawdown log-log plot of.1p vs. ~t. sure curve match. With the derivative plot, the heterogeneous nature
Four different time periods can be identified in the pressure of the response is obvious, eliminating the need for any further plot
response. for adjustments.
1. The wellbore-storage effect is always the first flow regime to Table 2 contains field data from a pressure buildup recorded in
appear. a fissured formation. The derivative of pressure suggests the hetero-
2. Evidence of well and reservoir heterogeneities then may follow. geneous behavior, and the combined log-log plot of pressure and
Such behavior may be a result of the effects of a fractured well, derivative (Fig. 13) is matched against the dual-porosity type curve
a partially penetrating well, a fissured formation, or a multilayered of Ref. 17. The buildup curve shown was generated with the flow
reservoir. history before shut-in (multirate curve). The differential was taken
3. After some production time, the system starts to exhibit a radial as the slope of the superposition plot. The double-porosity model
flow behavior, representing an equivalent homogeneous system used provides a fairly good description of the response, despite the
composed of all producing elements. discrepancy during part of the transition. Results of analysis are
4. Boundary effects may occur at late time. presented in Appendix C.
Conclusions
Transient Interporosity Flow Model. The derivative response of Transient test interpretation techniques have been reduced to the
the transient interporosity flow solutions is different from the identification of characteristic regimes that produce a straight line
pseudosteady-state curves: as for pressure. the derivative fJ' transi- when the pressure is plotted vs. time on various scales: radial flow
tion curves are obtained by displacing homogeneous CDe 2S curves with p vs. log(.:lt). wellbore storage and pseudosteady-state with
by a factor of two along pressure and time axes. 19 The theoretical t..p vs . .:It, linear flow with t..p vs. Kt, etc. With modern com-
semilog straight line of the transition regime (slope one-half of true puting facilities, there is no reason to limit the pressure analysis
semilog) is then changed into a constant-derivative 0.25 line. On to those restricted portions of the data during which a derivative
drawdown responses, the transient model does not show a derivative is constant. Such types of data, corresponding to pure specific re-
point below 0.25 during transition, as illustrated in Fig. 14. gimes, are often absent.
Fig. 14 presents for the same parameters (CDe 2S )f+ma= 1, fJ' = The method presented in this paper considers constant derivatives
10- 4 and w=IO- 2 , the drawdown response produced by two and changes of slope with a high definition. These transitional be-
matrix geometries, slabs and spheres. Though the pressure curves haviors are ignored on conventional straight-line plots and are often
look identical, the derivatives are different; the sphere model featureless on log-log pressure-vs.-time graphs. A diagnosis is per-
response does not reach the 0.25 straight line but remains above formed, with improved sensitivity, on the global response; the var-
it, whereas the slab curve is tangent to it but does not follow it for ious flow regimes are identified, according to a logical chronology .
a significant duration. New analytical solutions are needed for general reservoir model-
For buildups, the derivative during transition regime may exhibit ing to integrate characteristics neglected in traditional simplified
a lower value, down to 0.20, if the previous drawdown has not solutions.
reached total system flow at shut-in time. 19 Similar distortions The conclusions are as follows:
have been observed with a pseudosteady-state model. 17 1. The derivative approach improves the definition of the analysis
The discussion of the derivative curves of Fig. 14 illustrate the plots and therefore the quality of the interpretation.
drawback of the method that consists of drawing intermediate 2. The differentiation of actual data has to be conducted with care
straight lines on pressure plots. It is always possible to find several to remove noise without affecting the signal. The derivative ap-
reasonably straight portions on a standard pressure curve plotted proach does not produce errors or noise but only reveals them.
on any scale. This does not mean that an analysis of the intermediate- 3. The interpretation of pressure derivative is a single-plot proce-
straight-line characteristics is justified. The derivative approach dure. If enough data are available, pressure and time matches are