An Investigation of Recent Deconvolution

Methods for Well-Test Data Analysis

M. Onur, SPE, and M. Çınar,* SPE, Istanbul Technical University; D. Ilk, SPE, P.P. Valkó, SPE, and T.A. Blasingame,
SPE, Texas A&M University; and P.S. Hegeman, Schlumberger

Summary If such an objective could be achieved with some success, then,

as stated by Levitan, the deconvolved response would remove the
In this work, we present an investigation of recent deconvolution
constraints of conventional analysis techniques (Earlougher 1977;
methods proposed by von Schroeter et al. (2002, 2004), Levitan
Bourdet 2002) that have been built around the idea of applying a
(2005) and Levitan et al. (2006), and Ilk et al. (2006a, b). These
special time transformation [e.g., the logarithmic multirate super-
works offer new solution methods to the long-standing deconvo-
position time (Agarwal 1980)] to the test pressure data so that the
lution problem and make deconvolution a viable tool for well-test
pressure behavior observed during individual flow periods would
and production-data analysis. However, there exists no study pre-
be similar in some way to the constant-rate system response. As
senting an independent assessment of all these methods, revealing
also stated by Levitan, the superposition-time transform does not
and discussing specific features associated with the use of each
completely remove all effects of previous rate variations and often
method in a unified manner. The algorithms used in this study for
complicates test analysis because of residual superposition effects.
evaluating the von Schroeter et al. and Levitan methods represent
Unfortunately, deconvolution is an ill-posed inverse problem
our independent implementations of their methods based on the
and will usually not have a unique solution even in the absence of
material presented in their papers, not the original algorithms
noise in the data. Even if the solution is unique, it is quite sensitive
implemented by von Schroeter et al. and Levitan. Three synthetic
to noise in the data, meaning that small changes in input (measured
cases and one field case are considered for the investigation.
pressure and rate data) can lead to large changes in the output
Our results identify the key issues regarding the successful and
(deconvolved) result. Therefore, this ill-posed nature of the decon-
practical application of each method. In addition, we show that
volution problem combined with errors that are inherent in pres-
with proper care and attention in applying these methods, decon-
sure and rate data makes the application of deconvolution a chal-
volution can be an important tool for the analysis and interpreta-
lenge, particularly so in terms of developing robust deconvolution
tion of variable rate/pressure reservoir performance data.
algorithms which are error-tolerant. Although there exists a variety
of different deconvolution algorithms proposed in the past, only
Introduction those developed by von Schroeter et al., Levitan, and Ilk et al.
appear to offer the necessary robustness to make deconvolution a
Applying deconvolution for well-test and production data analysis viable tool for well-test and production data analysis. In this paper,
is important because it provides the equivalent constant rate/ our objectives are to conduct an investigation of these three de-
pressure response of the well/reservoir system affected by variable convolution methods and to establish the advantages and limita-
rates/pressures (von Schroeter et al. 2002, 2004; Levitan 2005; tions of each method.
Levitan et al. 2006; Ilk et al. 2006a, b; Kuchuk et al. 2005). With As stated in the abstract, the algorithms used in this study for
the implementation of permanent pressure and flow-rate measure- evaluating the von Schroeter et al. and Levitan methods represent
ment systems, the importance of deconvolution has increased be- our independent implementations based on the material presented
cause it is now possible to process the well test/production data in their papers; therefore, our implementations may not be identi-
simultaneously and obtain the underlying well/reservoir model (in cal to their versions. However, as is shown later, validation con-
the form of a constant rate pressure response). New methods of ducted on the simulated (test) data sets (von Schroeter et al. 2004;
analyzing well-test data in the form of a constant-rate drawdown Levitan 2005) sent to us directly by von Schroeter and Levitan
system response and production data in the form of a constant- shows that our implementations reproduce almost exactly the same
pressure rate system response have emerged with development of results generated by their original algorithms for these simulated
robust pressure/rate (von Schroeter et al. 2002, 2004; Levitan data sets.
2005; Levitan et al. 2006; Ilk et al. 2006a, b) and rate/pressure The paper is organized as follows: First, we describe the pres-
(Kuchuk et al. 2005) deconvolution algorithms. In this work, we sure/rate deconvolution model and error model considered in this
focus on the pressure/rate deconvolution for analyzing well-test data. work. Then, we provide the mathematical background of the von
For over a half century, pressure/rate deconvolution techniques Schroeter et al., Levitan, and Ilk et al. methods together with their
have been applied to well-test pressure and rate data as a means to specific features. We compare the performance of each method
obtain the constant-rate behavior of the system (Hutchinson and by considering three synthetic and one field well-test data sets.
Sikora 1959; Coats et al. 1964; Jargon and van Poollen 1965; Finally, we provide a discussion of our results obtained from
Kuchuk et al. 1990; Thompson and Reynolds 1986; Baygun et al. this investigation.
1997). A thorough review and list of the previous deconvolution
algorithms can be found in von Schroeter et al. (2004). The pri- Pressure/Rate Deconvolution Model
mary objective of applying pressure/rate deconvolution is to con-
vert the pressure data response from a variable-rate test or produc- The pressure/rate deconvolution model considered in this study [as
tion sequence into an equivalent pressure profile that would have well as in von Schroeter et al. (2002, 2004), Levitan (2005), Levi-
been obtained if the well were produced at a constant rate for the tan et al. (2006), and Ilk et al. (2006a, b)] is given by the well-
entire duration of the production history. known convolution integral (van Everdingen and Hurst 1949):
dpu共t − t⬘兲
兰 q 共t⬘兲
t
pm共t兲 = p0 − m dt⬘, . . . . . . . . . . . . . . . . . . . . (1)
0 dt
* Currently pursuing PhD degree at Stanford University.
where qm(t) and pm(t) are the measured flow rate and pressure at
Copyright © 2008 Society of Petroleum Engineers
any place in the wellbore including the wellhead or sandface, and
This paper (SPE 102575) was accepted for presentation at the 2006 SPE Annual Technical p0 is the initial pressure. In Eq. 1, pu is referred to as the drawdown
Conference and Exhibition, San Antonio, Texas, 24–27 September, and revised for publi-
cation. Original manuscript received for review 5 July 2006. Revised manuscript received 8
pressure response of the well/reservoir system if the well were
December 2007. Paper peer approved 12 December 2007. produced at a constant unit-rate. Thus, pu is the rate-normalized

226 June 2008 SPE Journal

pressure response in psi/(STB/D), and any reference rate other than volution in cases for which the expected error margins for rate and
unity can also be used for normalizing pu. Throughout this paper, pressure data are known or can be estimated.
pu is normalized by a constant rate of 1 STB/D. As to be discussed later, the three deconvolution methods each
Eq. 1 assumes that at the beginning of production, the wellbore/ have built-in flexibility to process the pressure for any subset of
reservoir system is in equilibrium and pressure is uniform through- flow and buildup periods in a test sequence by accounting for the
out the system at p0. In addition, Eq. 1 is an expression of the entire flow rate history from the start of the test to the end of the
principle of superposition (van Everdingen and Hurst 1949) that is flow or buildup period to be deconvolved. Note that only the
valid only for linear systems (e.g., single-phase flow of slightly deconvolution methods of von Schroeter et al. and Levitan have a
compressible fluid in porous media). built-in flexibility to allow the user to treat the rate and/or initial
It can be shown that Eq. 1 is valid if the wellbore-storage pressure as unknown in deconvolution; Ilk et al. does not have this
coefficient and the skin factor are constant during the entire history flexibility. Later, we will discuss the conditions under which es-
of pressure and rate measurements. If these constraints are satisfied timating rate or correcting measured rates and initial pressure by
by the test data, then pu(t) generated from Eq. 1 will represent the deconvolution of measured pressure data alone could be feasible.
constant unit-rate drawdown pressure response of the system in-
cluding the effects of the skin and wellbore-storage, Cw, which is Error Model. In this work, we compare the performance of the
caused by the wellbore volume below the point at which the rate three deconvolution methods based on Eq. 1 using simulated pres-
qm is measured (Kuchuk 1990). By considering simulated test data sure and rate data sets corrupted by noise. As in the von Schroeter
that have different wellbore-storage coefficients during different et al. work, we consider independent, identically distributed nor-
flow periods, Levitan demonstrates that Eq. 1 fails to produce a mal random errors with zero mean and specified standard devia-
physically meaningful response function pu(t), particularly during tions for pressure and rate data computed from, respectively,
the late-time portion of the response, when Eq. 1 is applied to the
pressures for the entire test sequence. Levitan (2005) and Levitan ␩p㛳⌬p㛳2
et al. (2006) also discuss other cases (e.g., interference effects and ␴p = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
commingled reservoirs) and note that Eq. 1 is not the correct
公M
superposition equation for such cases.
and
When dealing with inconsistent data sets, which are usually the
case in real test data, Eq. 1 should be used with caution (von
␩q㛳q㛳2
Schroeter et al. 2004; Levitan 2005). Here, inconsistent data may ␴q = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)
be simply defined as the data (acquired during the test sequence) 公N
that cannot be described by the deconvolution model of Eq. 1 (e.g.,
the data acquired under the situations where wellbore-storage co- Here, ⌬p represents the M-dimensional vector of true pressure
efficient and/or skin change, or multiphase flow effects occur near changes generated from a particular well/reservoir model, while
the wellbore during the test sequence). For inconsistent data sets, q represents the N-dimensional vector of true input flow rates.
Levitan and Levitan et al. suggest that a deconvolution algorithm (Note: throughout, a boldface lower case letter denotes a column
based on Eq. 1 should be applied by using the pressure data from vector, while a boldface capital letter denotes a matrix.)
an individual flow period(s) and by keeping rates and initial pres- 㛳x㛳2(⳱√x21+x22+ ⭈ ⭈ ⭈) is the l2-norm. The error levels as a fraction
sure fixed in each run. Typically, the individual flow period(s) is of the pressure change and rate are represented by ␩p and ␩q,
a pressure buildup portion of the test sequence. Levitan shows that respectively. For example, ␩p⳱␩q⳱0.05 corresponds to an error
when used in this mode, the algorithm produces reliable results for level of 5%. Given true pressure change and rate data, along with
pu(t), and hence deconvolution of test pressure data performed one specified ␩p and ␩q, Eqs. 2 and 3 are used to compute ␧ and ␦, the
flow period at a time provides for a consistency check. Based on M-dimensional and N-dimensional vectors of normal errors gen-
our understanding of his approach, Levitan seems to assume that erated for pressure change and rate data, respectively. These sets of
inconsistency (or quality) of rate data could be determined by random errors are added to the true pressure and rate data to
comparison of the constant unit-rate drawdown derivative curves generate the corrupted (or “measured”) pressure pm and rate qm
pertaining to each buildup period. His basic assumption is that if which will be used for deconvolution. That is,
the curves lay on top of each other, particularly at late times, then
the rates are consistent, unless the model changes during the test. pm = p + ␧, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4)
Should the curves not overlay, then one can think that there is a
problem with the rates, which has to be resolved before proceeding and
with deconvolution. However, it is not clear how such an approach
would work when both initial pressure and rate history could con- qm = q + ␦. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)
tain errors because Levitan et al. also show that initial pressure has
a profound effect on the late portion of deconvolved responses. This error generation process will provide the same standard de-
They recommend a trial-and-error procedure, wherein the initial viation of errors in both the absolute pressure pm and the pressure
pressure is treated as a constraint parameter, to determine initial change ⌬pm (⳱p0–pm). However, the specified error level (␩p) for
pressure from at least two buildup periods. An application of Levi- the pressure change data may not correspond to the same error
tan’s methodology to a synthetic test for which both initial and rate level in the pressure data pm because magnitudes of the pressure
data are uncertain can be found in the meeting version of this work and pressure change data may differ significantly.
[see Synthetic Example 3 (Çinar et al. 2006)]. One may wish to question the validity of using independent,
On the other hand, when applying the deconvolution model of identically distributed normal random errors with zero mean and
Eq. 1 to inconsistent data sets, von Schroeter et al. (2004) suggest specified standard deviations to model errors in both the pressure
a strategy based on applying deconvolution successively to in- and rate data. However, this is the most commonly used model to
creasing portions of the signal while monitoring rate and pressure interpret the errors in actual pressure and rate measurements. Also,
matches. In this strategy, rate history and/or initial pressure are the assumption of randomly distributed errors forms the basis of
treated as unknown in deconvolution. In our view, this strategy the all three methods considered in this study. However, we note
assumes that we have an a priori knowledge of expected error that the model may not be realistic in all cases and can lead to
margins in pressure and rate data, and most importantly, that such misleading results in deconvolution—particularly for tests where
a prior knowledge is correct. If so, then one can detect any incon- the rates are allocated based on a few measurements. It is possible,
sistency in rates as well as inconsistency between the data set and but beyond the scope of this paper, to use more general error
the deconvolution model by using a statistical evaluation of the models; for example, systematic and/or normal errors with zero
results. As will be shown with the example applications later, this mean and variable variance and with correlations (i.e., non-
strategy also proves useful for detecting inconsistencies in decon- diagonal covariance matrix for errors).

June 2008 SPE Journal 227

Description of Deconvolution N 㛳⌬pm㛳22
Methods Considered ␯def = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
M 㛳qm㛳22
Here, we present a description of each of the three pressure/rate
deconvolution methods. and

Von Schroeter et al. Deconvolution Method. The von Schroeter 㛳⌬pm㛳22

et al. method is based on three novel ideas: First, Eq. 1 is solved ␭def = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
M
not for the constant-unit-rate response pu(t), but for the z (or re-
sponse) function based on its derivate with respect to the natural In cases where the initial pressure p0 is known, ⌬pm is formed with
logarithm of time as defined by this p0. Otherwise, it is formed with the maximal pressure sample

冋 册 冋 册
in place of p0, as suggested by von Schroeter et al. In our appli-
dpu共t兲 dpu共␶兲 cations of the von Schroeter et al. method, we find that the default
z共␶兲 = ln = ln , . . . . . . . . . . . . . . . . . . . . . . . . (6)
d ln共t兲 d␶ value given by Eq. 9 for the relative error weight ␯ to be used in
where ␶⳱ln(t). The selection of z(␶) as a new solution variable Eq. 8 works fine in most cases.
ensures that the dpu(t)/d lnt is positive, a necessary condition that On the other hand, the regularization parameter ␭ has a strong
a constant-unit-rate response of the system should satisfy. This use influence on the z function. We found that the default value com-
of z(␶) converts Eq. 1 to a nonlinear convolution equation given by puted from Eq. 10 divided by the square of the Frobenius norm of
the curvature matrix D [see Eq. B-7 in von Schroeter et al. (2004)
兰
lnt
pm共t兲 = p0 − qm共t − e␶兲 ez共␶兲d␶. . . . . . . . . . . . . . . . . . . . . . . . (7) for the definition of the Frobenius norm] provides a good initial
−⬁
guess for determining an “optimal” level of regularization. This is
Second, a regularization based on the curvature of z(␶) is im- equivalent to the same level of regularization as in von Schroeter
posed to achieve some degree of smoothness of the solution z(␶) et al., where instead the curvature measure 㛳Dz−k㛳2 in Eq. 8 is
and to improve the conditioning of the deconvolution algorithm. normalized to unit Frobenius norm. Throughout this paper, we
Third, their formulation accounts for errors in both the rate and report ␭def as the ␭ value computed from Eq. 10 divided by the
pressure data by considering a special least-squares formulation square of the Frobenius norm of D, i.e.,
known as the Total Least-Squares (TLS) problem (von Schroeter
et al. 2002, 2004; Golub et al. 1999; Björck 1996). Within the TLS 1 㛳⌬pm㛳22
concept, the deconvolution problem can be posed as an uncon- ␭ˆ def = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
M 㛳D㛳2F
strained nonlinear minimization for which the objective function is
defined as (von Schroeter et al. 2002, 2004) whereas von Schroeter et al. report ␭def as the one computed from
Eq. 10. Thus, the difference is only in reporting.
E共p0, z, y兲 = 㛳p0e − pm − C共z兲y㛳22 + ␯㛳y − qm㛳22 + ␭㛳Dz − k㛳22 Similar to von Schroeter et al.’s suggestion, we start by using
. . . . . . . . . . . . . . . . . . . . . . . . . . . (8) a ␭ value equal to the value of ␭ˆ def computed from Eq. 11. As von
The arguments given in the parentheses of the objective function E Schroeter et al. (2004) state, determining the optimal value of ␭ in
represent the unknowns to be determined by minimizing E. As this way requires a trial-and-error procedure and is subjective.
with von Schroeter et al., we minimize the objective function of Similar to von Schroeter et al., we find that for a fixed number of
Eq. 8 with the variable projection algorithm given by Björck nodes, the optimal value of ␭ depends on well/reservoir model and
(1996). This algorithm is based on the separable least-squares on the error levels in the pressure and rate data.
problem and requires minimization of two (one linear, other non- A disadvantage that may be noted about the von Schroeter et al.
linear) least-squares objective functions. For the solution of these method is their assumption regarding the behavior of the z function
least-squares problems, we implemented the singular value decom- before the first node. They assume that the wellbore-storage unit-
position algorithm called SVDCMP found in Press et al. (1986). slope line is valid for the behavior of the z response before/at the
When updating the solution vector in Eq. 8, we use the line search first grid node. As is shown in “Synthetic Example 1” section, in
algorithm called LINMIN found in Press et al. (1986). We use the cases where this assumption is violated, the deconvolved response
same strategies as von Schroeter et al. for choosing grid nodes, at early times can be in error. As discussed in the next two sec-
starting values of p0, z, and y in the nonlinear minimization as well tions, the Levitan and Ilk et al. methods remove this restriction.
as the same termination criterion. The deconvolution method of von Schroeter et al. gives the user
In Eq. 8, y represents the N-dimensional column vector of an option to treat not only the z-response as unknown in decon-
computed (or true, unobserved) rates, and C is the M×N dimen- volution, but also treat the rate history and/or initial pressure as
sional matrix of response coefficients, D is the (L−1)×L constant unknown. (The same option is also true for the Levitan method,
matrix in curvature measure. z is the L-dimensional vector of re- but not for the Ilk et al. method.)
sponse coefficients, and k is the (L−1)-dimensional vector in cur- Unlike Levitan and Ilk et al., von Schroeter et al. (2002, 2004)
vature measure as given by kT⳱(1, 0, . . . , 0). The analytical provide a concrete error analysis based on Gaussian statistics and
expressions for the elements of the M×N matrix C are derived by show how to compute such statistics as a posteriori estimates of
approximating the flow rate qm(t) with stepwise constant-rate func- standard deviation of pressure and rate errors, and 95% confidence
tions [as prescribed in von Schroeter et al. (2002, 2004)]. The intervals for the estimates. Inspecting such statistics with the avail-
elements of the (L−1)×L constant matrix D are given by Eqs. B-6a able knowledge of expected error margins in pressure and rate data
and B-6b in von Schroeter et al. (2004). The solution zi in Eq. 8 is are indeed very useful in evaluating the validity of the results
approximated as a piecewise-linear function on a grid of points ␶i, obtained from deconvolution. In our implementation of the von
i⳱1, 2, . . . , L, usually with a uniform constant grid increment ⌬␶ Schroeter et al. method (and Levitan method), we also compute
[see Eq. 63 of von Schroeter et al. (2004)]. As suggested by von such statistics and show their utility in some of the synthetic and
Schroeter et al., we use 40 grid nodes and choose the first node as field data cases to be considered later.
10−3 hours and the last node as the total test duration, unless We also note that in the work of von Schroeter et al. the focus
otherwise stated. Furthermore, as consistent with the material pre- is the reconstruction of the derivative of the constant-unit-rate
sented by von Schroeter et al., L−1 curvature constraint equations pressure (i.e., z(t)) with respect to the logarithm of time (see Eq. 6).
are used in Eq. 8, and the wellbore-storage unit-slope line assump- They do not comment how to reconstruct the constant-unit-rate
tion at/before the first node is implemented. pressure response pu(t) from the estimated z response.
The ␯ and ␭ appearing in the right side of Eq. 8 are the relative Our results show that their method can be used to reconstruct
error weight and regularization parameter (von Schroeter et al. pu(t) from the estimated z response by use of the discretized ver-
2002, 2004), respectively. Providing a comprehensive error analy- sion of Eq. 7 (with all qms replaced by unity; see Eqs. 12 and 13)
sis based on Gaussian statistics, von Schroeter et al. recommend in cases where wellbore-storage effects exist at early-time portions
that the default values of these two parameters be computed from of the data. In cases where there is minimal or no wellbore-storage

228 June 2008 SPE Journal

effects, our results indicate that we can still reconstruct pu(t), but Schroeter et al. method because the integration limits in Eq. 7 and
in such cases, the z value estimated for the first grid node (i.e., z1 Eq. 15 are different. ỹ in Eq. 16 is an M-dimensional column vector
at t1) interestingly contains a “Dirac delta” component which rep- of flow rates at the times where the pressure measurements, pm, are
resents the skin effect. In other words, z has a discontinuity at the made. The elements of ỹ are also included in the elements of the
first node in the von Schroeter et al. method; estimated ez1 is equal unknown flow-rate vector of y. ␵p, ␵q, and ␵c represent the error
to the correct value of the pu(t) function at t1, but not to the correct bounds (or scale parameters) for the pressure, rate and curvature
value of the dpu(t)/d lnt function at t1, and estimated ez at the other constraints (Levitan 2005).
nodes at the very early times suffer from the “left-end” effects for Like Levitan, we use the algorithm for unconstrained minimi-
approximately a log-cycle due to the unit-slope line assumption zation given by Dennis and Schnabel (1983) to minimize the ob-
before/at the first grid node. This point will be further examined in jective function of Eq. 16. The Levitan method considers pu(t1) as
the “Synthetic Example 1” section. an unknown, in addition to the unknowns p0, y, and z to be esti-
An equation to construct the pu(t) response for the von Schroe- mated by minimization of Eq. 16. We use the same strategies
ter et al. method can be obtained by setting qm(t–e␶)⳱1 STB/D suggested by Levitan for choosing starting values of pu(t1), p0, z,
and replacing p0–pm(t) by pu(t) in Eq. 7 to represent the unit-rate and y in nonlinear minimization of Eq. 16, and use 70 uniformly
pressure change. The result is spaced nodes, as he suggested.
lnt
However, there are several points that are not stated or revealed

兰e z共␶兲
by Levitan, regarding specific implementation of his method based
pu共t兲 = d␶. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12) on Eqs. 15 and 16. We believe that our observations given here
−⬁ will be useful to those readers interested in implementing the Levi-
Assuming piecewise linear representation of the z function, and a tan method.
wellbore-storage unit-slope trend at/before the first node, we can The first point is related to the selection of the time for the first
derive the following discrete form of Eq. 12: node, t1. In his paper, Levitan does not describe how to choose the
value of t1 to be used for constructing the z-response. He only

冉 冊 states that if t1 is sufficiently small, the model pressure can be

i
共ezj − ezj−1兲
兺
tj
pu共ti兲 = ez1 + ln ; for i = 1, 2, . . . , L, represented as Eq. 15. In our applications, we usually choose t1 to
j=2 共zj − zj−1兲 tj−1 be less or equal to the minimum elapsed time (relative to the start
. . . . . . . . . . . . . . . . . . . . . . . . . . (13) of its flow period) of any point in the pressure signal. In the
“Synthetic Example 1” section, we also examine the consequences
where ti represent the times at which the z function is computed, of choosing small and large values of t1.
and L represents the number of nodes. Secondly, it is not clear from Levitan’s paper how the last term
Thus, after the z function is estimated by minimizing the ob- (or curvature constraint part) in the right side of the objective
jective function given by Eq. 8, Eq. 13 can be used to construct the function given by Eq. 16 is treated, particularly for the first node.
pu(t) function. If the assumption of a wellbore-storage unit-slope In his paper, Levitan states that he uses L−1 curvature constraints,
trend at/before the first node is satisfied, then it can be shown that where L is the number of nodes of the grid ␶i, and he states that he
uses the model prediction of a curvature constraint as given by von
dpu共t1兲
ez1 = = pu共t1兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (14) Schroeter et al. (2004) [see Eq. 9 of Levitan (2005)]. For the
d lnt purpose of discussion, here we recall Levitan’s Eq. 9:
However, as mentioned previously, the estimated ez1 from the von 1
Schroeter et al. encoding for the data sets with minimal or no ⍀ci共z兲 = 共z − 2zi + zi+1兲. . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
wellbore effect represents pu(t1), the unit-rate pressure drop at the ⌬␶ i−1
first node including the skin effect, if it exists, but not the value of
When i⳱1, we have z0 in this equation, and there is no such z0 in
dpu(t1)/d lnt.
his formulation. Therefore, if he is treating the curvature con-
straints as the same as von Schroeter et al. (2004), as stated in his
Levitan Deconvolution Method. Levitan’s method is based on
paper, then when i⳱1, we would expect that he uses [see also Eq.
the same concepts as proposed by von Schroeter et al. However,
B-6a of von Schroeter et al. (2004)]
Levitan’s method differs in two respects.
First, Levitan uses a deconvolution equation which removes the 1
restriction of the von Schroeter et al. assumption that the wellbore- ⍀c1共z兲 = 共z − z 兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
storage unit-slope line is valid before the first node. Levitan’s ⌬␶ 2 1
encoding for pressure/rate deconvolution equation assumes that the
In this paper, unless otherwise stated, we will present the results
time corresponding to the first grid node ␶1 (or equivalently t1) is
based on the implementation of Eq. 18 for i⳱1, and Eq. 17 for
sufficiently small so that Eq. 7 can be accurately approximated by
i⳱2, 3, . . . , L−1 because we believe this is consistent with the
lnt material presented in Levitan’s paper. There are other alternatives
pm共t兲 = p0 − qm共t兲pu共t1兲 − 兰 q 共t − e 兲 e
␶1
m
␶ z共␶兲
d␶. . . . . . . . . . . . . (15) that one can use for the treatment of the curvature constraint term
in Eq. 16. For example, one can use Eq. 17 with L−2 curvature
constraints by starting with i⳱2, 3, . . . , L−1. As is shown later,
Second, Levitan uses an unconstrained, nonlinear weighted we often find that the latter option provides a slightly better rep-
least-squares objective function involving the sum of three mis- resentation of the reconstructed z-response at early times than the
match terms for pressure, rate, and curvature: former option.
In addition, Levitan does not specifically show how he con-
㛳 㛳
2
1 p0e − pm − pu共t1兲ỹ − C̃共z兲y structs the pu(t) function. He only states that pu(t) can be recon-
E共p0, pu共t1兲, z, y兲 =
2 ␵p 2
structed from the estimated z by integrating Eq. 6 with the value of
pu(t1), estimated by minimizing the objective function given by
㛳
1 y − qm
㛳 㛳 㛳 . . . . . . . . . . . . . . . . (16)
2 2
1 Dz Eq. 16. There are several ways to perform the integration of Eq. 6
+ +
2 ␵q 2
2 ␵c 2
with the values of pu(t1), z1, z2, . . . , zL estimated from nonlinear
regression (e.g., composite trapezoidal rule of integration, and the
where C̃ is the M×N dimensional matrix of response coefficients rectangular rule of integration). Our point is that Levitan does not
for which analytical expressions can be derived by approximating specifically show how he reconstructs pu(t) from Eq. 6.
the flow rate qm(t) using stepwise constant-rate functions and are In our implementation of his method, we reconstruct the pu(t)
given in von Schroeter et al. (2002, 2004). Note that the matrix C̃ function by setting qm(t)⳱qm(t-e␶)⳱1 STB/D in Eq. 15, and re-
in the Levitan method is not identical to the matrix C in the von arranging the resulting equation to obtain:

June 2008 SPE Journal 229

 lnt level in data is not solely dependent on the sensor/tool resolution
pu共t兲 = pu共t1兲 + 兰e
␶1
z共␶兲
d␶. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (19) and measurement-error characteristics; there could be other types
of errors that cause the spread (variance or standard deviation) of
pressure and rate data to be larger than that caused by sensor/tool
Assuming piecewise linear representation of the z function, we can resolution or error characteristics alone. Therefore, a single choice
derive the following discrete form of Eq. 19: for ␵c (⳱0.05) together with the gauge resolution for ␵p (⳱0.01

冉 冊
i psi) could not be expected to work for all data sets, and it may be
共ezj − ezj−1兲
兺 共z − z
tj desirable to use different values of ␵p, ␵q, and ␵c than those sug-
pu共ti兲 = pu共t1兲 + ln ; for i = 1, 2, . . . , L.
j=2 j j−1 兲 tj−1 gested by Levitan, depending on the specific data quality and tool
characteristics.
. . . . . . . . . . . . . . . . . . . . . . . . . . (20) Finally, we make a comment regarding the relationship of ␵p,
Comparison of Eqs. 13 and 20 indicates that the ez1 used to ␵q, and ␵c in the Levitan method and ␯ and ␭ in the von Schroeter
construct the pu(t) function in the von Schroeter et al. method (Eq. et al. method. Ignoring the differences of encoding (see Eqs. 7 and
13) can be thought as the pu(t1) used to construct pu(t) function in 15), and assuming that errors in pressures and rates are indepen-
Levitan’s method (Eq. 20). However, recall that in Levitan’s dently, identically distributed normal random variables with zero
method, pu(t1) is an unknown model parameter to be determined mean and variances equal to ␵2p and ␵2q, respectively, it can be
by minimizing the objective function given by Eq. 16, while ez1 is shown that the objective functions are equivalent if we define
computed after z1 is determined by minimizing the objective func- ␯⳱(␵p/␵q)2 and ␭⳱(␵p/␵c)2. Minimizing E given by Eq. 16 is the
tion given by Eq. 16 or Eq. 8. It is guaranteed that ez1 to be used same as minimizing any constant multiple of it, so the necessary
in Eq. 13 will always be positive. However, there is no guarantee parameters are ␵p, ␵p/␵q, and ␵p/␵c; hence, the number of param-
for pu(t1) to be positive in Levitan’s method because it is treated as eters is exactly the same in both methods. Our results indicate that
an unknown model parameter to be estimated by unconstrained both objective functions yield almost identical results if these ra-
minimization of the objective function given by Eq. 16. Thus, it is tios are kept the same in both objective functions.
possible that in some cases, pu(t1) could converge to a negative We can make the following observations based on the expres-
value, as happens for our Synthetic Example 2 and field example sions ␯⳱(␵p/␵q)2 and ␭⳱(␵p/␵c)2. For example, in cases where we
cases to be discussed later. do not have a prior knowledge of data quality, the default values
In cases where pressures pertaining to only buildup portions are of ␯ and ␭ computed from the equations presented by von
matched to reconstruct the constant unit-rate drawdown responses, Schoreter et al. (see Eqs. 9 and 11) can be used in the expressions
it can be shown that in the Levitan encoding of Eq. 15, the sen- ␯⳱(␵p/␵q)2 and ␭⳱(␵p/␵c)2 to determine the default values for ␵p
sitivity of the model pressure with respect to pu(t1) depends on and ␵q to be used in the Levitan objective function (Eq. 16) with
only the flow rate value at the instant of shut-in. To show this, we a presumed value of ␵c. For example, if we compute ␯⳱0.01
differentiate Eq. 15 with respect to pu(t1) to obtain: (psi.d/stb)2, and ␭⳱0.04 psi2 from Eqs. 9 and 11 for a given data
set, then for a presumed value of ␵c⳱0.05, the expressions will
dpm共t兲 give ␵p⳱0.01 psi and ␵q⳱0.1 stb/d. (Note that the Levitan’s sug-
= −qm共t兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (21) gested values of ␵c⳱0.05 and ␵p⳱0.01 psi give ␭⳱0.04 psi.2) On
dpu共t1兲
the other hand, in cases where we have a prior knowledge of
Therefore, if we apply deconvolution to only buildup pressures standard deviations of errors in rate and pressure data (i.e., we
acquired at t b1, t b2, ⭈ ⭈ ⭈ , t N
b
b
, where t b1 represent the time for the know the values of ␵p and ␵q a priori), then the expressions ␯⳱(␵p/
instant of shut-in, then qm(t) in Eq. 21 will be all zero, except at the ␵q)2 and ␭⳱(␵p/␵c)2 with a presumed value of ␵c and with the
time for the instant of shut-in. As a result, the sensitivity of pm to known values of ␵p and ␵q can be used to estimate the default
pu(t1) is only nonzero at the instant of shut-in, i.e., values of ␯ and ␭ to be used in the von Schroeter et al. objective

再
function (Eq. 8).
dpm共t兲 −qm共t b1兲, for t = t b1
= . . . . . . . . . . . . . . . . . . . . . . . (22)
dpu共t1兲 0, for t ⬎ t b1 Validation of Our Algorithms Based on the von Schroeter et al.
and Levitan Methods. Here, we validate our implementations of
This shows that the sensitivity of model pressure to pu(t1) depends these authors’ methods using the synthetic data provided to the
on only a single flow-rate information at the instant of shut-in, first author of the paper by von Schroeter and Levitan, through
when processing the pressures for a buildup period. personal communication.
Our investigation indicates that an improper selection of t1 and First, we compare the results generated from our implementa-
of curvature constraint ␵c, and/or magnitude of errors in pressure/ tion with those provided by von Schroeter. The data set used for
rate data and time synchronization errors in rate and pressure data this comparison is the “Simulated Example” of von Schroeter et al.
are possible causes of driving an unconstraint minimization algo- (2004) for the case where pressure and rate data contain 5% and
rithm to a negative value for pu(t1). A possible way to avoid this 10% errors (i.e., ␩p=0.05 and ␩q=0.10), respectively. The simu-
is to use a constraint when minimizing Eq. 16 in the form of a lated model is for a vertical well with wellbore-storage and skin
one-sided inequality constraint, such as pu(t1)>0. For example, the effects, near a sealing fault, and all quantities in this example are
imaging procedure suggested by Carvalho et al. (1996) can be used dimensionless (von Schroeter et al. 2004). In generating the results
to achieve this. shown in Table 1 for estimated rates and Fig. 1 for estimated
Regarding the error bound (or scale) parameters ␵p, ␵q, and ␵c unit-rate responses, we used his parameters supplied, i.e., t1⳱2,
in Eq. 16, Levitan states that these are not free parameters that can ␯⳱44.8676, and ␭⳱␭ˆ def⳱1.21351 (which is equal 424.868 com-
be manipulated at will to control least-squares minimization, and puted from Eq. 10 divided by the Frobenius norm, which is
that these parameters are based on prior knowledge of data quality. 350.063 for this example). The number of nodes is 21, and initial
Based on his experience, he suggests the use of ␵p⳱0.01 psi pressure is fixed at its true value of 60. We have an excellent
(which is a typical value for the resolution of quartz gauges) and agreement (correct up to three significant figures) between data
␵c⳱0.05 in Eq. 16, but does not provide any default value to be generated from our algorithm and data provided by von Schroeter.
used for the rate error bound ␵q. Levitan states that choosing the The pu(t) curve shown in Fig. 1 represents our estimated values
values of the error bounds for rates is not straightforward and generated by Eq. 13 because von Schroeter did not provide us the
recommends that larger values of ␵q be assigned to the flow peri- pu(t) data. For this data set, the deconvolved derivative response
ods where we expect large uncertainty in flow rate data. (Fig. 1) deviates considerably from the true behavior [see Fig. 2
We agree that ␵p and ␵q should be determined based on knowl- of von Schroeter et al. (2004)] at early and late times. The reason
edge of data quality; however, we also note that it is usually for the odd behaviors seen at the deconvolved derivative response
difficult to have a correct prior knowledge of error levels in pres- at early and late times is mainly caused by use of the default val-
sure and particularly in rate data. This difficulty is because error ues of the parameters ␯ and ␭ in Eq. 8, which appear to be not

230 June 2008 SPE Journal

appropriate for reconstructing the true derivative response for this use of a numerical derivative routine based on the Bourdet method
data set. (1989) without smoothing. In Fig. 2, we have also reported the
Next, we compare the results generated from our implementa- value of pu(t1) obtained from our implementation and Levitan’s
tion of Levitan method with those provided by Levitan. The data value. As can be seen, our results are in excellent agreement with
set used for this comparison is “Example 1” of Levitan (2005). The Levitan for his simulated “Example 1.”
simulated model is for a vertical well with wellbore-storage and Based on all the results presented here, we conclude that our
skin effects, located in the middle of a channel with two parallel independent implementations of the von Schroeter et al. and Levi-
no-flow boundaries. Pressure and rate data are free of errors. The tan methods reproduce almost exactly the same results provided to
data sent to us only included unit-rate pressure, not the unit-rate us by von Schroeter and Levitan. We do not claim that our imple-
logarithmic derivative responses. He also did not provide us with mentations of the methods are identical to those of the authors;
the values of ␵p, ␵q, ␵c, and t1 (time value for the first node), but his however, on the basis of the material presented in their papers and
tabulated values of unit-rate pressures started from a time value of the data sets provided to us, we believe we have validated our
0.01 h. We used t1⳱0.01 h, ␵p⳱0.01 psi, ␵c⳱0.05, and 70 nodes. implementations of the algorithms.
We kept rates and initial pressure fixed, and used all pressure data
for the entire test sequence to generate the unit-rate responses from Ilk et al. Deconvolution Method. The Ilk et al. (2006a) method
our algorithm. A comparison of the results is shown in Fig. 2. The considers the following form of the convolution integral:
constant unit-rate drawdown derivate response, dpu/d lnt, shown as t
dpu共t⬘兲
兰 q共t − t⬘兲
solid circular data points in Fig. 2 for Levitan, was generated by
⌬p共t兲 = dt⬘ . . . . . . . . . . . . . . . . . . . . . . . . . (23)
dt
0

Fig. 1—Comparison of unit-rate responses from our implemen-

tation of von Schroeter et al. (2004) method with those gener-
ated by von Schroeter for “Simulated Example” considered in Fig. 2—Comparison of unit-rate responses from our implemen-
von Schroeter et al. (2004) for the case with ␩p=0.05 and tation of Levitan’s method (2005) with those generated by Levi-
␩q=0.10. tan for his “Example 1” in his paper (2005).

June 2008 SPE Journal 231

and uses quadratic (second-order) B-splines, denoted by B2 where W is the M×M diagonal weighting matrix with diagonal
throughout, with logarithmically distributed knots to represent the elements equal to wj, for j⳱1, 2, . . . , M, and E(c) is the objective
unknown dpu(t)/dt, simply denoted by p⬘u(t), function. The solution of Eq. 31 can be given by:
K ĉ = 共XTWTWX兲−1XTWTW⌬pm. . . . . . . . . . . . . . . . . . . . . . . . . . (32)
p⬘u共t兲 = 兺
i=l
ciB2i 共t兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (24)
Here, a few remarks are in order regarding the weighted least-
2
squares objective function of Ilk et al., where the weights are
B (t) is the second order B-spline being nonzero only in the inter-
i defined by Eq. 30. If we view the least-squares only as a pure
val bi<t<bi+3 for i⳱0, ±1, ±2, . . . , where b is the basis; typically curve-fitting method, it may be reasonable to define the weights as
Ilk et al. use b⳱1.6. If l is the index of the first knot and K is the the reciprocal of the observed pressure drop values. In this case,
index of the last knot, then there are K−l+1 B-splines involved in the solution may even provide reasonable estimates of the constant
the interpolation of the unknown p⬘u(t) response function. Inserting unit-rate responses. However, when the objective function is
Eq. 24 in Eq. 23, the following equation is obtained: viewed within the concept of the maximum likelihood method of

冋 册
t K
estimation based on Gaussian statistics, then the statistics of the

兰 兺 c B 共t⬘兲
0 i=l
2
i i q共t − t⬘兲dt⬘ = ⌬p共t兲, . . . . . . . . . . . . . . . . . . . (25)
estimated parameters will be correct and the estimates would be
optimal only if the weights are defined as the inverse of standard
deviation of errors in the data [see, for example, Bard (1974)].
or, equivalently, Therefore, we do not expect that using a weighting matrix based on

冋 册
Eq. 30 as proposed by Ilk et al. should yield correct statistics (e.g.
t
K 95% confidence intervals for the unit-rate responses), even though
兺 兰 B 共t⬘兲q共t − t⬘兲dt⬘
i=l
ci
0
2
i = ⌬p共t兲, . . . . . . . . . . . . . . . . . . (26) the resulting estimates of the unit-rate response may be reasonable.
However, we were not able to verify this issue because the version
of the algorithm that we used in this study does not provide a
where c is the (K−l+1)-dimensional vector of unknown coefficients. statistical analysis of the results.
Selecting the most appropriate basis for logarithmic distribu- The solution ĉ from Eq. 32 does not provide sufficient regu-
tion of B-splines is very influential on the deconvolved response larization as the noise level increases in the data. Hence, for their
functions (Ilk 2005). In other words by selecting the basis, the regularization approach, Ilk et al. consider the following two con-
number of B-splines is set—the number of B-splines has a direct ditions per each spline interval:

冋冉 兺 冊 冉 兺 冊 册
effect on the resolution of the desired function. Fewer B-splines
K K
can result in over-smoothing and introduce bias; greater B-splines
can result in oscillations in the deconvolved response functions, t ciB2i 共t兲 − t ciB2i 共t兲 = 0 . . . . . . . (33)
especially in well-test derivative functions. In addition, Ilk et al. i=l t=bk i=l t=bk+1/2
note that the earliest part of the unit-rate derivative response is and

冋冉 兺 冊 冉 兺 冊 册
critical, especially if the observed variable-rate response contains
a skin effect (but minimal or no wellbore-storage effects) because K K

in such cases the unit-rate response contains a Dirac delta compo- t ciB2i 共t兲 − t ciB2i 共t兲 = 0, . . . . . (34)
nent. To reconcile this condition, they add four additional “anchor” i=l t=bk+1/2 i=l t=bk+1
B-splines to the left of the first observed time point in the signal for k⳱l,l+1,. . . ,K–1. The regularization conditions (Eqs. 33 and
(Ilk 2006a). 34) require that the value of the logarithmic derivative of the
If there are M pressure-change observations collected in the constant-rate response differ only “slightly” between the knot and
M-dimensional vector ⌬pm that were observed at times (t1, t2, . . . , the middle location and can be represented as
tM), then the problem can be translated into an over-determined
system (i.e., the number of measurements are greater than the Xrc = 0, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (35)
number of sought B-spline coefficients, Mkn) of linear equations as: where Xr is the (2K–2l)×n regularization matrix. Then, by intro-
Xc = ⌬pm, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (27) ducing the regularization parameter ␣, Ilk et al. append the system
of regularization equations (Eq. 35) to the system of equations
where X is the M×n sensitivity matrix (or “design” matrix). [Note given by Eq. 29 to obtain a new system of equations which may be
that n (⳱K–l+1) represents the number of B-splines.] Its elements written as
are defined as:
X*c = ⌬p*m, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (36)
tj

Xj,i = 兰 B 共t⬘兲q共t − t⬘兲dt⬘.

2
i j . . . . . . . . . . . . . . . . . . . . . . . . . . . . (28)
where X* is the (M+2K–2l)×n augmented matrix obtained by join-
ing the M×n matrix(1−␣) WX and the (2K−2l)×n matrix␣ Xr, and
0
⌬p*m is a M+(2K−2l)-dimensional vector obtained by joining the
As mentioned before, Eq. 27 is an over-determined linear system M-dimensional vector (1−␣)W⌬pm and the (2K−2l)-dimensional
of equations and can be solved in least-squares sense. It is also (zero) vector 0.
possible to solve the system of equations in weighted least-squares Solving Eq. 36 in a least-squares sense is equivalent to mini-
sense. For this case, by introducing a weight matrix W, we modify mizing the following objective function:
Eq. 27 as:
Ẽ共c兲 = 共1 − ␣兲2E共c兲 + ␣2cTXTr Xrc, . . . . . . . . . . . . . . . . . . . . . . (37)
WXc = W⌬pm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (29) where E(c) is the objective function given by Eq. 31. The solution
Ilk et al. usually prefer a diagonal weighting matrix, W, with the of Eq. 37 can be given by:
diagonal elements defined by: ĉ = X+a ⌬pm, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (38)
wj = 1 Ⲑ ⌬pm,j. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (30) where the n×M matrix X+a represents the pseudoinverse of the
In other words, the reciprocal of measured pressure drop is used— augmented matrix and is given by
it is assumed that the responses are corrupted by a constant level X+a = 共1 − ␣兲2关共1 − ␣兲2XTWTWX + ␣2XTr Xr兴−1XTWTW. . . . . . (39)
of relative error. In this case, the goal of B-spline deconvolution is
to find c that minimizes the following weighted least-squares ob- The over-determined system (Eq. 38) can be solved after comput-
jective function: ing the pseudoinverse X+a by singular value decomposition with a
default cut-off for the smallest singular value. Further details can
E共c兲 = 共Xc − ⌬pm兲TWTW共Xc − ⌬pm兲, . . . . . . . . . . . . . . . . . . . (31) be found in Ilk (2005).

232 June 2008 SPE Journal

 We note that the solution ĉ given by Eq. 38 will not satisfy Eqs. functions. However, the rates can also be modeled in a constant
33 and 34 exactly because the system of equations (Eq. 36) is step-wise fashion. In this case, the rate function is given by
over-determined. The influence of Eqs. 33 and 34 on the solution N
will depend on the size of the regularization parameter ␣. If ␣ is set
to zero, there is no regularization, and the solution (Eq. 38) reduces q共t兲 = 兺 关q − q
k=1
k k−1兴␪共t − t qk 兲, . . . . . . . . . . . . . . . . . . . . . . . . . (44)
to the solution given by Eq. 32 for the no-regularization case. In
the presence of random noise and/or other inconsistencies, a posi- and can be solved in the real-time domain without Laplace trans-
tive ␣ is selected based on an informal interpretation of the dis- formation using the superposition principle and the analytic inte-
crepancy principle—i.e., the value of the regularization parameter gral of quadratic B-splines. However, for “smoothly” changing
is increased until the calculated (model) pressure difference begins rates Ilk et al. prefer Eq. 43.
to deviate from the observed pressure difference in a specific man- Application of the Ilk et al. method requires not only the mea-
ner. The mean and standard deviation of the arithmetic difference sured rate and pressure data, but also specification of the starting
of the computed and input pressure functions are also computed. times of the rate segments, t qk , the basis for the spline-knots, b, the
Increasing ␣ to the point where the problem is over-regularized regularization parameter, ␣, and an estimate of the initial reservoir
will yield a solution that is oversmooth and dominated by bias. pressure p0.
By using the c coefficients obtained by minimizing Eq. 37 (or An advantage of the method is that it can treat variable rate
equivalently solving Eq. 38), the constant-unit-rate pressure and its without using step-wise approximation. This can be useful, par-
logarithmic derivative responses are reconstructed by the follow- ticularly when dealing with continuously measured bottomhole
ing equations, respectively: flow rates. When a profile of piecewise constant surface rates like
in a well-test sequence is considered, there is little benefit from
K

兺 ĉ B
using the Ilk et al. method, and such data can be equally well
pu共t兲 = i i,int共t 兲,
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (40) processed by the von Schroeter et al. and Levitan methods.
i=l
One of the disadvantages is that a variable rate profile must be
and dissected into continuous segments. Also, the method as presented
K
here does not allow for the estimation of initial pressure and rates.
dpu共t兲
d lnt
=t 兺 ĉ B 共t兲,
i=l
2
i i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (41)
In addition, unlike the deconvolution methods of von Schroeter
et al. and Levitan, the method cannot ensure positivity of the
logarithmic derivative of the constant-unit-rate pressure response
where ĉi, for i⳱l, l+1, . . . , K, represents the coefficients estimated of the system. Using regularization based on Eqs. 33 and 34 is an
by minimization, and the integral of the second order B-spline, indirect attempt to satisfy the positivity condition as well as pro-
B2i,int(t), is obtained analytically. vide smoothness in the response function p⬘u(t). In certain cases, we
In contrast to the von Schroeter et al. and Levitan methods, the observed negative values of p⬘u(t) (Eq. 24) and hence, the logarith-
Ilk et al. method does not assume that the rate function is given in mic derivative values (Eq. 41) were negative.
a constant step-wise manner. Instead, Ilk et al. allow for a general
rate function whose Laplace transform, q(s), is analytically de- Comparisons of Methods With Simulated
fined. The rate profile is not required to be an analytic function, but Test Data
segments of the rate profile are represented by individual functions To test the proposed deconvolution methods, we consider three
which can be easily transformed into the Laplace domain. synthetic (simulated) test examples. The examples are defined as
The elements of the matrix X (Eq. 28) are calculated by us- follows:
ing numerical inverse Laplace transformation (here denoted 1. A simulated well-test example for an infinite-conductivity
by L−1[F(s)][t]): fracture well without wellbore-storage effects in a reservoir
bounded by two parallel (no-flow) boundaries.
Xj,i = L−1关q共s兲B2i 共s兲兴 关tj兴. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (42)
2. A simulated well-test example [Example 2 from Levitan
Here, B2i (s) is the Laplace transform of the ith B-spline and it is (2005)], where a different wellbore-storage coefficient is modeled
known in closed form. The B-splines with indices i for which bi>tj for the final pressure buildup sequence.
are excluded, because those splines start later than the time of the 3. A simulated well-test example for which we have con-
given observation. The success of this approach relies heavily on tinuous measurements of flow rates.
the accuracy of the numerical Laplace transform inversion. As Synthetic Examples 1 and 3 were generated by the Saphir soft-
such, the multi-precision GWR algorithm (Abate and Valkó 2004) ware (2005).
is used to obtain the elements of the matrix X.
To obtain the Laplace transform of the rate, the Ilk et al. Synthetic Example 1. As noted previously, the von Schroeter
method considers various options. The one that they prefer is to et al. method is based on the assumption that the wellbore-storage
dissect the rate into N segments with starting times t qk , k⳱1, . . . , unit-slope line is valid for the behavior of the z response before the
N. In each segment they describe the incremental contribution by first grid node. In our experience, the wellbore-storage unit-slope
an exponential term and hence the rate is obtained in the form trend at the start of a transient is rare. Either the storage is poorly

冋 冉 冊册
defined (because of small storage, or data acquisition not being fast
N
t − t qk
兺
enough), or the storage is changing. Therefore, here, we investi-
q共t兲 = a qk + bqk exp − ␪共t − t qk 兲, . . . . . . . . . . . (43) gate the consequences of using the von Schroeter et al. method on
k=1 cqk
data with no unit-slope trend at the very early time, in comparison
where ␪(t) is the Heaviside (unit-step) function. The parameters for with the use of the Levitan and Ilk et al. methods, which remove
each segment are found by a linear least-squares fit combined with the unit-slope line assumption. For this investigation, we consider
“direct search” on the nonlinear parameter c kq. [Here, “direct a simulated well-test example (see Fig. 3 for pressure/rate data) for
search” is related with the optimization algorithm to find the op- an infinite-conductivity fracture well without wellbore-storage ef-
timum value of the function which is called the “golden section fects in a reservoir bounded by two parallel (no-flow) boundaries
search” [see Cheney and Kincaid (2003)]. This is a sequential (see Table 2 for input model parameters). Pressure data were
search method which makes use of previous information to locate generated by assuming a gauge with a resolution of 0.01 psi, and
subsequent experiments. A reasonable range for this time constant rate history does not contain any errors. The data set is consistent
is between t1 and tM. The fewer segments used the more smooth with the deconvolution model of Eq. 1.
the resulting rate approximations will be. A shut-in period should The constant unit-rate drawdown responses generated based on
be always included as an individual segment, and it is possible to matching the entire pressure history for all periods from our imple-
represent the shut-in period with exponentials. In simple terms, the mentations of the von Schroeter et al. and Levitan methods are
Ilk et al. rate approximation is the superposition of exponential shown in Fig. 4a. The parameters used for the von Schroeter et al.

June 2008 SPE Journal 233

 Fig. 3—Pressure and rate data for Synthetic Example 1.

method are ␯⳱1 (psi.d/stb)2 and ␭⳱0.04 psi2, and for Levitan are
␵p⳱0.01 psi, ␵q⳱0.1 stb/d, and ␵c⳱0.05. The minimum elapsed
Fig. 4a shows that our implementation based on the von Schroe-
time of any point in the pressure signal is 10−3 h. Hence, for both
ter et al. method does not correctly reconstruct the logarithmic
methods, we have selected this value of elapsed time as the value
derivative of the constant unit-rate drawdown response for the first
of time for the first node (i.e., t1⳱10−3 h), and 70 nodes were used.
log-cycle. However, the effect of assuming a wellbore-storage
unit-slope trend at/before the first node for the von Schroeter et al.
method disappears fast, and this assumption does not cause any
artifacts in the late-time portion of the derivative response. Al-
though it may not be readily apparent from Fig. 4a, our algorithm
for the von Schroeter et al. method yields the same values for both
pu(t) and dpu/d lnt functions at the first node [i.e., dpu(t1)/d
lnt⳱pu(t1)⳱ez1], as expected. Most interestingly, the estimated ez1
is equal to the correct value of the constant unit-rate drawdown
pressure drop including the skin effect at t1 [i.e., pu(t1)], but not
equal to the correct value of the logarithmic derivative of the
unit-rate response at t1 [i.e., dpu(t1)/d lnt]. Therefore, there is a jump
(or discontinuity) at the first node for the von Schroeter et al. method.
Fig. 4a also shows two implementations for the Levitan
method; the first with L−1 curvature constraints (based on Eq. 18
for the first node and Eq. 17 for the remaining nodes), and the
second with L−2 constraints (based on Eq. 17 for all nodes). Both
cases reconstruct the unit-rate responses quite accurately for all
times, though the implementation based on L−2 curvature con-
straints gives slighlty better results.
Fig. 4b shows the results obtained from the Ilk et al. method.
For this method, four additional “anchor” B-splines were added to
the left of the first observed time point in the signal. In Fig. 4b, we
also show the results for the case if the “anchor” B-splines are not
added, although Ilk et al. suggest that the anchor splines be added
for all data sets with/without storage effects (Ilk et al. 2006a). The
total number of B-spline knots is 37 for the case without “anchor”
B-splines. No regularization was considered (i.e., ␣⳱0). Fig. 4b
indicates that the Ilk et al. method with “anchor” B-splines also
constructs the unit-rate responses quite accurately for all times.
Next, we investigate the effect of the value of t1 on deconvo-
lution for our implementations of the von Schroeter et al. and
Levitan methods. We consider two different values of t1; Fig. 5
shows the results for t1⳱1×10−4 h, and Fig. 6 shows results for
t1⳱1×10−2 h. The unit-rate responses shown for the von Schroeter
et al. and Levitan method were generated by using the same values
of the “weight” parameters that were used to generate the results in
Fig. 4a. From the results of Figs. 5 and 6, we conclude that t1 has
an effect on the early time portion of the unit-rate derivative re-
sponses generated from both methods. For this example, the effect
is more pronounced (approximately a log-cycle) on the unit-rate
derivative response generated from the von Schroeter et al.
method. Levitan’s method with L−2 curvature constraints pro-
Fig. 4—Comparison of unit-rate responses generated (with duces the correct unit-rate responses for all times if t1 is chosen
t1=10−3 h) from our implementations for (a) the von Schroeter smaller than the minimum elapsed time in the pressure signal.
et al. and Levitan methods, and (b) for the Ilk et al. method for Thus, we conclude that the effect of t1 is only on the very early
Synthetic Example 1. data, approximately a log-cycle for both methods and recommend

234 June 2008 SPE Journal

Fig. 5—Comparison of unit-rate responses generated with Fig. 6—Comparison of unit-rate responses generated with
t1=10−4 h for Synthetic Example 1. t1=10−2 h for Synthetic Example 1.

70, respectively, while the total number of B-spline knots used for
that t1 be chosen equal or less than the minimum elapsed time in
the Ilk et al. algorithm is 21. The same t1 value of 10−2 h, which
the signal.
is equal to the minimal elapsed time in the pressure signal, was
used for the three methods. Initial pressure and rates were fixed at
Synthetic Example 2. Pressure and rate data for this simulated test
their true values in deconvolution. The default values for ␯ and ␭
example are shown in Fig. 7. The test data were provided by
computed from Eqs. 9 and 11 for the von Schroeter et al. objective
Levitan (through a personal communication) and do not contain
function (Eq. 8) are ␯def⳱0.0085 (psi.d/stb)2 and ␭ˆ def⳱2.0643
any errors. The simulation used two different values of the well-
psi2 for the first buildup period, and ␯def⳱0.0046 (psi.d/bbl)2 and
bore-storage coefficient; Cw⳱0.05 RB/psi for the first three peri-
␭ˆ def⳱1.3576 psi2 for the second buildup period. The default val-
ods and 0.001 RB/psi for the last buildup period. The reservoir is
ues of ␵p⳱0.01 psi and ␵c⳱0.05 recommended by Levitan were
constrained by two parallel no-flow boundaries (Levitan 2005).
used in Eq. 16. Regularization was not used for the Ilk et al.
The data set constitutes an example of inconsistent test data with
algorithm (i.e., ␣⳱0 in Eq. 37) because pressure and rate data do
the deconvolution model of Eq. 1—the individual flow periods in
not contain any noise.
the example are consistent with the deconvolution model of Eq. 1,
As shown in Figs. 8 and 9, each algorithm gives almost iden-
but jointly they are not. Our objective is to test the performances
tical results for the pu(t) function and its logarithmic derivative for
of the three algorithms for this inconsistent data set and also test
both buildup periods. All methods work well with their default
the ability of the algorithms when applied to process pressure data
parameters (because we applied them to the periods where data are
pertaining to only individual flow periods.
consistent with the deconvolution model of Eq. 1) and are free of
Figs. 8 and 9 present comparisons of the results of deconvo-
errors. Although it is not shown, when we applied the deconvolu-
lution obtained from the first buildup (50 to 100 hr) and the second
tion algorithms to process the pressure data for the entire test
buildup (200 to 250 hr) periods by processing them separately with
sequence at once (with keeping initial pressure and rates at their
the three deconvolution algorithms. For each algorithm, the flow
true values in deconvolution), all of the algorithms failed to pro-
rates and initial pressure are fixed at their given true values. There-
duce the correct solution, and we obtained a very similar response
fore, this actually represents the strategy proposed by Levitan et al.
function to that of Levitan [as shown in Fig. 5 of Levitan (2005)].
in deconvolution. The recommended default values of the required
parameters for each algorithm are used. The total number of nodes
used for the von Schroeter et al., and Levitan algorithms is 40 and

Fig. 8—Comparison of constant-unit-rate pressure responses

Fig. 7—Pressure and rate data for Synthetic Example 2; Ex- obtained by deconvolution of the first buildup period; Synthetic
ample 2 of Levitan (2005). Example 2.

June 2008 SPE Journal 235

 With the application of this strategy, we are also able to detect that
there is indeed inconsistency between the data for the second
buildup and the deconvolution model because all deconvolved
derivative responses, except the one from the start to the end of the
test (i.e., from 0.01 to 250 h), show the same well/reservoir be-
havior (Fig. 10). The pressure match obtained by deconvolution of
all pressure data from the start to the end of the test is not accept-
able (Fig. 11). The pressure matches obtained by deconvolution of
other portions of the pressure data were perfect (not shown). Es-
timated initial pressures are 4,999.816 psi by deconvolution of data
from 0.01 to 50 h; 4,999.986 psi by deconvolution of data from
0.01 to 100 h; 4,999.991 psi by deconvolution of data from 0.01 to
200 h; and 5,044.511 psi by deconvolution of data from 0.01 to
250 h, compared to the true initial pressure of 5,000 psi. Estimated
flow rates for the first and second flow periods are almost the same
as the input values of 1,000 STB/D by deconvolution of all por-
tions of the pressure data, except by deconvolution of pressure data
from 0.01 to 250 h. For that case, the estimated flow rates for the
first and second flow periods are 1143.68 and 793.46 STB/D,
respectively. Thus, the estimated initial pressure and estimated
flow rates differ significantly from the true values if the pressure
data for the second buildup are included in deconvolution. In sum-
Fig. 9—Comparison of constant-unit-rate pressure responses mary, these results show that for this example, the von Schroeter et al.
obtained by deconvolution of the second buildup period; Syn- strategy also detects the inconsistency between data and the decon-
thetic Example 2. volution model by checking the pressure and rate matches as well as
the unit-rate responses obtained for different subsets of the data.
These results verify that when all three algorithms are applied to
subsets of the pressure/rate data that are consistent with the de- Synthetic Example 3. The true pressure and rate data as well as
convolution model of Eq. 1, the algorithms produce meaningful their noisy versions with different error levels for this simulated
results. In addition, the results show that for this test example, the test example are shown in Fig. 12. The model is for an unfractured
Levitan et al. strategy detects the inconsistency between data and vertical well in a double-porosity reservoir with a no-flow circular
the deconvolution model because of changing wellbore-storage boundary and including wellbore-storage and skin effects. The
coefficient. input parameters are given in Table 3. The true production rate as
We also applied the von Schroeter et al. and Levitan algorithms a function of time was produced with the following formula:
successively to increasing portions of the signal, but treating initial
pressure and rates as unknown and monitoring pressure and rate q共t兲 = 共350 − 300e−t Ⲑ 20兲␪共t兲 + 共−350 + 350e−共t−100兲 Ⲑ 50兲␪共t − 100兲
matches. Note that this is the strategy proposed by von Schroeter + 共300e−t Ⲑ 20 − 350e−共t−100兲 Ⲑ 50兲␪共t − 150兲. . . . . . . . . . . . . . (45)
et al. (i.e., jointly estimate initial pressure and rates together with
the unit-rate derivative response). In this application, we use the We consider continuous measurements of flow rate data in this
von Schroeter et al. algorithm with ␯ and ␭ at their default values example so that we can investigate the performance of the decon-
(not reported here) computed from Eqs. 9 and 11 by using the volution algorithms based on constant stepwise approximation of
maximal pressure value of 4,992.419 psi. This value of initial flow rates. In addition, we consider different levels of normally
pressure was also taken as the initial guess for the unknown initial distributed errors generated from Eqs. 2 and 3 and added to flow
pressure; the true (or given) flow rates were taken as initial guess rate and pressure data to investigate the robustness of the algo-
for the flow rates. rithms. The standard deviations of errors in pressure data with
Fig. 10 shows derivative responses deconvolved over increas- 0.5%, 2%, and 5% error levels are ␴p⳱10.42, 41.88, and 101.69
ing portions of the data set from the start to the end of the test.

Fig. 10—Comparison of constant-unit-rate derivative responses Fig. 11—Pressure match obtained by deconvolution of data
deconvolved (using the von Schroeter et al. algorithm) over from 0.01 to 250 h, using the von Schroeter et al. algorithm;
increasing portions of the data set; Synthetic Example 2. Synthetic Example 2.

236 June 2008 SPE Journal

Fig. 12—Pressure and rate data for Synthetic Example 3; both
true and corrupted data sets of pressure and rate. We apply the deconvolution algorithms to process the pressure
data for the whole test sequence in one pass. In our applications
psi, respectively, while the standard deviations of errors in with the von Schroeter et al. and Levitan methods, we only fix the
rate data with 1% and 5% error levels are ␴q⳱2.44 and 11.24 initial pressure at its true value of 8,000 psi. For the Ilk et al.
STB/D, respectively. algorithm, in addition to fixing the initial pressure, we fix the rate
Fig. 13 presents a comparison of the true flow rate history and history approximated by fitting piecewise exponentials to the noisy
the constant stepwise approximations used to represent these data rate data because flow rate cannot be treated as unknown. We set
for application in the von Schroeter et al. and Levitan algorithms. t1⳱10−3 h, the minimal elapsed time in the pressure signal, for the
Although it may not be clear from Fig. 13, we have approximated von Schroeter et al. and Levitan methods; we start to construct
the continuous flow rate history by 125 piecewise constant steps. B-splines with four “anchors” added to the left of this t1 in the Ilk
Noisy rate data containing 1% and 5% random errors were simi- et al. method.
larly approximated for the algorithms. In the Ilk et al. algorithm, Our objective is to investigate the robustness of the three meth-
piecewise exponential rates (Eq. 43) were used, and the parameters ods by considering pressure and rate data sets that are contami-
for each segment were found by linear least-squares fit, as ex- nated with different levels of Gaussian errors. In Figs. 15 through 17,
plained previously. For example, for rate data containing 5% ran- we show the deconvolution results generated for three different
dom noise, we use the following formula: combinations of error imposed on the input rate and pressure data.
For the von Schroeter et al. (Fig. 15) and the Levitan (Fig. 16)
q共t兲 = 共349.7 − 300.8e−t Ⲑ 20兲␪共t兲 methods, piecewise constant rates with errors are used and treated
as unknowns in the deconvolution process. For the Ilk et al.
+ 共−339.7 + 336.8e−共t−100兲 Ⲑ 48兲␪共t − 100兲 method (Fig. 17), the rate history is fixed and fitted using piece-
wise exponential functions (e.g., see Eq. 46 for 5% error in rate
+ 共−10 − 118.7e−共t−150兲 Ⲑ 48兲␪共t − 150兲, . . . . . . . . . . . . . . . . (46)
data, and Fig. 14).
for 0<tⱕ300. Fig. 14 shows a comparison of the rate data con- For each method we have developed a “best fit” using the
taining 5% error level with the rate data approximated from Eq. 46. regularization mechanism for a particular algorithm. The control
Although not reported here, our piecewise constant and exponen- and regularization parameters are presented in Table 4. The ␯-
tial representations of the continuous flow rate with or without parameters given in Table 4 for the von Schroeter et al. method
noise do honor the cumulative production during the test period represent the default values computed from Eq. 9. The ␭-
reasonably well.

Fig. 14—Comparison of flow-rate data with 5% error level with

Fig. 13—Comparison of true flow rate with its piecewise con- its piecewise exponential approximation given by Eq. 46, Syn-
stant approximation, Synthetic Example 3. thetic Example 3.

June 2008 SPE Journal 237

Fig. 15—Deconvolved constant-unit-rate pressure responses Fig. 16—Deconvolved constant-unit-rate pressure responses
generated by the von Schroeter et al. method for Synthetic Ex- generated by the Levitan method for Synthetic Example 3.
ample 3.
data and use these values for ␵p and ␵q because this is a synthetic
parameters represent 10␭ˆ def (with ␭ˆ def computed from Eq. 11) example. However, how this could be achieved in practice is an
values. Thus, we actually use the strategy recommended by von important question.
Schroeter et al. (i.e., for fixed ␯, start with a positive default level The results of Figs. 15 through 17 indicate that the von Schroe-
␭ˆ def and multiply it by powers of 10). In Fig. 15 we present our ter et al. and Levitan methods perform better than the Ilk et al.
results only for 10␭ˆ def because this value provides the best results method for this simulated example, particularly when the level of
for this strategy. error in rate/pressure data increases. Fortunately, each method ap-
The results shown in Fig. 16 for the Levitan method were pears, from this synthetic example to be competent (i.e., robust) for
generated by setting ␵p and ␵q equal to the standard deviations of practical applications in well-test and production data.
errors in pressure and rate data for each data set considered, and Fig. 18 shows reconstructed pressure signals obtained from all
adjusting the value of ␵c to find its “optimal” value giving the best three methods together with the given “measured” pressure for
unit-rate response curve. We note that the value of ␵c⳱0.05 sug- data set 3 (i.e., 5% error levels in p and q) and the true (clean)
gested by Levitan together with the ␵p and ␵q values given in Table 4 pressure signal. Fig. 19 shows reconstructed rate signals obtained
did not produce smooth unit-rate responses for any of the three from the von Schroeter et al. and Levitan methods (recall in the Ilk
data sets considered in the experiments. In addition, for data set 3 et al. method rate is fixed, so it is not shown in Fig. 19) together
(5% error levels in p and q), using ␵p⳱102 psi, ␵q⳱11.2 STB/D, with the “measured” rate for this data set, and the true (clean) rate
and ␵c⳱0.05 yielded a negative pu(t1) value (not shown here). We signal. As can be seen, reconstructed pressures from all three
have not varied the values of ␵p and ␵q in these experiments be- methods agree well with the true (clean) pressures. The average
cause in view of the maximum likelihood concept (Bard 1974), root-mean-square (rms) errors for the pressure match for the von
these parameters in the weighted least-squares objective function Schroeter et al., Levitan, and Ilk et al. methods are 94.1, 93.1, and
considered by Levitan (Eq. 16) should represent the standard de- 100.2 psi, respectively, which are close to the true standard devia-
viations of error in pressure and rate data, respectively. We can tion of errors ␴p⳱101.7 psi for “measured” pressure data contain-
certainly compute the standard deviation of errors in pressure and ing 5% error level. Similarly, the average rms errors for rate match
for the von Schroeter et al. and Levitan methods are 7.7 and 8.2
STB/D, respectively, which are close to the true standard deviation
of errors (␴q⳱11.24 STB/D) for “measured” rate data containing
5% error level.
Field Example Application
In this section, we consider one field well-test example to illustrate
application of the deconvolution methods under investigation.
The pressure and rate for this example are presented in Fig. 20.
The test consists of four distinct step-rate changes and a shut-in
acquired over a 16-hr period from a geothermal well producing
from a liquid-dominated, highly permeable fractured/faulted res-
ervoir system. The flow in the reservoir is single-phase water with
a salinity of approximately 5000 ppm, and a dissolved CO2 content
approximately 0.4% by weight [see Onur et al. (2007) for details].
In the field, the reservoir thickness is variable and not known very
well, and the permeability-thickness product (kh) ranges from 100
Darcy-m to 1900 Darcy-m (Onur et al. 2007). The well is vertical.
The pressure was measured with a downhole quartz gauge, posi-
tioned at the middle of the open interval, with a specified resolu-
tion of 0.01 psi. The rate measurements were taken at the surface
using a weir, and were accounted for the steam phase at the sur-
face. The temperature was also recorded downhole and was nearly
Fig. 17—Deconvolved constant-unit-rate pressure responses constant at 107.8°C throughout the test. In the well, we had also
generated by the Ilk et al. method for Synthetic Example 3. run a static survey and a flowing survey (conducted with a rate

238 June 2008 SPE Journal

equal to the rate for the fourth flow period shown in Fig. 20) of shown in Fig. 21, derivative data for the flow periods are much
pressure/temperature vs. depth, and these surveys indicated that noisier than those for the buildup period, and derivative data for
there was no gas or steam phase at the gauge location. some of the flow periods are not interpretable.
The measured initial pressure is approximately 280 psi, but Given the issues stated previously and considering that we have
there is an uncertainty in the value. The conventional (multirate) only one buildup period, first we process the pressures for the
test analysis of each flow period (Fig. 21) gives an indication of an entire test sequence in one pass to estimate the initial pressure and
inconsistent test example with changing wellbore-storage effects rates jointly with the unit-rate derivative response functions. In a
(owing to nonisothermal and multiphase flow effects in the well- sense, this is the strategy proposed by von Schroeter et al. How-
bore) within each period as well as non-Darcy flow (i.e., a rate- ever, we must realize that in this approach, we may end up ad-
dependent skin profile). The non-Darcy flow is typical of the geo- justing the rates incorrectly because of rate-dependent skin for the
thermal system under consideration and results from a high- sake of matching (or honoring) the entire measured pressure data.
permeability fracture network intersecting the wells. Based on the For this application, we use our algorithm based on the von Schroe-
results of Fig. 21, we also suspect that measured rates for some ter et al. method. The maximal pressure in the test sequence of
periods may not be accurate (e.g., the derivative responses at late 279.872 psi is used as an estimate of initial reservoir pressure. The
times for the third and fourth flow periods do not overlay with the minimal elapsed time in the test sequence of 2.78×10−3 h is used
derivative responses for the second flow and buildup periods). for the first node (t1) to reconstruct constant unit-rate drawdown
Otherwise, we would expect rate-normalized derivatives for all responses. For this value of initial pressure and t1, the default
periods to overlay because they should be independent of rate- values of ␯ and ␭ are ␯ def ⳱3.16×10 −8 (psi.d/stb) 2 and
dependent skin after storage effects die out. Although the same ␭ˆ def⳱1.92×10−3 psi.2
level of smoothing was applied for each of the derivative responses

Fig. 18—Comparison of reconstructed pressure together with Fig. 19—Comparison of reconstructed rate together with “mea-
“measured” (containing 5% error level) and true (clean) rate; sured” (containing 5% error level) and true (clean) rate; Syn-
Synthetic Example 3. thetic Example 3.

June 2008 SPE Journal 239

 Fig. 20—Pressure and rate data; field example.

Fig. 22 shows pressure derivative estimates obtained for five

different sets of ␯ and ␭—in four of the sets, ␭ is increased from
␭ˆ def by successive powers of 10, leaving ␯ at its default value, Fig. 21—Conventional rate-normalized pressure change and de-
while in one of the sets, we increase ␯ from ␯def by factor of 10, rivatives for each flow period in the test sequence; field ex-
leaving ␭ at its default value. (The conventional rate-normalized ample.
buildup derivative based on the superposition of measured rates is
also shown in the same figure for comparison.) For the sets where
␯ is fixed at its default value and ␭⳱10n×␭ˆ def for n⳱0, 1, 2, 3, because the derivative responses are based on different rate histo-
estimates for initial pressure are almost identical, ranging from ries—the conventional buildup derivate is based on measured
280.081 to 280.087 psi. Also, rms errors for the pressure and rate rates, while the derivative estimate for ␯=␯def and ␭⳱101×␭ˆ def is
matches are almost identical; rms for pressure matches ranging based on the adjusted rates shown in Fig. 23. For now, we post-
from 0.071 from 0.080 psi, and rms for rate matches ranging from pone our discussion of this issue.)
2,684 to 2,688 STB/D, as n increases from 0 to 3. For the set with Thus, we choose the derivative estimate for ␯=␯def and
␯⳱101×␯def, and ␭⳱␭ˆ def, the estimated initial pressure is 280.40 ␭⳱101×␭ˆ def for further analysis. The rate and pressure matches
psi, and rms errors for pressure and rate matches are 0.084 psi and obtained with this choice are shown in Figs. 23 and 24, respec-
2004 STB/D, respectively. These results indicate that increasing ␯ tively. The pressure match has an rms error of 0.072 psi and looks
with ␭ fixed at its default has little effect on the pressure and rate excellent; however, the rms error is higher than the gauge resolu-
matches. The derivative estimate for the set with ␯=␯def and tion of 0.01 psi. The estimated initial pressure is 280.081 psi with
␭⳱␭ˆ def is not interpretable because of its oscillations, and also the a 95% confidence interval of ±0.049 psi, which indicates small
derivative estimate for the set with ␯⳱101×␯def, and ␭⳱␭ˆ def uncertainty in initial pressure and that the true value may be inside
shows an odd behavior after 1 h. The derivative estimate for ␯=␯def the interval [2.80.032, 280.130 psi]. We do not have a prior knowl-
and ␭⳱101×␭ˆ def best matches the conventional buildup derivative edge of error level in rate data, so it is difficult for us to judge
for times until approximately 3 h, though both are based on dif- whether the rate match shown in Fig. 23 (with an average rms error
ferent rate histories. (We find this good agreement to be puzzling of 2691.2 STB/D) is acceptable: The rates for the first two flow
periods are changed by approximately 30%, which is quite large,
while the rates for the third and fourth flow periods are changed by
approximately 11%. The 95% confidence intervals for the esti-
mated rates are quite narrow, indicating small uncertainty in the
estimated rates. However, none of the measured rates lie within the

Fig. 22—Comparison of reconstructed unit-rate derivative re-

sponses by the von Schroeter et al. method, by processing the
pressures for the entire test sequence in one pass with initial
pressure and rates as unknown for various values of ␯ and ␭; Fig. 23—Rate match for ␯=␯def and ␭=101×␭ˆ def and 95% confi-
field example. dence intervals for the estimated rates; field example.

240 June 2008 SPE Journal

 Fig. 25—Comparison of reconstructed unit-rate derivative re-
Fig. 24—Pressure matches obtained by deconvolution of pres- sponses from deconvolutions based on pressures for second,
sures for the entire test sequence with unknown initial pressure third, and buildup periods with initial pressure fixed at 280.081
and rates, and for the buildup pressures alone with fixed initial psi and rates fixed at their estimated values shown in Fig. 23;
pressure of 280.081 psi and measured rates. Both deconvolu- deconvolutions were performed using the von Schroeter et al.
tions were performed using the von Schroeter et al. method method with for ␯=␯def and ␭=101×␭ˆ def; field example.
with for ␯=␯def and ␭=101×␭ˆ def; field example.
deconvolution based on individual flow periods helps us to detect
95% confidence intervals. This indicates the rate errors cannot be the inconsistency. However, deconvolution does not help us to
characterized as normally distributed random variables, which is prove or disprove whether the initial pressure and particularly flow
one of the basic assumptions of the deconvolution model. Based on rates estimated simultaneously with the constant unit-rate draw-
these results and the fact that rate-dependent skin and changing down derivative response by deconvolution of all measured pres-
wellbore-storage effects exist in the test, the data set cannot be sures are correct.
consistent with the deconvolution model of Eq. 1. Fig. 26 presents further interesting results for this field ex-
Next, we consider deconvolutions by using the pressure data ample. In Fig. 26, we compare the estimated derivative responses
from an individual flow period(s) and by keeping rates and initial obtained from deconvolutions of buildup pressures alone with
pressure fixed in each run. For this purpose, we perform decon- measured and estimated rates. For all cases, in addition to rates,
volutions by using the pressure data from the second and third flow initial pressure is also fixed at 280.081 psi. The conventional rate-
and buildup periods separately. In a sense, this strategy can be normalized buildup derivatives based on the superposition time of
viewed as the strategy proposed by Levitan for inconsistent data measured and estimated rates are also shown in the same figure for
sets, though we process pressure data for the flow periods too. comparison. The results of Fig. 26 indicate that all derivative re-
Here, we will use the rate history (Fig. 23) and initial pressure of sponses are reasonably consistent with each other, and that the
280.081 psi estimated previously by processing pressures for the choice of rate history does not have a significant influence on the
entire test sequence in deconvolution with ␯=␯def and ␭⳱101×␭ˆ def deconvolved unit-rate and conventional rate-normalized deriva-
to examine whether the inconsistency detected by the conventional tives from buildup pressures. Particularly, it is quite puzzling to us
multirate analysis (Fig. 21) was caused by the incorrect measured why the estimated derivative response obtained by processing the
rates or the model changes. From now on, estimated rates are pressures for the entire test sequence (all four flow periods and
referred to those shown as gray lines in Fig. 23. If deconvolutions buildup period) by treating initial pressure and rates as unknown in
based on these individual periods with the estimated rates produce deconvolution agrees well with the estimated derivative response
unit-rate drawdown derivative responses that are consistent with
each other at least at late times, then the data and deconvolution
model may be said to be consistent, and the inconsistency observed
in Fig. 21 is caused by incorrect measured rate history. Otherwise,
the test data and the deconvolution model are inconsistent mainly
because of the model changes in each period. For each deconvo-
lution, we use our algorithm based on the von Schroeter et al.
method with ␯=␯def and ␭⳱101×␭ˆ def, where ␯def and ␭ˆ def were
computed by using the pressure data and rate history for the period
under consideration—their values change depending on the period
considered in deconvolution. (We also performed deconvolutions
with ␯=␯def and ␭⳱␭ˆ def, but the results were similar to those
obtained with ␯=␯def and ␭⳱101×␭ˆ def.) Fig. 25 shows a compari-
son of the deconvolution results for the three periods. The esti-
mated derivatives for the second and third flow periods are very
different from that for the buildup period. Hence, the same decon-
volution model cannot be valid for these flow periods and the
buildup period. This is not surprising to us because we have
nonisothermal and multiphase effects (steam and noncondensable
gas phases above its flashing point) in the wellbore reflected as a
Fig. 26—Comparison of reconstructed unit-rate derivative re-
complicated wellbore-storage phenomena, as well as rate- sponses from various deconvolutions with rates fixed at their
dependent skin. Hence, we do not expect the pressure transient estimated (in Fig. 23) and measured (Fig. 20) values with initial
behavior of a flow period to be consistent with that of a buildup pressure fixed at 280.081 psi; deconvolutions were performed
period even if the initial pressure and rate history were known with using the von Schroeter et al. method with for ␯=␯def and
certainty or estimated correctly. In summary, for this example, ␭=101×␭ˆ def; field example.

June 2008 SPE Journal 241

obtained by processing only the pressures for the buildup period in
deconvolution with initial pressure fixed at 280.081 psi and rates
fixed at their measured values.
In an attempt to clarify this point, we generated a pseudo-field
test sequence with the flow rate history shown in Fig. 20 and
rate-dependent skin and wellbore-storage effects—wellbore-
storage coefficient is constant and the same for each period. For
the purpose of our discussion, we did not add errors to rate history.
Our results not shown here indicate that processing the pressures
for the entire test sequence at once (or also subsets including flow
periods) by jointly estimating initial pressure and rates in decon-
volution produces correctly the constant unit-rate drawdown de-
rivative response and the initial pressure. However, the deconvo-
lution algorithm adjusts flow rate history (actually incorrectly) for
the sake of providing a perfect match of the measured pressure
history, similar to the pressure match obtained in Fig. 24 for the
field example (solid curve in Fig. 24). Processing pressures for the
buildup period with rates and initial pressure fixed at their true
values also produces the constant unit-rate drawdown derivatives
correctly, but the pressure match for the flow periods is not good,
which should be expected because of rate-dependent skin, as simi-
lar to the pressure match obtained in Fig. 24 for the field example negative value* when pu(t1) is treated as an unknown model pa-
(dashed curve). rameter in our implementation of the Levitan method. As dis-
In summary, we admit that the field example considered here is cussed in detail previously, when pu(t1) is treated as an unknown
complex in the sense that pressures include rate-dependent skin model parameter in our implementation of his method, the resulting
and changing wellbore-storage effects; additionally, measured value may be negative because the minimization is unconstrained.
rates contain errors that do not follow Gaussian statistics and our The prior geologic model (not shown here) indicates that the
lack of information on expected error margins on rate data. How- well is nearly centered between two parallel faults, but no infor-
ever, on the basis of our deconvolution analyses of this data and mation exists whether the faults are no-flow or conductive. The
the pseudo-field example with rate-dependent skin, we believe that deconvolved unit-rate response indicates that these faults may be
estimated initial pressure of 280.081 is reasonably correct and that no-flow because of a well-defined 1/2 slope line after 1 hour (Fig.
the unit-rate derivative response derived from buildup pressures 27). The deconvolved derivative response at early times from
with this initial pressure represents the actual behavior of the geo- 0.003 to 0.2 hr gives an indication of a partially or fully penetrating
thermal reservoir system under consideration. well with changing wellbore-storage effect. In addition, the –1
Next, we compare the performance of each deconvolution slope line in the time interval from 0.3 to 1 hr exhibited by the
method in generating the constant unit-rate drawdown response deconvolved derivative response may indicate that the well is near
from buildup pressures alone, using the initial pressure value of a highly conductive fault or permeability-thickness product in-
280.081 psi and the measured flow rate history as fixed in decon- creasing away from the well. Unfortunately, we do not have the
volution. The deconvolved pressure drop and derivative responses analytical models in our catalogue to account for all these flow
obtained from the von Schroeter et al., Levitan, and Ilk et al. regimes together. Therefore, we consider rather a simpler model
methods are shown in Fig. 27, and the deconvolution results are based on a partially-penetrating well producing [with changing
compared to the conventional pressure buildup derivative. The wellbore-storage (Hegeman et al. 1993) and rate-dependent skin
algorithmic parameters for deconvolution are again given in Table 5. effects] between two no-flow parallel faults at least to honor the
In this case, for our implementation of the Levitan method, we 1/2 slope line observed after 1 hour on the deconvolved derivative
fixed pu(t1) in Eq. 16 by using the value of the deconvolved unit- response (Fig. 27). Fig. 28 presents the match of the unit-rate
rate derivative response estimated at the first node from our imple- derivative response derived from deconvolution of the buildup
mentation of the von Schroeter et al. method. This action was pressures, while Fig. 29 presents the match of the buildup pressure
necessary, because for this application, we estimated pu(t1) to be a change and its derivative (based on superposition time). Fig. 30
presents a comparison of the pressure data generated from this
model with the test pressure data during the whole test sequence.
The results shown in Figs. 28 through 30 are based on measured
rates and the initial pressure of 280.081 psi. The model cannot
reproduce the measured pressures for the flowing periods, even
though we included rate-dependent skin effects (Fig. 30). As dis-
cussed previously in detail, this discrepancy may be caused by
errors in measured flow rates; though we do not know how to
correct them or if the estimated (or adjusted) rates by deconvolu-
tion are correct. However, the model reproduces the constant unit-
rate drawdown buildup responses derived from deconvolution of
buildup pressures (Fig. 28) and rate-normalized buildup responses
based on superposition time (Fig. 29) fairly well at late times. It
does not, however, reproduce the estimated derivative responses
well at early times in the time interval from 0.03 to 0.5 hr. We

* Based on a personal communication between Levitan and the first author of this paper
through an e-mail on 16 October 2006, by sending his results in a figure (digital data were
not provided by Levitan) generated from his implementation of the algorithm for the this field
Fig. 27—Comparison of deconvolved responses derived from example, Levitan showed that his implementation of the algorithm did not produce the
buildup pressure data using the von Schroeter et al., Levitan, problem [i.e., a negative value of pu(t1)] that our implementation did. Based on this com-
and Ilk et al. algorithms with the conventional pressure buildup munication, it was also brought to our attention that some details of Levitan’s algorithm were
not revealed in his paper (2005) for purposes of protection of intellectual property rights.
pressure change and its logarithmic derivative (normalized by Without complete details of the algorithm, we could not fully evaluate it and reproduce his
the last rate prior to buildup); field example. results for the field example.

242 June 2008 SPE Journal

Fig. 28—Model match of the unit-rate responses derived from Fig. 29—Model match of buildup pressure change and its de-
deconvolution of buildup pressures using measured rates rivative using measured rates and initial pressure of 280.081
and initial pressure of 280.081 psi. The model is a partially pene- psi; field example.
trating well in a reservoir with two parallel no-flow faults,
field example.
but with the estimated rates from deconvolution (Fig. 23). The
results were similar to those obtained by using measured rates. For
believe that this is because the model we used cannot incorporate example, with the model used to generate the results given in Figs.
a finite-conductivity fault nearby the well. We note that we obtain 28 through 30, we obtained the permeability-thickness product as
a much better match for the early-time portions of the derivative 1175 Darcy-m, mechanical skin as –2.3, and the distances to the
responses by using the model based on a finite-conductivity fault no-flow faults as 330 m. The model matches for deconvolved and
(for a fully penetrating well with changing wellbore-storage ef- buildup responses were similar to those shown in Figs. 28 and 29,
fects) given by Abbaszadeh and Cinco-Ley (1995), as shown in but the model match for the pressures during the flow periods was
Fig. 31. The model fails, however, to reproduce the late-time por- much better than that shown in Fig. 30 based on measured rates.
tions of the buildup after 1 hr, as expected, because it assumes that
the zones on both sides of the fault plane are of infinite extent. To Discussion
match both the early- and late-time portions of the buildup re- Based on the results of our three synthetic and one field examples,
sponses we need a finite-conductivity fault model that considers we propose, as a general statement, that the von Schroeter et al.
one of the zones as bounded by two parallel no-flow boundaries. (2002, 2004), Levitan (2005), and Ilk et al. (2006a) methods make
The permeability-thickness product (∼295**−957 Darcy-m), me- deconvolution a viable tool, and are very useful for the analysis of
chanical skin (∼−1.0**−–2.5), and distances to the finite conduc- variable rate/pressure well test and production data. We believe
tivity fault (∼30** m) and no-flow faults (∼340 m) estimated from that these deconvolution methods open a new era in the interpre-
both models were reasonable and consistent with the available tation of well-test and production data.
geologic model and estimated parameters from the tests conducted However, the proper use of these methods requires an under-
in other nearby wells. Finally, we should note that we also per- standing of their assumptions, and also an understanding of the
formed history matches of the test data by using the same models, deconvolution problem itself, particularly its ill-posed nature. In
practice, we never know a priori to what extent a given data set is
consistent with the deconvolution model (Eq. 1), and hence, we
concur with both the von Schroeter et al. (2002, 2004) and Levitan
** Estimated values based on the finite-conductivity fault model (Abbaszadeh and Cinco-
Ley 1995). (2005, 2006) perspectives to avoid problems (or failure) and to iden-
tify any inconsistencies in the data, when applying deconvolution.

Fig. 30—Model match of measured pressure data during the Fig. 31—Model match of the unit-rate responses derived from
whole test sequence, using measured rates and initial pressure deconvolution of buildup pressures using measured rates and
of 280.081 psi. The model is a partially penetrating well in a initial pressure of 280.081 psi. The model is a finite conductivity
reservoir with two parallel no-flow faults; field example. fault (Abbaszadeh and Cinco-Ley 1995); field example.

June 2008 SPE Journal 243

 The von Schroeter et al. perspective is to jointly estimate rates with pressure data, and inaccuracy in initial pressure. However, for
and/or initial pressure together with the unit-rate drawdown de- the cases where a well-test sequence or production data do not
rivative response, and then check whether the results honor the contain buildup periods or contain only a single buildup period, the
data to within expected error margins, coupled with the statistical use of this strategy is limited because a single buildup or flow (or
analysis (e.g., rms errors in pressure and rate as well as 95% drawdown) period do not contain enough information to identify
confidence intervals) based on Gaussian statistics. As shown pre- the correct value of initial pressure and rate history required for
viously, their perspective works very well if we have consistent reconstruction of correct constant unit-rate drawdown pressure.
data sets and most importantly a priori knowledge of error levels However, variable-rate drawdown data, if of “good quality,” can
in rate and pressure, but more importantly in rate. In cases of bear significant information, and does not rule out successful use
inconsistent data sets as in the field example case considered, the of deconvolution.
inconsistency of the data with the deconvolution model, as well as Here, we provide some further general remarks on deconvolu-
the validity of such prior knowledge, may also be detected by tion. Regardless of the deconvolution method used, the quality of
checking pressure and rate matches coupled with the statistical a deconvolved response depends on (i) the validity of the pressure-
analysis. However, as we faced in the case of our field example rate data pair with the deconvolution model of Eq. 1, (ii) data
application, for which the data subsets are not consistent because subsets used for deconvolution, (iii) regularization parameters
of changing wellbore-storage and rate-dependent skin effect, we (“weights”) used, and (iv) possible use of constraints [i.e., con-
do advise caution that adjusted rates obtained by processing pres- straining initial pressure or rates (as fixed) in deconvolution]. In
sures based on the von Schroeter et al. strategy may not necessarily addition, deconvolution could be very sensitive to small errors in
be correct because deconvolution adjusts the rates for the sake of the data. For example, small time lags or time synchronization
“honoring” measured pressures. Therefore, we do not view this errors between rates and pressure could easily lead to large aber-
deconvolution strategy or deconvolution in general as a tool to rations or oscillations in the deconvolved response, as pointed out
correct the rates from pressure data alone for possible measure- by von Schroeter et al. (2004) and demonstrated by Levitan et al.
ment errors. In fact, we could easily show that there is no decon- (2006). Some data may not be representative of reservoir flow
volution method or algorithm that can estimate the rate history because of wellbore/surface effects, and even some rate data may
from the pressure data, or “correct” the measured rate, even in the be missing. Hence, due care must be taken in preparing and se-
case of consistent data sets unless we know the spread of the error lecting data to be used in deconvolution. When working with very
(or expected error margins) in rate data. Here, we will give two large datasets, data filtering and flow rate decimation and simpli-
simple examples to show this point. fication become critical in practical use of deconvolution.
Suppose we have a constant rate drawdown followed by a In summary, deconvolution is an interpretation tool for the data
buildup. Furthermore, let us assume that pressure/rate data are itself, and it depends on the user running it, and therefore requires
consistent with the deconvolution model of Eq. 1. Further, suppose data preparation as well as the user’s judgment. We believe that the
that we process the entire drawdown and buildup pressure data by deconvolved response must always be compared with the conven-
deconvolution with the drawdown rate and unit-rate (or z–) re- tional derivative response, and more importantly must be consis-
sponse as unknown. Can the deconvolution methods (von Schroe- tent with geological prior information, and be validated through
ter 2002, 2004; Levitan 2005, 2006) estimate the true (unknown) pressure history matching based on the possible well/reservoir
drawdown rate if an initial guess used for the rate is different from model(s). Furthermore, we think that it is too much to expect a
the true value? The answer is “no,” because the algorithms will deconvolution algorithm to correct both rate history and initial
(and should) give back the same initial input rate without provid- pressure. It is also unfortunate that the late-time portion of the
ing any correction, but the unit-rate responses (pressure change deconvolved unit-rate response is so sensitive to the value of initial
and derivative) will be vertically displaced depending on the input pressure. Irrespective of all these issues, we believe that it should
value of drawdown flow rate, though pressure match for both the also be the task of technology to provide more accurate measure-
drawdown and buildup periods will be perfect. The reason is that ments of rates and initial pressure to be used in the deconvolution
we do not have a unique solution for the z-response and for the methods considered in this work.
drawdown rate, and hence any value of drawdown flow rate with
the estimated z-response (based on this flow rate) will reproduce Conclusions
the same pressure history. We can give another example: Suppose As mentioned previously, the algorithms used in this study for
that we have a test with rate 1, buildup, rate 2, buildup. Again, we evaluating the von Schroeter et al. (2002,2004) and Levitan (2005)
believe that there is no deconvolution method that could estimate methods represent our independent implementations based on the
the rate history, or correct the measured rate. At best, it might be material presented in their papers. These implementations form the
able to estimate the ratio of the two rates, or it might identify an basis of our conclusions.
inconsistency in the two measured rates, but it would never know Based on our investigation, we can state the following specific
which is correct and which would be wrong. A prior knowledge of conclusions related to each algorithm:
error level in rate data will be a prerequisite for one to be able to 1. Our algorithm based on the von Schroeter et al. method:
determine the appropriate values of rates estimated from decon- • Algorithm ensures positivity in deconvolved pressure and de-
volution. Of course, if we have further inconsistency of pressure rivative responses.
data (e.g., in the form of changing skin from flow period to flow • Algorithm addresses pressure and rate uncertainties.
period) with the deconvolution model of Eq. 1, then matching • Algorithm allows the treatment of initial reservoir pressure and
pressure data pertaining to only the buildups or drawdowns or both rates as unknown in deconvolution.
drawdown and buildups while letting flow rate and initial pressure • Algorithm requires only two input parameters, namely ␯ and
go unknown in deconvolution could easily lead to nonunique re- ␭—the relative error weight and the regularization parameter.
sults. In such cases, inspecting statistics of the deconvolution re- The prescribed (default) values are typically sufficient. The
sults may prove useful, as shown with the synthetic and field data parameter ␯ can be estimated directly from measured pressure
example applications. and rate data available as the ratio of the “pressure error” to
The Levitan perspective is to apply deconvolution for each “rate error” and can be fixed at its default value, while the
flow (preferably buildup) period separately with rates and initial regularization parameter ␭ is adjusted manually by powers of
pressure fixed in each run in order to check the consistency of data ten of ␭ˆ def (Eq. 11) computed from pressure data in deconvo-
with the deconvolution model of Eq. 1. As shown with the syn- lution until a smooth enough response, interpretable without
thetic and field data applications, this strategy also works very well losing its dominant features, is obtained.
if the test sequence contains a few buildups in the sense that we • Assumption of a wellbore-storage unit-slope trend at/before
can identify the correct reservoir response at late times, and detect the first node causes artifacts on the unit-rate derivative re-
any inconsistencies of the pressure/rate data sets with the decon- sponse near the first node for data sets with no storage or
volution model of Eq. 1, identify any inconsistency of rate data minimal storage effects, but these artifacts disappear after the

244 June 2008 SPE Journal

 first log-cycle. For data with no storage or minimal storage correct confidence intervals), though the resulting estimates of
effects, the value of ez1 estimated at the first time node repre- the unit-rate response may be reasonable.
sents the constant unit-rate drawdown pressure drop including • Algorithm does not ensure positivity in deconvoled responses.
the skin effect, not the value of constant unit-rate drawdown • The regularization parameter ␣ in the algorithm is adjusted
derivative response. manually, and the regularization approach used by Ilk et al.
2. Our algorithm based on Levitan method: tend to make the algorithm less tolerant to the errors in pres-
• Algorithm ensures positivity in deconvolved deriva- sure and rate data, compared to the von Schroeter et al. and
tive responses. Levitan algorithms.
• Algorithm addresses pressure and rate uncertainties.
• Algorithm allows the treatment of initial reservoir pressure and
rates as unknown in deconvolution. Nomenclature
• Algorithm requires three input (control) parameters: ␵p, ␵q, and ⳱ coefficients in rate function, Eq. 43, L3/t, bbl/ a
␵c, where ␵p and ␵p represent the standard deviation of errors day [cm3/s]
(“error band” or noise levels) for pressure and rate data and ␵c b ⳱ coefficients in rate function, Eq. 43, or base of
represents the “weight” for the curvature constraint. The val-
logarithmically evenly distributed knots
ues of ␵p and ␵q to be used in the algorithm are determined
B2i (t) ⳱ second-degree B-spline starting at bi
from our prior knowledge of noise levels in pressure and rate
data, and the curvature constraint weight is adjusted manually. B2i,int(t) ⳱ integral of 2nd degree B-spline starting at bi
As shown, if each measured pressure point in the pressure data ct ⳱ total compressibility, Lt2/m, 1/psi, [1/atm]
set has the same standard deviation of error ␵p, and each mea- c ⳱ vector of unknown B-spline coefficients, Eq. 27
sured rate point in the rate data set has the same standard Cw ⳱ wellbore-storage coefficient, L4t2/m, stb/psi [cm3/atm]
deviation of error ␵q, and if we ignore the differences in en- C̃ ⳱ design matrix for pressure match in Levitan’s
coding between the von Schroeter et al. and Levitan methods, objective function, Eq. 16.
the objective functions considered by these authors are iden- C ⳱ design matrix for pressure match in von Schroeter
tical provided that we define ␯⳱(␵p/␵q)2 and ␭⳱(␵p/␵c)2. Hence, et al.’s objective function, Eq. 8
only the ratios ␵p/␵q and ␵p/␵c are required, and as a result the D ⳱ matrix in curvature measure
number of control parameters is the same in both methods.
e ⳱ vector of size m with each component equal to 1
• Algorithm removes the assumption of unit-slope trend at/
before the first node, and hence can reconstruct the unit-rate ez1 ⳱ exponential of response function, z1 at the first node
responses correctly at the very early times even if the data do E ⳱ objective function, Eqs. 8, 16, and 31
not exhibit wellbore-storage unit-slope line, provided that time Ẽ ⳱ objective function, Eq. 37
value t1 for the first node is chosen less than or equal to the h ⳱ thickness, L, ft, [cm]
minimum elapsed time in the pressure signal. Use of L−2 k ⳱ permeability, L2, md [Darcy ]
curvature constraint equations in the objective function (Eq. k ⳱ vector in curvature measure
16) provides slightly better representation of the reconstructed K ⳱ index of the last knot for B-spline
unit-rate response than that of L−1 curvature constraint equations. l ⳱ index of the first knot for B-spline
• There is no positivity constraint on the constant unit-rate draw- L ⳱ number of nodes for the deconvolved response
down pressure response at the first grid node [i.e., pu(t1)],
Lxf ⳱ fracture half length, L, ft [cm]
which is treated as an unknown model parameter in the algo-
rithm. Therefore, the resulting value may be negative. As seen L1 ⳱ perpendicular distance to the first no-flow boundary,
in our Synthetic Example 2 and field data application, this can L, ft [cm]
cause an artifact at the reconstructed unit-rate responses at the L2 ⳱ perpendicular distance to the second no-flow
very early times. boundary, L, ft [cm]
• When the algorithm is applied to individual buildup periods, L−1 ⳱ Laplace inverse transform operator
the sensitivity of the model pressure with respect to pu(t1) M ⳱ number of pressure data points
depends only on the flow rate at the instant of shut-in. n ⳱ number of unknown coefficients cis, Eq. 34 or
• An improper selection of t1 value and of curvature constraint number of B-splines
␵c, and/or measurement errors in pressure/rate and time syn-
chronization errors in rate and pressure data, are possible N ⳱ number of flow periods or segments
causes of driving the unconstrained nonlinear minimization p ⳱ pressure, m/Lt2, psi [atm]
algorithm to a negative value for pu(t1). p0 ⳱ initial pressure, m/Lt2, psi [atm]
3. Ilk et al. algorithm: pu ⳱unit-rate drawdown-pressure response, m/L3t3, psi
• Algorithm offers the flexibility of modeling rates as continu- /(stb/d) [atm/(cm3/s)]
ous measurements. p ⳱ vector of pressures, m/Lt2, psi [atm]
• Algorithm handles the earliest part of the unit-rate derivative ⌬p ⳱ pressure change, m/Lt2, psi [atm]
response for the data with a skin effect (but minimal or ⌬p ⳱ (vector of) pressure change, m/Lt2, psi [atm]
no wellbore-storage effects) by adding additional “anchor”
B-splines. Four additional B-splines added to the left of the ⌬p* ⳱ (vector of) augmented pressure changes, Eq. 36,
first observed time point in the signal are sufficient for virtu- m/Lt2, psi [atm]
ally any scenario. q ⳱ flow rate, L3/t, stb/day [cm3/s]
• Algorithm considers a linear weighted least-squares objective re ⳱ the distance from the well to the circular boundary,
function based on a diagonal weighting matrix to account for L, ft [cm]
errors only in pressure data. The weights in diagonal elements rw ⳱ wellbore radius, L, ft [cm]
of the matrix are defined as the reciprocal of the observed s ⳱ Laplace transform variable, 1/t, hour-1 [s-1]
value of measured pressure drop (i.e., the measured pressures S ⳱ mechanical skin, dimensionless
are corrupted by a constant level of relative error). However, in
view of the concept of maximum likelihood estimation, the t ⳱ time, t, hour [s]
weights should actually represent the inverse of standard de- t⬘ ⳱ time, t, hour [s] (integration variable)
viation of errors in data to have the correct assessment of t1 ⳱ time for the first node, t, hour [s]
statistics of the match, and hence using a weighting matrix as W ⳱ weighting matrix
proposed by Ilk et al. may yield incorrect statistics (e.g., in- X ⳱ sensitivity matrix, Eq. 27

June 2008 SPE Journal 245

 X* ⳱ augmented matrix, Eq. 36 References
X+a ⳱ pseudoinverse matrix of the augmented matrix, Eq. 39 Abate, J. and Valkó, P.P. 2004. Multi-Precision Laplace Transform Inver-
y ⳱ (vector of) estimated rates, Eq. 8, L3/t, stb/day [cm3/s] sion. International Journal for Numerical Methods in Engineering 60
ỹ ⳱ (vector of) estimated rates (Levitan algorithm), L3/t, (5): 979–993. DOI: 10.1002/nme.995.
stb/day [cm3/s] Abbaszadeh, M. and Cinco-Ley, H. 1995. Pressure-Transient Behavior in
z ⳱ response function defined by Eq. 6 a Reservoir With a Finite-Conductivity Fault. SPEFE 10 (1): 26–32;
Trans., SPE, 299. SPE-24704-PA. DOI: 10.2118/24704-PA.
z1 ⳱ value of response function at the first node
Agarwal, R.G. 1980. A New Method To Account for Production Time
z ⳱ (vector of) response coefficients Effects When Drawdown Type Curves Are Used To Analyze Buildup
␣ ⳱ regularization parameter (Ilk et al. algorithm), and Other Test Data. Paper SPE 9289 presented at the SPE Annual
unit-less Technical Conference and Exhibition, Dallas, 21–24 September. DOI:
␦ ⳱ (vector of) absolute rate errors, L3/t, bbl/day [cm3/s] 10.2118/9289-MS.
␧ ⳱ (vector of) absolute pressure data errors, m/Lt2, Bard, Y. 1974. Nonlinear Parameter Estimation. New York City: Aca-
psi [atm] demic Press, Inc.
␵ ⳱ error bounds used in Eq. 16 Baygun, B., Kuchuk, F.J., and Arikan, O. 1997. Deconvolution Under
␩ ⳱ error levels, Eqs. 2 and 3, fraction Normalized Autocorrelation Constraints. SPEJ 2 (3): 246–253. SPE-
28405-PA. DOI: 10.2118/28405-PA.
␪ ⳱ Heaviside function
Björck, Å. 1996. Numerical Methods for Least Square Problems. Phila-
␭ ⳱ regularization parameter in Eq. 8 or interporosity delphia, Pennsylvania: Society for Industrial and Applied Mathematics.
flow coefficient, m2/L2t4, psi2 [atm2] Bourdet, D. 2002. Well Test Analysis: The Use of Advanced Interpretation
␭def ⳱ default value for regularization parameter defined by Models. Amsterdam: Elsevier Science B.V.
Eq. 10, m2/L2t4, psi2 [atm2] Bourdet, D., Ayoub, J.A., and Pirard, Y.M. 1989. Use of Pressure Deriva-
␭ˆ def ⳱ default value for regularization parameter defined by tive in Well-Test Interpretation. SPEFE 4 (2): 293–302; Trans., SPE,
Eq. 11, m2/L2t4, psi2 [atm2] 287. SPE-12777-PA. DOI: 10.2118/12777-PA.
␮ ⳱ viscosity, m/Lt, cp [cp] Carvalho, R., Thompson, L.G., Redner, R., and Reynolds, A.C. 1996.
Simple Procedures for Imposing Constraints for Nonlinear Least
␯ ⳱ relative error weight used in Eq. 8, m2/L8t2,
Squares Optimization. SPEJ 1 (4): 395–402. SPE-29582-PA. DOI:
(psi.d/stb)2 [(atm.s/cm3)2]
10.2118/29582-PA.
␯def ⳱ default value for relative error weight defined by Eq. Cheney, E.W. and Kincaid, D.R. 2003. Numerical Mathematics and Com-
9, m2/L8t2, (psi.d/stb)2 [(atm.s/cm3)2] puting. Pacific Grove, California: Brooks Cole Publishing Company.
␴ ⳱ standard deviation of errors Çinar, M., Ilk, D., Onur, M., Valkó, P.P., and Blasingame, T.A. 2006. A
␶ ⳱ variable of integration or logarithm of time Comparative Study of Recent Robust Deconvolution Algorithms for
␾ ⳱ porosity, fraction Well-Test and Production-Data Analysis. Paper SPE 102575 presented
␻ ⳱ storativity ratio for the double porosity model, at the SPE Annual Technical Conference and Exhibition, San Antonio,
dimensionless Texas, 24–27 September. DOI: 10.2118/102575-MS.
Coats, K.H., Rapoport, L.A., McCord, J.R., and Drews, W.P. 1964. De-
⍀ ⳱ data model predictions for curvature constraints, Eqs.
termination of Aquifer Influence Functions From Field Data. JPT 16
17 and 18 (12): 1417–1424; Trans., AIME, 231. SPE-897-PA. DOI: 10.2118/897-
㛳 㛳2 ⳱ l2-norm of a vector PA.
Dennis, J.E. and Schnabel, R.B. 1983. Numerical Methods for Uncon-
strained Optimization and Nonlinear Equations. Englewood Cliffs,
Subscripts New Jersey: Prentice-Hall Inc.
c ⳱ curvature Earlougher, C.R. Jr. 1977. Advances in Well Test Analysis. Monograph
def ⳱ default Series, SPE, Richardson, Texas, 5.
Golub, G.H., Hansen, P.C., and O’Leary, D.P. 1999. Tikhonov Regular-
m ⳱ measured
ization and Total Least Squares. SIAM J. Matrix Anal. Appl. 21 (1):
F ⳱ Frobenius norm 185–194. DOI: 10.1137/S0895479897326432.
p ⳱ pressure Hegeman, P.S., Hallford, D.L, and Joseph, J.A. 1993. Well Test Analysis
q ⳱ rate With Changing Wellbore Storage. SPEFE 8 (3): 201–207; Trans., SPE,
u ⳱ unit-rate 295. SPE-21829-PA. DOI: 10.2118/21829-PA.
Hutchinson, T.S. and Sikora, V.J. 1959. A Generalized Water-Drive
Analysis. Trans., AIME 216: 169–178. SPE-1123-G.
Superscripts Ilk, D. 2005. Deconvolution of Variable Rate Reservoir Performance Data
b⳱ buildup Using B-Splines. MS thesis, College Station, Texas: Texas A&M Uni-
versity.
T
⳱ transpose of a matrix
Ilk, D., Valko, P.P., and Blasingame, T.A. 2006a. Deconvolution of Vari-
^⳱ estimated (from least squares) able-Rate Reservoir-Performance Data Using B Splines. SPEREE 9
—
⳱ Laplace transform (4): 582–595. SPE-95571-PA. DOI: 10.2118/95571-PA.
–1 ⳱ inverse of a matrix Ilk, D., Anderson, D.M., Valko, P.P., and Blasingame, T.A. 2006b. Analy-
sis of Gas-Well Reservoir Performance Data Using B-Spline Decon-
volution. Paper SPE 100573 presented at the SPE Gas Technology
Acknowledgments Symposium, Calgary, 15–17 May. DOI: 10.2118/100573-MS.
The first author thanks Drs. Thomas von Schroeter of Imperial Jargon J.R. and Van Poollen, H.K. 1965. Unit Response Function From
College and Michael Levitan of BP for providing data to verify our Varying Rate Data. JPT 17 (8): 965–969; Trans., AIME, 234. SPE-
versions of their algorithms developed during the course of this 981-PA. DOI: 10.2118/981-PA.
work, and for providing their comments to our questions. We also Kuchuk, F.J. 1990. Applications of Convolution and Deconvolution to
are thankful to the anonymous referees for their careful reviewing Transient Well Tests. SPEFE 5 (4): 375–384; Trans., SPE, 289. SPE-
of the manuscript and useful suggestions. The first author also 16394-PA. DOI: 10.2118/16394-PA.
thanks Kappa for providing Saphir free for educational and re- Kuchuk, F.J., Carter, R.G., and Ayestaran, L. 1990. Deconvolution of
search purposes for the Department of Petroleum and Natural Gas Wellbore Pressure and Flow Rate. SPEFE 5 (1): 53–59. SPE-13960-
Engineering at ITU. PA. DOI: 10.2118/13960-PA.

246 June 2008 SPE Journal

Kuchuk, F.J., Hollaender, F., Gok, I.M., and Onur, M. 2005. Decline nical University in Turkey and MS and PhD degrees from the
Curves from Deconvolution of Pressure and Flow-Rate Measurements University of Tulsa, all in petroleum engineering. He has served
for Production Optimization and Prediction. Paper SPE 96002 pre- on the editorial committees of SPEREE and SPE Journal. Murat
sented at the SPE Annual Technical Conference and Exhibition, Dallas, Çınar is a PhD student in the Department of Energy Resources
Engineering at Stanford University. His research interest in-
9–12 October. DOI: 10.2118/96002-MS.
cludes well-test analysis, geothermal engineering and in-situ
Levitan, M.M. 2005. Practical Application of Pressure/Rate Deconvolution combustion. He holds a BS degree and MS degrees from Istan-
to Analysis of Real Well Tests. SPEREE 8 (2): 113–121. SPE-84290- bul Technical University, Turkey, both in petroleum engineering.
PA. DOI: 10.2118/84290-PA. Dilhan Ilk is a PhD student in the Department of Petroleum
Levitan, M.M., Crawford, G.E., and Hardwick, A. 2006. Practical Consid- Engineering at Texas A&M University. His research interest in-
erations for Pressure-Rate Deconvolution of Well-Test Data. SPEJ 11 cludes well-test/production-data analysis, reservoir engineer-
(1): 35–47. SPE-90680-PA. DOI: 10.2118/90680-PA. ing, and numerical analysis. He holds a BS degree from Istanbul
Onur, M. et al. 2007. Analysis of Well Tests in Afyon Ömer-Gecek Geo- Technical University, Turkey, and an MS degree from Texas
thermal Field, Turkey. Proc., 32nd Workshop on Geothermal Reservoir A&M University, both in petroleum engineering. Peter P. Valkó
is a professor of petroleum engineering at Texas A&M Univer-
Engineering, Stanford University, Stanford, California, 22–24 January.
sity. Previously, he taught at academic institutions in Austria
Press, H. Flannery, B.P., Teukolsky, S.A., and Vetterling, W.T. 1986. and Hungary and worked for the Hungarian Oil Co. (MOL). His
Numerical Recipes: The Art of Scientific Computing. Cambridge, UK: research interests include design and evaluation of hydraulic
Cambridge University Press. fracture stimulation treatments, rheology of fracturing fluids,
Saphir Ver. 3.20.10. software. 2005. Sophia Antipolis, France: Kappa En- and performance of advanced and stimulated wells. Valkó
gineering. has coauthored three books and published several dozen re-
Thompson, L.G. and Reynolds, A.C. 1986. Analysis of Variable-Rate search papers. He holds BS and MS degrees from Veszprém
Well-Test Pressure Data Using Duhamel’s Principle. SPEFE 1 (5): University, Hungary, and a PhD degree from the Institute of
Catalysis, Novosibirsk, Russia. Tom Blasingame is a professor of
453–469. SPE-13080-PA. DOI: 10.2118/13080-PA.
petroleum engineering at Texas A&M University, where he
van Everdingen, A.F. and Hurst, W. 1949. The Application of the Laplace works in the areas of reservoir engineering and analysis of res-
Transformation to Flow Problems in Reservoirs. Trans., AIME 186: ervoir performance. He holds BS, MS, and PhD degrees from
305–324. SPE-949305-G. Texas A&M University, all in petroleum engineering. Blasingame
von Schroeter, T., Hollaender, F., and Gringarten, A.C. 2004. Deconvolu- is a Distinguished Member of SPE, and received both the 2005
tion of Well Test Data as a Nonlinear Total Least-Squares Problem. SPE Distinguished Service Award and 2006 Lester C. Uren
SPEJ 9 (4): 375–390. SPE-77688-PA. DOI: 10.2118/77688-PA. Award. He has chaired Forums, Applied Technology Work-
shops (ATWs), and committees. Peter Hegeman is a project
manager and engineering advisor in the Reservoir Sampling
Mustafa Onur is a professor and head of the Department of and Pressure Discipline of Schlumberger Sugar Land Product
Petroleum and Natural Gas Engineering at Istanbul Technical Center. His interests include well testing, pressure-transient
University, Turkey. His interests include well testing, pressure- analysis, formation testing, and production systems analysis. He
transient analysis, formation testing, and geothermal reservoir holds BS and MS degrees in petroleum engineering from the
engineering. He holds a BS degree from the Middle East Tech- Pennsylvania State University.

June 2008 SPE Journal 247