SPE

Society of Petroleum Engineers

SPE 20550

Pressure Desuperposition Technique for Improved Late-Time

Transient Diagnosis
by C. A. Ehlig-Economides, R. W. Ambrose and J. A. Joseph, Schlumberger Well Services

Copyright 1990, Society of Petroleum Engineers Inc.

This paper was prepared for presentation at the 65th Annual Technical Conference and Exhibition of the Society of Petroleum Engineers held in New Orleans, LA, September 23-26,1990.

This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the pa~er,
as resented have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material, as presented: does not necessarily re .ect
an p osition ~f the Society of Petroleum Engineers, its officers, or members. Papers presented at SPE meetings are subject to publication review by Editorial Committees of the Society
of ~~roleum Engineers. Permission to copy is restricted to an abstract of not more than 300 words. Illustrations may not be copied. The abstract should contain conspiCuous acknowledgment
of where and by whom the paper is presented. Write Publications Manager, SPE, P.O. Box 833836, Richardson, TX 75083-3836. Telex, 730989 SPEDAL.

ABSTRACT INTRODUCTION

Diagnosing the reservoir model from a field data set is
one of the main challenges in well test analysis. The log-
log pressure and pressure derivative plot is now firmly
established as a primary diagnostic tool. This paper pro-
vides a new derivative formulation that enhances model
diagnosis and addresses techniques for reducing noise in
the derivative response.
Transients initiated by a sufficiently large increase in the
surface rate closely represent the ideal reservoir response.
However, when a transient is initiated by a decrease in
surface rate, distortions due to superposition effects from
previous surface flow rate changes are frequently present
in the late-time derivative and can lead to incorrect char-
acterization of reservoir boundaries.
To improve the late-time response, the pressure data are
corrected for effects of a variable surface rate history be-
fore differentiation. The resultant desuperposition deriva-
tive avoids assuming a model for the reservoir behav-
ior by using the late-time character of a representative
transient. The desuperposition derivative appears as an
equivalent drawdown (injection) response regardless of
the nature of the perturbation that was actually used to
induce the particular transient. This technique is par-
ticularly useful in reservoir and limits tests designed to
delineate heterogeneities or boundaries; it also applies to
multirate and multilayer tests.
Noise in the derivative can either mask or give a false im-
pression of the true reservoir behavior. Overly smoothing
noisy data may obscure the response and lead to erro-
neous model diagnosis. This study addresses logarithmic
filtering and sampling techniques for reducing noise in
the data.
References and illustrations at end of paper.
From its inception, the log-log pressure derivative plot
has played a fundamental role in reservoir model diagno-
sis. 1 ,2 For tests where downhole flow rates are acquired,
its counterpart is the convolution derivative. 3 Experience
with these diagnostic plots has shown that the shape of
the classical derivative response depends on the overall
well test sequence. These discrepancies typically occur in
late time and can interfere with the correct characteriza-
tion of boundary effects.
Model diagnosis entails pattern recognition, during which
the shape of the derivative of the transient test data is
identified with a known reservoir model.' The comparison
may be achieved visually or automatically by applying
artificial intelligence techniques. 5 ,6
Since the late-time derivative shape can be influenced by
the test sequence and the well flow rate history and may
not necessarily match a shape in the lookup catalogue
of theoretical drawdown responses, misdiagnosis of the
reservoir model is a possibility. Misdiagnosis will likely
be corrected during subsequent analytical modelling 7 be-
cause the data are matched with a simulated reservoir
model response that accounts for the entire rate history.
However, a match might be obtained using an improper
model resulting in incorrect reservoir characterization.
Pressure and convolution derivatives are currently com-
puted by differentiating with respect to radial superpo-
sition and convolution time groups, respectively.l-s This
paper describes a desuperposition technique which ren-
ders the derivative as an equivalent drawdown or injection
response which can be directly compared with theoretical
type curves. The desuperposition technique is explained
for pressure, convolution, and deconvolution derivatives,
and a simulated example shows its application.
* Mark of Schlumberger

201
2 EHLIG-ECONOMIDES, c.A., AMBROSE, R. W. AND JOSEPH, J. A.
In multirate tests, noise in the data can be enhanced by
the derivative. A section on operational considerations
addresses the nature of various kinds of noise in data and
describes techniques for logarithmic filtering and sam-
pling of the data that reduce noise in the derivative re-
sponse. The field example compares data processing tech-
niques and illustrates how the desuperposition derivative
works in practice.
NEW DERIVATIVE FORMULATION
Each time a step change is made in the surface rate, a
transient condition is induced in the wellbore and reser-
voir. The effects of each surface rate change endure for-
ever, but diminish continously over time. The behavior
of the transient response following anyone surface rate
change is distorted by the superposition effects of previ-
ous variations in the surface flow rate. S
Superposition effects may be caused either by events oc-
curring at one or more instants in time at the test welt,
or by events or conditions in nearby wells which interfere
with the test well. Interfering wells cause superposition
in space. 9 This section concerns correcting the derivative
diagnostic plot for effects of superposition in time, that
is, for distortions in the transient response caused by the
flow rate history before the start of the transient. The
Slider lO correction provides a means to account for su-
perposition in time in pressure buildup analysis for wells
that reach pseudosteady state flow before being shut in.
This work generalizes this idea and applies it to log-log
diagnosis.
The desuperposition derivative is computed as the deriva-
tive of the desuperposed pressure function with respect to
an appropriate time function. For the desuperposed pres-
sure derivative or the desuperposed deconvolution (us-
ing continuously measured sandface rates) derivative, the
differentiation is conducted with respect to In(At). The
desuperposed convolution derivative is computed from
the desuperposed unit rate-normalized pressure with re-
spect to the sandface rate-convolved time function. The
desuperposition is computed using the late time trend
from a representative transient which is believed to have
minimal late-time data distortion due to superposition
in time. A more complete description of the technique is
provided in the following sections.
Superposition Time Function Computation
From the superposition principle,s for At = t - tn(t >- tn)
and given a series of n step changes in the surface flow
rate, AQj = (Qj+l - Qj)' for j = 1, ... , n, with Qo = 0, the
downhole pressure drawdown is given by the following
expression:
n
Pi - Pwf(tn + At) = L(Qj+l - Qj)APdd(t n - tj + At), (1)
j=o
where APdd(t) is the unit transient drawdown pressure
response to a step surface rate change from zero to a
finite flow rate. For the first drawdown,
202
SPE 20550
The dimensionless wellbore pressure drawdown, Pw n (t n),
in Eq. (2) is given by any solution of the diffusivity equa-
tion with suitable initial and boundary conditions. When
all rate changes are made at the surface, APdd must ac-
count for the wellbore transient responses (storage, etc.).
Equation (1) can be used to simulate a drillstem test in
a newly drilled well. When a transient test is conducted
on an established well, the entire flow history is rarely
known. However, in Eq. (1), it is possible to drop all
terms with t n - t k >->- At and AQk ~ AQn' for j = 1, ... , k:
n
Pi - Pwf(tn + At) "" L(Qj+l - Qj)APdd(tn - t j + At). (3)
j=k
In addition, for any transient response to a given step
rate change, a plot of the pressure change, APwf(At) =
IPwf(tn + At) - Pwf(tn)l, as a function of elapsed time,
At, mimics APdd(At) for At :5 O.I(t n - tn-d whenever the
changes in surface flow rate are of approximately equal
magnitudes. In terms of the drawdown response, the
pressure change is given by the following:
n
APwf(At) "" I L(Qj+l - Qj) [APdd(tn - t j )
j~ W
-APdd(tn - t j + At)] I.
The above expression accounts for n+I rate changes from
some initial time, to, with an unknown rate history prior
to to.
The pressure change can also be computed as
with
When Pn can be expressed by a particular function of
time for t n - t n- l :5 t :5 t n - t l , this time function is sub-
stituted into Eq. (6). Since infinite-acting radial flow
is commonly observed as the late-time behavior in many
transient tests, we have
1 apJ.t atk
Pn(t) = - kh lInt + In 4> 2 + 0.80907]. (7)
2 J.tCtTw
The superposition time function for infinite-acting radial
flow is given by
1 n-l ( tn - tj )
t, upIA = (Qn+l _ Qn) L(Qj+l - Qj)In t n _ t. + At . (8)
J=o J
If the actual late-time response is not infinite-acting ra-
dial flow, then t, upe should be computed with a different
function. For example, if the reservoir is closed,l1
SPE 20550 PRESSURE DESUPERPOSmON TECHNIQUE FOR IMPROVED LATE-TIME TRANSIENT DIAGNOSIS 3

(t) = 21rCXtk t + ~ln 2.2458A, (9)

on a log-log plot of the pressure change and the pressure
PD 1JJ.lctA 2 C ATw2 derivative versus elapsed time, the trends represented by
Eqns. (13) and (14) are straight lines. However, be-
Hence, for pseudo-steady state flow, the following super-
cause of the constant b which is absent in the derivative,
position time function should be used in for t supe in Eq.
the straight line trend may be apparent in the pressure
(5):
derivative and not in the pressure change.
For pressure buildups, and for transients after the first
drawdown, the current practice is to compute the deriva-
tive with respect to the infinite-acting radial superposi-
Alternatively, if the late-time response is controlled by tion time function, t sup , given by 1
parallel no-flow planar boundaries, as in an elongated or
channel-shaped reservoir, 12 then t sup = tsupIA -In(~t). (16)

(11) When the late-time unit drawdown response is given by

Eq. (7), then in late time,
giving rise to
d[~Pwf(~t)] _ (Qn+l - Qn)BJ.l
(17)
1rCXtk 1 dt sup - 2cx pkh
1JJ.l Ct W2 (Qn+l - Qn)
n-1 In early time, all but the last term in the time function
L(Qj+l - Qj) [Jt n - tj - Jt n - tj + ~t] , can be neglected, and using Eq. (2):
j=o
(12) d[~Pwf(~t)l = [(Q _Q )1 cxpBJ.l d[PwD(~t D)] (18)
dt sup n+l n 2kh dln(~t) .
If a gas cap, an aquifer, or an injector-producer pattern
result in an effective constant pressure boundary, ~Pdd is
a constant in late time. When the late-time response in the pressure change is
not given by Eq. (7)' another approach is needed to cor-
The correct function to use for t supe can be determined rect for superposition effects before differentiating, If the
from transient data acquired for a period of time long pressure transient response is available for a sufficiently
enough that it has minimal superposition effects from the long duration, then the integral in Eq. (6) can be com-
previous flow history of the well. Once this function has puted from the late-time data as follows:
been defined from such a representative transient, super-
position effects can be greatly reduced or eliminated in -1 n-1 rtn-tj+At d[Pwf(T)]
derivatives computed for other transients acquired in the Psupe = (Q _ Q ) L(Qj+l - Qj) it t dr dr.
n+1 n j=o n- j
same well. The function is determined from a derivative
(19)
response for the representative transient, and the method
for determining this function is explained in the next sec- When ~Pwf(~t) = me(~t)a+b for t ~ t n -t n - 1, the follow-
tion. ing results:
n-1
Desuperposition Pressure Derivative Psupe = (Q me Q ) L(Qj+l - Qj)[(tn - tj)a
n+l - n j=o (20)
If the pressure change, ~Pwf(~t), follows a trend repre- - (t n - t j + ~t)a].
sented by a linear function of (~t)a, e.g.,
From Eq. (5),
(13)
~Pwf(~t) = Psupe - (Qn+l - Qn)~Pdd(~t), (21)
the derivative of the pressure change with respect to the
logarithm of time is given by
The expression (Qn+l - Qn)~Pdd(~t) correctly represents
d[~Pwf(~t)l a a theoretical drawdown response. As such, for each tran-
dln(~t) = ma(~t) . (14)
sient, the log-log diagnostic plot should consist of IPsupe -
~Pwf(~t)1 and its derivative with respect to In(~t).
When ~Pwf(~t) = mln(~t) + b,
From a practical point of view, the values for me and a in
d[~Pwf(~t)]
(15) Eq. (20) are determined as follows: For a long duration
dln(~t) = m.
transient response initiated by a large rate change, plot
Eqns. (13-15) illustrate why the derivative of pressure d[APw!(AtlJ
dt on Iog- Iog coor d'mates. If t h
e i t 'Ime d
ate- '
eflva-
change with respect to the logarithm of elapsed time has tive d";,ta trend is a straight line, the value a is computed
been used for diagnosing flow regime in drawdown tran- as the slope of a line fit through the late-time derivative.
sient tests. Nearly all of the commonly recognized tran- If a i- 0, the value me is computed as the slope of a line
sient flow regimes are represented by one of the above fit thought the same data on a plot of ~Pwf(~t) versus
equations; the common exceptions are dual porosity and (~t)a. When the late-time behavior is infinite-acting ra-
dual permeability behavior. It is important to note that
203
4 EHLIG-ECONOMIDES, c.A., AMBROSE, R. W. AND JOSEPH, J. A.
dial flow, a = 0, and me is the slope on a plot of APwf(.At)
versus In(At).
The correction P.upe is effective as long as the trend given
by me(At)a + b represents the behavior of Apdd(t) for the
duration of time from the beginning of each step change
in the production/injection or test flow history to the
end of the transient being corrected. If the selected rep-
resentative transient is not sufficiently long, the transient
behavior observed at the end of the selected representa-
tive transient will not include additional boundary effects
which would only be visible much later in time. In that
case, the correction will not apply for rate changes that
occurred long enough before the end of the transient be-
ing corrected to sense boundary effects unaccounted for in
P.upe' and the computed desuperposition derivatives may
still show some separation. Hence, If the desuperposition
derivatives for a series of transients do not agree in late
time, this may be a sign that there are later-time boundary
effects not visible in anyone transient response. 13
To illustrate the points in this section, Fig. 1 shows the
theoretical pressure change and derivative responses for
a well located very near one of two parallel sealing faults
in a reservoir in contact with an aquifer at some dis-
tance from the well. Straight line trends apparent in the
deriative (and the respective log-log slope) are wellbore
storage (1), infinite-acting radial flow before and after
near fault is sensed (0), channel linear flow (1/2), and
channel linear flow with one constant pressure boundary
(-1/2). (The apparent plateau seen in late time does not
represent radial flow; instead, it represents the transition
from infinite to semi-infinite channel flow.)
A 48-hour test of this well would sense channel linear
flow in late time. However, if surface rate changes occur-
ring more than 1000 hours before the test were used in
the desuperposition, some distortion of the late-time half
slope trend would appear on the desuperposition deriva-
tive for the 48-hour transient.
Clearly, no one test would sense all of the responses shown
in Fig. 1. However, the part of the response which ap-
pears depends on the reservoir and fluid properties. A
shorter test would sense fewer of the flow regimes. A
reservoir with higher mobility (k/ p,) or smaller total com-
pressibility would sense the flow regimes earlier, possibly
missing the first radial flow and sensing the aquifer in
late time. The desuperposition used depends on which
late-time straight line behavior appears at the end of the
transient for the particular test.
Desuperposition Convolution Derivative
When, in addition to pressure, the continuous downhole
flow rate, q, is measured in response to a step change in
the surface rate, the sandface flow rate exhibits a tran-
sient variation due to various effects including wellbore
storage, wellbore phase redistribution, or wellbore cross-
flow (for commingled flow). For pressure buildup tests,
the sandface flow rate variation is known as afterflow.
Until the sandface flow rate eventually reaches the same
value as the downhole equivalent value of the surface flow
rate, wellbore-dominated contributions to the pressure
204
SPE 20550
transient response are significant and mask the reservoir
flow regimes. The continuously measured sandface flow
rates are used to compute a representation of the reservoir
response with negligible or considerably reduced wellbore
storage effects.
Discretizing the continuously varying sandface rates into
a series of small step changes, and defining APwf(At) =
IPwf(t n + At k ) - PWf(tn)l, where At k = t k - tn, for t k >- tn,
results in
k-l
APwj(At) = P.upe + 2)qi+l - qi)Apdd(t k - til, (22)
i=o
or alternatively,
APwj(At) = P,upe
k-l
+ 2)qo - qi)[Apdd(tk - ti-l) - Apdd(t k -til]
i=l
(23)
The radial sandface rate-convolved time function, tSFRC'
introduced by Meunier, aI.,14 is extended to include the
well tiow jll~;Lory as fU1IUWS:
1
tSFRC = (qo - qk) ll(Qn+l - Qn)lt.up/A
+~
(tk-t.) (24)
L.,,(qo - qi)ln t _ ~l
i=l k ,
+(qo - qk)ln(tk - t k _ 1 )]
with t' UP / A defined as in Eq. (8).
When Apud(t) is equivalent to PD(t) defined by Eq. (7)'
(25)
where Aq(At) = Iqo - q(At)l.
The above unit rate-normalized pressure change when
plotted versus tSFRC provides a straight line, and per-
meability is computed from the slope of the line when
the infinite-acting radial flow regime has been identified
in the transient. 14 However, in a similar fashion as with
the pressure derivative, the convolution derivative is com-
puted as the derivative of the unit rate-normalized pres-
sure change with respect to tSFRC. 15 ,3 Computed as such,
the convolution derivative response preserves the trends
in early time of the unit rate-normalized pressure change
and completely corrects for superposition only when the
late-time trend is infinite-acting radial flow.
To correct for superposition when the late-time trend is
not infinite-acting radial flow, the function to use for P,upe
is determined from the late-time unit rate-normalized
pressure change data in the same manner as for pressure
data described in the previous section. The derivative of
APw/(At) • h . 1 d
IAq(Atl!(Qn+l-Qn) p.upel WIt respect to t SF RCo IS p otte
SPE 20550 PRESSURE DESUPERPOSITION TECHNIQUE FOR IMPROVED LATE-TIME TRANSIENT DIAGNOSIS
versus (ll.t) on log-log coordinates, with
(26)
Desuperposition Deconvolution Derivative
The convolution derivative is computed by applying the
approximation in Eq. (7) to each change in the sandface
flowrate. The deconvolution computation makes no such
assumption.
From Eq. (22),
ll.Pwj(ll.t)
-:---'--'--:-:-/2~---:::-:--P.upe=
ll.q(ll.t) (Qn+l - Qn)
Ef-=-;(qi+l - qi)ll.Pdd(tk - til
ll.q(ll.t)
(27)
For diagnosis of the correct reservoir model, it is sufficient
to compute ll.Pdd(ll.t) for N logarithmically spaced values
of (ll.t). That is, compute ll.Pdd(ll.tj ) for tj = to + (t l -
t o)lO(i-l)Ot for j = 1, ... , N, and Ot = In (tt~t:o ) /(N - 1).
For this computation, values of ll.Pwj(ll.t) and ll.q(ll.t) are
needed for times tN_i+l = tk-t i for i = 1, ... ,N. This data
sample requires logarithmic filtering, as described in the
section on operational considerations.
For the N values of ll.tN_i+l' Eq. (27) provides N equa-
tions in N unknowns. Once ll.Pdd has been computed for
the logarithmically-spaced points in time, the derivative
of ll.Pdd with respect to In(ll.t) is used for diagnosing the
reservoir model.
Sinaulated ~xanaple
This example represents the response of a well centered in
a reservoir with an impermeable circular boundary. Both
downhole pressure and flowrate responses were simulated
for the test sequence shown in Fig. 2. Reservoir and fluid
properties are provided in Table 1.
Figure 3 shows a comparative diagnostic plot 16 for this
example. The plot shows the pressure change and com-
puted derivatives for each transient normalized by the
magnitude of the step flow rate change that initiated the
transient. For each transient, the pressure change was dif-
ferentiated with respect to the radial superposition time
function and the unit rate-normalized pressure was differ-
entiated with respect to the sandface rate-convolved time
function. Although pressure and convolution derivatives
agree for each transient, after about 2 hours there is con-
siderable separation in the computed derivatives from one
transient to another, because the late-time reservoir re-
sponse is pseudo-steady state, and not radial flow.
Since the first transient was initiated by a single step
change in surface rate, this represents the theoretical
drawdown response. Hence, to determine the superposi-
tion correction, P8UPe' the slope of the late-time derivative
205
5
trend on log-log coordinates is determined, as in Fig. 4.
Then, in Fig. 5, the slope of the line drawn through the
same data plotted on cartesian coordinates is determined.
The value of me is computed as the cartesian slope di-
vided by the magnitude of the surface flow rate change
that initiated the transient.
The desuperposition derivatives for pressure and rate-
normalized pressure are shown in Fig. 6. As seen in
the figure, all the derivatives now reproduce the same
late-time trend.
OP~RATIONAL CONSID~RATIONS
The method described in this paper uses a straight line
picked from the late time data for further processing. As
such, the quality of initial processing becomes especially
important because errors in early processing can be mag-
nified in the subsequent step. Noise that appears in late
time can degrade the final data quality through an un-
representative straight line pick.
Noise is any data characteristic caused by conditions not
associated with a reservoir model. It may be indistin-
guishable from model behavior; this occurs when the
noise is of the same time scale as features in the model.
When this happens, it may be best to account for these
effects by incorporating them directly into the model.
Noise of this type can be cause by sun and moon tides,
surface rate fluctuations or systematic slowing of rate,
flow in the pumped well annulus, wellbore phase redistri-
bution, commingled flow and gauge drift.
When noise is of a time scale different from the time scale
predicted by the constructed model of behavior, it may
be removed statistically. This can be accomplished by
replacing the pressure measurement at each data point
with the output from a low pass filter, where the cutoff
point of the low pass filter is above the highest frequency
expected from the signal.
The solution to the diffusivity equation predicts that the
characteristic time scale for reservoir behavior is a log-
arithmic function of time. This means that the highest
frequency expected from the signal decreases as the log-
arithm of time, and therefore the cutoff point of the low
pass filter is reduced logarithmically with time. This is
logarithmic filtering, which can be implemented by trans-
forming the time of measurement into logarithmic time,
and then applying a low pass filter with constant charac-
teristics.
With implementation of low pass filtering in logarithmic
time, sampling in logarithmic t;mf' i~ sngi!/'stI'G Arrod-
ing to the Shannon 17 sampling theorem: "If a function
contains no frequencies higher than W cps, it is com-
pletely determined by giving its ordinates at a series of
points spaced 1/2W seconds apart." Since the function
has been low pass filtered in logarithmic time, Shannon's
theorem is applied in logarithmic time. This is logarith-
mic sampling, in which the signal is represented on a grid
of points equally spaced in logarithmic time. A time grid
of this nature can be described as follows:
(28)
6 EHLIG-ECONOMIDES, C.A., AMBROSE, R. W. AND JOSEPH, J. A. SPE 20550

where t i is the time of the i-th grid point, to is the time at In the next step, logarithmic filtering and sampling are
the beginning of the transient, Ot o is the log-time of the used to demonstrate the desuperposition derivative on
first sample, and 6t is the log-time of the sample interval. each of the transients shown in the test sequence in Fig.
7. The rate change for the last transient is of the largest
Operationally, logarithmic filtering and logarithmic sam-
magnitude, and hence is likely to provide a late-time re-
pling are performed simultaneously. A grid is calculated
sponse most representative of theoretical drawdown be-
in logarithmic time, the original data is transformed into
logarithmic time, and low pass filtering is effected to com- havior.
pute the pressure values on the computed grid of points. Figure 10 shows the comparative diagnostic plot for the
This procedure is described in detail in Ref. 18. pressure change and pressure derivative responses. Al-
though the three transients (1, 3, and 5) initiated by step
increases in the surface rate appear to follow about the
FIELD EXAMPLE same trend in late time, transients 2 and 4 show consid-
erable distortion of the late time trend. As for the sim-
The data in this field example were originally published ulated example, a representative transient, in this case
in Ref. 16. The test was a layered reservoir test of six transient 5, is selected for determination of the late-time
zones in an oil reservoir thought to be bounded by parallel slope of the pressure derivative on log-log coordinates, as
sealing faults. In this paper, the focus is on diagnosis in Fig. 11. Then the late-time slope is determined for the
using the desuperposition derivative. same set of data points in Fig. 12. Finally, Fig. 13 shows
The multirate test seqence is shown in Fig. 7. From that the desuperposition derivatives all follow about the
2000 to 4000 points were acquired during each of the five same trend in late time.
transients shown. Hence, the data were resampled before
The interpretation in Ref. 16 indicated that there was
beginning the transient interpretation to reduce the num-
evidence of an additional boundary not characterized by
ber of points. Figure 8 shows a comparison, for one of the
anyone transient response during the test. The slight
transients, between logarithmic filtering and logarithmic
spread in the desuperposition derivatives may be an in-
sampling strategies and simple subset resampling. The
dication of the existence of a third boundary, the effects of
square symbols in the figure present the pressure change
which are apparent only for an elapsed time longer than
data after logarithmic filtering using a sample that is
the 2-hour transients, but shorter than the time since the
equally spaced in 10g(Ll.t). It is a sparse logarithmic grid,
beginning of the production history.
with only 71 points. The triangles represent a 500 point
subset of the raw data with no logarithmic filtering, with
the 500 point subset lying on an approximate logarithmic
IMPLICATIONS FOR TEST DESIGN
grid. As can be seen, the pressure function represented
by these two sets of points appears to be the same. Note
that although 500 points were used in the calculation,
some were omitted from the plot because they were so
close together; therefore, fewer than 500 points appear in
the graphic.
The derivative of the 500 point subset is represented by
the dotted line. Because of the high number of points,
the high frequency content of the data has been retained.
It cannot be reservoir behavoir because the frequency is
too high. Rather, it must be noise, and may be primarily
due to tool vibration and flow rate fluctuation, since these
data were acquired when the well was flowing. To remove
this noise, smoothing was applied to the derivative. 2 The
results of this are represented by the plus-sign symbol.
The remaining curve of figure 8 is the derivative com-
puted directly from the logarithmically filtered and re-
sampled curve represented by only 71 points, and is rep-
resented by the circles. As can be seen, this curve is vir-
tually the same as the smoothed curve with 500 points.
In Fig. 9, the same comparison is extended to the con-
volution derivatives, with the same favorable result.
The conclusion is that with logarithmic filtering and sam-
pling, a sparse sample of the data can sufficiently repre-
sent the transient for calculating the derivative and con-
volution derivative. This method provides a more effi-
cient way to achieve the result otherwise obtained by
keeping a large number of points, performing a precise
calculation, and then smoothing to reduce noise in the
computed signal.
206
SPE 20550 PRESSURE DESUPERPOSITION TECHNIQUE FOR IMPROVED LATE-TIME TRANSIENT DIAGNOSIS 7
The desuperposition derivative is used in a new technique cations of the logarithmic filter were investigated. Ex-
for MLT test interpretation that provides a representa- amples showed that logarithmically filtered data repro-
tion of the conventional single layer response for each duce the same derivative response with fewer points than
layer tested. The implications for MLT test design are smoothed derivatives of a subset of the data.
discussed further in Ref. 13.
The new formulation underscores the importance of test
The desuperposition derivative is particularly useful for design for transient tests conducted to characterize reser-
reservoir limits tests. Pressure buildup data can be am- voir boundaries. A representative transient, initiated by
biguous when outer boundary effects are present. Some- a single step surface rate change, or with a step of much
times the the effect of superposition is a downturn in the larger magnitude than previous rate changes in the well
pressure derivative that mimics the effect of a gas cap flow history, will mimic the behavior of a theoretical draw-
or aquifer. As previously indicated,16 model diagnosis is down type curve. Other decreasing rate and buildup
improved by designing the test to include a long draw- transients can then be corrected for superposition using
down or injection transient acquiring downhole pressure the desuperposition derivative technique.
and flow rate.
Even when downhole flow rates are not measured, if the
flow to the surface during the limits test can be main-
NOMENCLATURE
tained at a constant rate, then the late-time trend can a slope of late-time line on log-log plot
be taken from the drawdown transient data. In turn, A reservoir area
this trend can be used to desuperpose the buildup data. b constant term in linear function
Drawdown pressures continue to vary in late time, keep- B formation volume factor
ing the signal-to-noise ratio high enough to avoid prob- C storage coefficient
lems of gauge sensitivity or drift. Downhole flow rate CA reservoir shape factor
data in late time remain nonzero during a flowing tran- Ct total compressibility
sient, above the low flow rate range at which the flowme- d differentiation operator
ter becomes inaccurate. The buildup data typically have I(t) time function
a higher signal-to-noise ratio in early time. Hence, the h formation thickness
desuperposed buildup will be the best representation of k formation permeability
the entire transient response. In natural logarithm
m slope of linear function
The desuperpostion derivative may not apply for tests
me slope of late-time line on specialized plot
with buildup transients that are much longer than the
N number of unknowns in Eq. (27)
previous drawdown. Since the trend used for the desu-
P pressure
perposition must continue through the length of both
Pi initial reservoir pressure
the drawdown and the buildup, straight line trends in
P.upe generalized superposition pressure function
the drawdown due to wellbore storage/skin, partial pen-
etration, a vertical fracture, or dual porosity may not PWD dimensionless wellbore pressure drawdown
Pwf flowing bottomhole pressure
last long enough. When the well is shut in during any
one of these transient features, the buildup response may Pdd drawdown pressure response to single step
have considerable distortion, including a crossover of the rate change
pressure change and pressure derivative responses in late q downhole flow rate
time. The desuperposition derivative will not apply for qo downhole flow rate at start of transient
slug, Impulse", or closed chamber tests. qj downhole flow rate at time t j
Q surface flow rate
Qj surface flow rate at end of jth transient
CONCLUSIONS Tw wellbore radius
S mechanical skin factor
The desuperposition derivative introduced in this paper t time
corrects for superposition effects using the late-time data to start time for transient
from a representative transient. The late-time charac- tD dimensionless time
ter of the desuperposition derivative is consistent for all tj start time for jth transient
transients in a multirate test. t sFRC sandface rate convolution time function
tSFRCO sandface rate convolution time function
Superposition effects caused by the previous flow history defined in Eq. (26)
have been distinguished from the effects of afterflow or infinite-acting radial superposition time
sandface flow rate variation. The desuperposition convo- function
lution or deconvolution derivative formulations account generalized superposition time function
for both types of flow rate variation. infinite-acting radial superposition time
Noise in transient data inhibits correct diagnosis of a function defined in Eq. (8)
reservoir model. Depending on the data acquisition elec- superposition time function for closed
tronics, logarithmic filtering and sampling can provide a circular reservoir defined in Eq. (10)
better representation of the transient signal than a sub- superposition time function for elongated
set of the raw data points. The signal processing impli- reservoir defined in Eq. (12)

207
8 EHLIG-ECONOMIDES, CA., AMBROSE, R. W. AND JOSEPH, J. A. SPE 20550

w elongated reservoir width 7. Ehlig-Economides, C.A., Joseph, J.A., Ambrose, R.W.,

W sampling frequency Jr. and Norwood, C.: "A Modern Approach to Reservoir
Test Interpretation," paper SPE 19814 presented at the
Greek Symbols 1989 SPE Annual Technical Conference and Exhibition,
San Antonio, Oct. 8-11.
units conversion factor for dimensionless
pressure 8. Odeh, A.S. and Jones, L.G.: "Pressure Drawdown Anal-
units conversion factor for dimensionless ysis, Variable-Rate Case," JPT (Aug. 1965) 960-4.
time 9. Onur, M., Serra, K., and Reynolds, A.C.: "Analysis of
gas gravity (air=l) Pressure Buildup Data Obtained at a Well Located in a
fluid viscosity Multi-Well System, paper SPE 18123, presented at the
porosity 1988 SPE Annual Technical Conference and Exhibition,
variable of integration Oct. 2-5.
log-time of sample interval
log-time for first sample 10. Slider, H.C.: Practical Petroleum Reservoir Engineering
elapsed time in one transient Methods, PennWell Books, Tulsa, (1976).
pressure change 11. Matthews, C.S. and Russell, D.G.: Pressure Buildup and
downhole flow rate change Flow Tests in Wells, Monograph Series, SPE, Richard-
= surface flow rate change son, TX, (1967) 1.

Subscripts 12. Ehlig-Economides, C.A. and Economides, M.J.: "Pres-

sure Transient Analysis in an Elongated Linear Flow Sys-
i summation index for continuously measured tem," SPEJ (Dec. 1985) 839-47.
downhole flow rate
13. Ehlig-Economides, C.A.: "Model Diagnosis for Layered
j su=ation index for step surface flow rate
Reservoirs," , to be presented at the 1990 SPE European
changes
k
Petroleum Conference, The Hague, Netherlands, Oct. 22-
su=ation limit for continuously measured
24.
downhole flow rate
n number of the current transient 14. Meunier, D., Wittmann, M.J., and Stewart, G.: "Inter-
pretation of Pressure Buildup Test Using In-Situ Mea-
surement of Afterflow," J. Pet. Tech. (Jan. 1985).
Acknowledgment
15. Bourdet, D., and Alagoa, A.: "New Method Enhances
Well Test Interpretation," World Oil (Sept. 1984).
The authors wish to thank Schlumberger management
for permission to present this paper. 16. Joseph, J.A., Ehlig-Economides, C.A. and Kuchuk, F.J.:
"The Role of Downhole Flow and Pressure Measurements
in Reservoir Testing," paper SPE 18379 presented at the
REFERENCES
1988 SPE European Petroleum Conference, London, Eng-
1. Bourdet, D., Whittle, T.M., Douglas, A.A. and Pirard, land, Oct. 16-19.
Y.M.: "A New Set of Type Curves Simplifies Well Test 17. Shannon, C. K: "Co=unication in the Presence of
Analysis," World Oil (May 1983) 95-106. Noise," Proc., IRE (1949) 37, NO.1, 10-21.
2. Bourdet, D., Ayoub, J.A., and Pirard, Y.M.: "Use of 18. Ambrose, R.W.: internal report
Pressure Derivative in Well Test Interpretation," SPEFE
(June 1989) 293-302. 19. Ehlig-Economides, C.A.: "Testing and Interpretation in
Layered Reservoirs," JPT (Sept. 1987) 1087-90.
3. Ehlig-Economides, C., Joseph, J., Erba, M. and Vik,
S.A.: "Evaluation of Single-Layer Transients in a Mul- Tw = 0.25 ft
tilayered System," paper SPE 15860 presented at the 1986
SPE European Petroleum Conference, London, England, 4> = 0.20
Oct. 20-22. h = 100 ft
4. Ehlig-Economides, C.: "Use of the Pressure Derivative
for Diagnosing Pressure-Transient Behavior," JPT (Oct. Ct = 0.00001 psi- 1
1988) 1280-2. It = 1 cp
5. AI-Kaabi, A.U., McVay, D.A., and Lee, W.J.: "Using an
Expert System to Identify the Well Test Interpretation FV F = 1 RB/STB
Model," paper SPE 18158 presented at the 1988 SPE k = 100 md
Annual Technical Conference and Exhibition, Houston,
Oct. 2-5. C = 0.01 RB/psi
6. Allain, O.F. and Horne, R.N.: "Use of Artifical Intelli- Te = 1000 ft
gence for Model Identification and Parameter Estimation
in Well Test Interpretation," paper SPE 18160 presented Pi = 5000 psi
at the 1988 SPE Annual Technical Conference and Exhi- Table 1-Reservoir and fluid properties for simulated
bition, Houston, Oct. 2-5.
'-- L. example. .._ _ _ _ _.__.. _.~ .._...J

208
SPE 20550 PRESSURE DESUPERPOSmON TECHNIQUE FOR IMPROVED LATE-TIME TRANSIENT DIAGNOSIS 9

Flow Regime Identification Plot - Transient

10',-------------------------,
10'
10'
'iii
0- 10'
ui
0-
J
o
L.
() 10'
Ql
>
'';:;
o
>
.~ 101
o Lote- Ti Slope:= 1

Od
End of Test
<J
0-
10' /
10-3 Wellbore Storage 8

10' 'r-----,---....---..---....- - - r - - - r - - - . . , - - J
10' 10' 10' 10' 10' 10' 10' 10 10 ' 10" 10' 10'
lit, hr

Fig. 1-Test design simulation. Fig. 4-Determination of late-time log-log derivative slope.
5000....-_..=Dc:ec:.te:::r..:.m:..::.:..in:.=0c:.ti:.=oc:,:nc....::o.:...f....:L:.:a:.:t..=e..:.-..:.T-=im=e..:...=S::.:lo:rp:.:e:....,:o::.nc....::S.r::p.::e.::c·:.::la:::.li.::z.::ed:::...:.P..:.1.::o-,--t_,
2000.00

lJ 4960
LL
I':::::-
mi)
m 4920

0.0 4880
5000.00

'iii 4840
0....- 0-
ICfl
mO- ci:. 4800
000 0 0 a o o
4600.00J>-- -ljl
4760 Slape = 0.7454 psi/hr
2000.00
4720

4680

0.0 f----,---,,---,------,.--,---,---r--,.....A----,.-""---1 4640+-,....-,---,....--,--,--,--,--,--,--,--r=""I

o 20 40 60 80 100 120 140 160 180 200 o 10 w ~ W ~ w ro W 00 m 00 ~
Time, hr M, hr

Fig. 2-Test sequence for simulated data, Fig. 5-Determination of late-time slope on specialized
plot.
Comparative Dia nostic Plot Comparative Dia nostic Plot
1
10' [IJ Pressure Change, Transient 1 10 l!l Pressure Change, Transient 1
o Pressure Derivative o Pressure Derivative
6 Convolution Derivative 6 Convolution Derivative
~ Pressure Change, Transient 2 ~ Pressure Change, Transient 2
c) Pressure Derivative (') Pressure Derivative
~ Convolution Derivative
'iii 10'
~ Convolution Derivative
'* Pressure Change, Transient 3 * Pressure Change, Transient .3
0: Pressure Derivative
0- o Pressure Derivative
uf ~ ~~~~~~~ioCh~;~~Oii:;nsient 4 (If ~~~~;~~~~iOCh~~~;Oii:~nsient 4
§- 8 ~~~~~~rio~e8~~i~~tive
~ 10-'-1 "':;""::=======---'
§- 8 ~~~:~~fio~e8~~i~~five
~ 10-'~ L:::....:=======---_----'
()
Representative TranSient,
~,
Ql
.2: \ ........ Ql
>
:;:; ~ .
~ 10-
';::
Ql
2

. ~~""IIiIJIIIll,,--"";;".Ii~
~o ~@ C;;
.;::
Ql
10' • :20~~~~jII"'_"""_""_,,,,•• ./,,!,~,,~
o o
No Late-Time Derivative Separation
/
Separation in Late Time Due to Superposition Effects
D OD
o

10-' 10-' 10' 10' 10' 10-' 10-' 10' 10' 10'
lit, hours M, hours

Fig. 3-Current derivative formulation applied to simulated Fig. 6-Desuperposition derivatives for simulated data.
data.
209
10 EHLIG-ECONOMIDES, CA, AMBROSE, R. W. AND JOSEPH, J. A. SPE 20550

Test Sequence Comparative Dia nostic Plot

1.5000E+04 10-'-,----:-----.:~=.=:.:.:::...===:.;============;l
CJ Pressure Change, Trans. 1
o
L~
0
Pressure Derivative
6 Pressure Change, Trans.
0
c) Pressure Derivative
0
C) Pressure Change, Trans. 3

0.0
o
r- 0

""""""""'0
~
*
(l

+
Pressure Derivative
Pressure Change, Trons. 4
Pressure Derivative
x Pressure Change, Trans.
Pressure Derivative
5440.00

\Representative Transient (+)

o
5240.00 o

'~i'1J I I 1 I
o
Derivative Separation Due to Superposition Effects

4000.00f...-..-.---.r-~~1 10.0 17.5 20.0 22.5 25.0

2.5 5.0 7.5 12.5 15.0 10-'J,---"-~~....- - -....- - -....-~~....-~-....-----,"
Time, hr 10-' 10' 10' 10' 10'
llt, hours
Fig. 7-Test sequence for field data.
Fig. 1O-Current derivative formulation applied to log-
Dia nostic Plot - Transient 4
filtered field data,

10'
10,,-_ _---'--F:..:low Regime Identification Plot - Transient 5
I [lJ Pressure Chcmge
I0 PreSSLJr e Derivative
-Ui
0.
oS 10'
>
:g -Ui
>
·C 0. ,
Q) 10
o oS
>
:;:;
-0 10°
c o
o >
'C:
0. Q)
<J o
D
C
0 ,
[!J Pressure 10
oA Pressure
Pressure ~.~ ,~, c.".
0.
<J
Pressure
t Pressure

10" 10 C 10'
lit, hours

Fig. 8-Comparison showing effects of data processing

strategies on pressure and pressure derivative. lit, hr

Dia nos tic Plot - Transient 4

Fig. 11-Determination of late-time log-log derivative
slope.
10' 6 ~~~~~:~f:~~ g~~[~~f:~~: ~gD-~~.teL~g_somPle. No Smoothing
+ Convolution Derivative, 500 PL Log-Sample, 0.1 Smoothing

'Ui
0. ,
10
ui
0.
J
o
c'3
Q) 10'
,2:
-0
>
'C:
Q)
010'
~
0.
<J

10' 10' 10'

lit, hours

Fig. 9-Comparison showing effects of data processing

strategies on the convolution derivative.

210
SPE 20550 PRESSURE DESUPERPOSITION TECHNIQUE FOR IMPROVED LATE-TIME TRANSIENT DIAGNOSIS 11

Determination of Late-Time Slope on Specialized Plot 10."1 ~C:::o.:..:m.1:p::.:a::..r~a::.:ti.:.ve::._:D~i~ag~n~o::.:s::.:t~ic~P~1o::::t======:::;-

5420
8 r,~~~~~~~ g~;~r~67;vrrnns
6. Pressure
1
Chongt:, Trans. 2

.. C!J Pressure
C) Pressure
~ Pressure

(I
Pressure
Derivative
Change, Trons. j
Derivative
* Pressure Derivative
Change, Trans 4

5400 x Pressure Change. Trans ~

+ Pressure Derivative

'en 5380
0-

ei
Late- Time Slape = - 24.2 psi/hr v2 "0
C
5360 o

/ 0-
<J 10"
.. Little Late-Time Derivative Separation
5340

5320+--r----r--r-----r--.-----r--.,---~-~ 10'-J,----...-----r---.....,---~---...,....---...j
1
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 10" 10 ' 10' 10' 10' 10

(lit)'!'. hr
v2 L1t. hours

Fig. 12-Determination of late-time slope on specialized Fig. 13-Desuperposition derivatives for field data.
plot.

211