SPE 95571
Deconvolution of Variable-Rate Reservoir Performance Data Using B-Splines
D. Ilk, P.P. Valko, SPE, and T.A. Blasingame, SPE, Texas A&M U.
Copyright 2005, Society of Petroleum Engineers
This paper was prepared for presentation at the 2005 SPE Annual Technical Conference and
Exhibition held in Dallas, Texas, U.S.A., 9 – 12 October 2005.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
This work presents the development, validation and applica-
tion of a novel deconvolution method based on B-splines for
analyzing variable-rate reservoir performance data. Variable-
rate deconvolution is a mathematically unstable problem
which has been under investigation by many researchers over
the last 35 years. We believe that this work is an important
addition to existing well test/production data analysis methods
where few deconvolution methods are practically applicable.
We use B-splines for representing the derivative of unknown
unit-rate drawdown pressure and numerical inversion of the
Laplace transform is utilized in our formulation. When
significant errors and inconsistencies are present in the data
functions, the direct and indirect regularization methods (i.e.,
mathematical "uniformity" processes) are used. We provide
examples of under and over-regularization, and we discuss
procedures for ensuring proper regularization.
We validate our method with synthetic examples generated
with and without errors (for this work we provide cases with
10 percent error, but we have considered cases with as high as
40 percent error). Upon validation, we then demonstrate our
deconvolution method using a variety of field cases —
including traditional well tests, permanent downhole gauge
data as well as production data. Our work suggests that the
new deconvolution method has broad applicability in variable
rate/pressure problems — and can be implemented in typical
well test and production data analysis applications.
Objectives
The following objectives are proposed for this work:
z To develop and validate a new deconvolution method
based on B-spline representations of the derivative of
unknown constant rate drawdown pressure response (i.e.,
the undistorted pressure response).
z To create a practical and robust deconvolution tool that
can tolerate relatively large (random) errors in the input
rate and pressure functions. The proposed process should
also be capable of tolerating small systematic errors in the
input functions (via calibrated regularization).
z To apply this new method to traditional variable-rate/
pressure problems, such as wellbore storage distortion,
long-term production data, permanent downhole (pres-
sure) gauge data, and well tests having multiple flow
sequences.
Introduction
The constant-rate drawdown pressure behavior of a well/reser-
voir system is the primary signature used to classify/establish
the characteristic reservoir model. Transient well test proce-
dures are typically designed to create a pair of controlled flow
periods (a pressure drawdown/buildup sequence), and to
convert the last part of the response (the pressure buildup) to
an equivalent constant-rate drawdown via special time
transforms. However, the presence of wellbore storage,
previous flow history, and rate variations may mask or distort
characteristic features in the pressure and rate responses.
With the ever increasing ability to observe downhole rates, it
has been long recognized that variable-rate deconvolution
should be a viable option to traditional well testing methods
because deconvolution can provide an equivalent constant-rate
response for the entire time span of observation. This po-
tential advantage of variable-rate deconvolution has become
particularly obvious with the appearance of permanent down-
hole instrumentation.
First and foremost, variable-rate deconvolution is mathematic-
cally ill-conditioned — while numerous methods have been
developed and applied to deconvolve "ideal" data — very few
deconvolution methods perform well in practice. The ill-con-
ditioned nature of the deconvolution problem means that small
changes in the input data cause large variations in the
deconvolved constant-rate pressures. Mathematically, we are
attempting to solve a first-kind Volterra equation (see Lamm1),
that is ill-posed. However, in our case the kernel of the Vol-
terra-type equation is the flowrate function (i.e., the generating
function) and it is not known analytically, but rather, is
approximated from the observed flowrates, and in practical
terms, this issue adds to the complexity of the problem.2
2 SPE 95571
In the literature related to variable-rate deconvolution we find
the development of two basic concepts. One concept is to
incorporate an a priori knowledge regarding the properties of
the deconvolved constant-rate response. The observations of
Coats et al.3 on the strict monotonicity of the solution lead
Kuchuk et al.2,4,5 to impose a "nonpositive second derivative"
constraint on pressure response. In some respects this
tradition is maintained in the work given by Schroeter et al.5,6,
7
when they incorporate non-negativity in the "encoding of the
solution". We note that in the examples given, this concept
(non-negativity/monotonicity of the solution) requires non-
standard numerical methods (linear programming, non-linear
least-squares minimization).
The second concept is to use a certain level of regularization5,6
— where "regularization" is defined as the act/process of
making a system regular or standard (smoothing or eliminating
non-standard/irregular response features). Regularization can
be performed indirectly by representing the desired solution
using a restricted number of "elements" — or directly by
penalizing the "non-smoothness" of the solution. In either
case, the additional degree of freedom (the regularization
parameter) has to be established — where this is facilitated by
the discrepancy principle (effectively tuning the regularization
parameter to a maximum value, while not causing intolerable
deviation between the model and the observations). In some
fashion, each deconvolution algorithm developed to date
combines these two concepts (non-negativity/monotonicity of
the solution or regularization).
In addition to these main features, individual deconvolution
algorithms may have other distinctive features. From a
numerical standpoint, recent deconvolution methods tend to
use advanced techniques to solve the underlying system of
linear equations (e.g., Singular Value Decomposition, the
equivalent concept of pseudoinverse, etc. (see ref. 8)).
One class of approaches makes extensive use of transforma-
tions (i.e., the Laplace transform, the Fourier transform, etc.).
Several cases consider "numerical Laplace transformation of
observed tabulated data" combined with numerical inversion
(Roumboutsos and Stewart,9 Onur and Reynolds10). In addi-
tion, Cheng et al.11 consider the application of the Fourier
transformation to solve the deconvolution problem.
Error minimization methods — typically driven by some type
of least squares algorithm include Schroeter's "total least
squares" method (Schroeter et al.5, Levitan6), as well as a
group of methods which rely on the use of spline functions (in
various steps of the deconvolution) algorithm (e.g., Gilly and
Horne12 and Mendes et al.13).
Formally, it would be easy to place our proposed method into
existing categories. Our proposed approach uses B-splines,
numerical inversion of the Laplace transform is also used for
certain components of the solution, regularization is provided
indirectly (i.e., by the number of knots used in the selected B-
spline) and directly (by penalizing the non-smoothness of the
logarithmic derivative of the reconstructed constant-rate
response).
However, in detail our approach is radically different from any
of the the existing methods — in contrast to previous methods
our approach does not fit B-splines to observed data. Rather,
B-splines are used to represent the unknown response solution
(i.e., as a linear combination of B-splines). With respect to the
Laplace transform, our approach does not use the Laplace
transform to transform the observed pressure data series nor
does our approach use numerical inversion to provide the
objective response (i.e., the constant rate response) as do other
spectral methods. In our approach, the application of nume-
rical inversion of the Laplace transform is restricted to the
construction of the sensitivity matrix for the least-squares
problem.
Our new technique represents the derivative of the unknown
constant-rate drawdown pressure response as a weighted sum
of B-splines, using logarithmically-distributed knots. For the
discrete flowrate function, we use piecewise constant,
piecewise linear, or any other appropriate representation for
which the Laplace transform can be easily obtained. Taking
the Laplace transform of the B-splines, we then apply the con-
volution theorem in the Laplace domain and we calculate the
sensitivities of the observed pressure response with respect to
the B-spline weights by numerical inversion of the Laplace
transform.
Finally, the sensitivity matrix is used in conjunction with a
least-squares criterion to yield the sought B-spline weights and
the unit-rate response (the spline formulation) — which then
yields the well testing derivative functions in the real domain.
The combined approach of B-splines, Laplace domain
convolution, and the least-squares procedure are innovative
and robust — and should find numerous applications in the
petroleum industry (production and well test data analysis,
inversion of water influx behavior, etc.).
Paradoxically, this new deconvolution approach is made
possible by our ability to solve a certain class of convolution
problems with any desired accuracy, in particular, by the
availability of a robust numerical inversion procedure for the
Laplace transform (i.e., the Gaver-Wynn-Rho algorithm14,15).
Statement of the Problem
Duhamel's principle states that the observed pressure drop is
the convolution of the input rate function and the derivative of
the constant-rate pressure response — at t=0 the system is
assumed to be in equilibrium (i.e., p(r,t=0) = pi). For refer-
ence, the convolution integral is defined as:
t
∫0 q(t − τ ) p' (τ )dτ = ∆p(t) ..............................................(1)
u
The goal of variable-rate deconvolution is to estimate the unit
(i.e., constant) rate reservoir response using observed measure-
ments of pressure drop and flowrate. This objective leads us
to the inverse problem in which the time dependent "input
signal" (constant-rate pressure response) must be extracted
from the output signal (pressure drop response distorted by the
specified variable-rate profile). To reconstruct the unknown
function (i.e., the derivative of the unit rate reservoir pressure
response (K(t) = pu'(t)), we require the solution of the Volterra
SPE 95571
integral equation of the first kind (i.e., the convolution inte-
gral), where the role of the convolution kernel is represented
by the sandface rate.
Lamm1 shows that for q(t-τ)τ=t = 0 — i.e., when the rate does
not jump instantaneously at the first moment, the left-hand-
side is "smoothing at least of order two," and hence the
problem is severely ill-conditioned. In addition, the sandface
flowrate is also observed in discrete points and is always
corrupted in practice by some form of error — therefore, the
results will further vary depending on the way how we make
the transition from discrete observations into the "convolution
kernel": q(t-τ).
As noted by Lamm, the convolution operation has a natural
smoothing effect, therefore significantly different K(t) func-
tions can (and may) reconstruct essentially the same ∆p(t)
response for a given rate history. Ideally, K(t) ≥ 0 and
decreases monotonically (Coats, et al.3 (showed that
derivatives of higher order must remain monotonic)). Un-
fortunately, even using downhole measurements, measured
(observed) flowrates contain effects that may cause deviation
from monotonic behavior (e.g., wellbore storage or other
variable-rate conditions). If the response contains a skin effect
and no wellbore storage, then the impulse response (K(t)) will
contain a Dirac delta component.16 Therefore, the conceptual
representation of the K(t) function is not trivial, a point often
established by numerous failed attempts to represent this
function. In particular, many algorithms substitute the con-
volution integral in Eq. 1 by the trapezoidal rule (or some si-
milar approach) that is difficult to justify near the origin.
Development of a New B-Spline Based
Deconvolution Method
Spline functions8 are piecewise polynomial functions which
are defined on subintervals connected by points called knots
(ti) and have the additional property of continuity (of the
function and its derivatives). In this work we represent the
unknown K(t) function as a second order spline with loga-
rithmically spaced knots. To reveal characteristic reservoir
behavior, the number of knots should be on the order of at
least 2–6 knots per log cycle (for typical kernels of interest
(i.e., input flowrate functions observed in reservoir engineer-
ing)). Therefore, we never use more knots, (but we may be
forced to reduce the number of knots in cases where the data
quality does not justify looking for sophisticated signatures.)
A spline function can effectively be represented using a linear
combination of basic spline functions called B-splines8. Once
the knots (or continuity points) are set, generation of B-splines
is easy because of their intrinsic recurrence relation. Distribut-
ing the knots logarithmically:
t i =b i , b > 1 i = 0 ,±1,±2 ,....
With a suitable selected b > 1 basis, the B-spline of degree 0
is defined by:
Bi0 (t ) = {
1
0
t i < t < t i +1
otherwise .......................................................... (2)
Higher degree B-splines are generated recursively using:
3
⎡ t − ti ⎤ k −1 ⎡ t i + k +1 − t ⎤ k −1
Bik (t ) = ⎢ ⎥ Bi (t ) + ⎢ ⎥ Bi +1 (t ) .................(3)
⎣ ti+k − ti ⎦ ⎣ t i + k +1 − t i +1 ⎦
Where k = 1,2,…. The "i-th" B-spline has a non-zero value
(support) between ti and ti+k, as is illustrated in Appendix A.
Any "k-th" degree spline function (over the given system of
knots) can be represented as a linear combination of k-th order
B-splines.
u
S (t ) = ∑ c i Bik (t ) ...................................................................(4)
i =l
Where the number of B-splines involved is u-l+1. For second
order splines having support in an interval (ti1, ti2), we obtain
l=i1-2, u= i2-1, and the number of B-splines involved is n=i2-
i1+2. By linearity, the Laplace transform of Eq. 4 is:
u
S ( s ) = ∑ c i Bik ( s ) ..................................................................(5)
i =l
We will represent K(t) as a weighted sum of B-splines of
degree 2, defined over logarithmically evenly-spaced knots:
u
K (t ) = ∑ c i Bi2 (t ) ...................................................................(6)
i =l
Substituting Eq. 6 into Eq. 1, we have:
t u
∆p (t ) = ∫ ∑ c i Bi2 (τ ) q(t − τ ) dτ ..............................................(7)
0 i =l
If we have m drawdown observations collected in a vector ∆p ~
~ ~ ~
that were observed at times ( t1 , t2 ,... tm ), then the problem
can be written as an overdetermined system of linear
equations:
~ .....................................................................(8)
W X c = W ∆p
Leading to the linear least-squares problem:
~ − X c) T W T W ( ∆ p
~ − Xc) ........................................(9)
min ( ∆p
c
Where X is the m x n sensitivity matrix and c is the n-vector of
unknown coefficients and WTW is the m x m diagonal
weighting matrix. (In general, we select the weights as the
measured pressure drawdowns assuming a constant level of
relative errors in the drawdowns.) Naturally, we assume that
m >> n and the observations are distributed approximately
evenly with respect to the logarithmically spaced knots.
At this point we can assume that the observed rates can be
conveniently represented by a representative function for
which the Laplace transform can easily be obtained. We de-
note the Laplace transform of function f(t) by the operator L
that acts on a function of time and creates a function of the
Laplace variable s:
f ( s ) = L [ f (t )][s ] ................................................................(10)
For the inverse Laplace transformation we use the notation
[ ]
f (t ) = L-1 f ( s ) [t ] ..............................................................(11)
When we cannot construct the function f(t) analytically, the
inversion can be still done numerically, that is we can
4 SPE 95571
calculate the value of f(t) at any given t using an appropriate
numerical inversion algorithm for the Laplace transform.
In the Laplace domain Eq. 7 can be written as
∆p (s ) = ∑ c i q ( s ) Bi2 ( s ) .................................................... (12)
u
i =l
And hence the elements of the sensitivity matrix become
u
⎣i = l ⎦
[ ]
X j ,i = L−1 ⎡⎢ ∑ q (s )Bi2 ( s ) ⎤⎥ t j ............................................ (13)
~
We note that calculating the elements of the sensitivity matrix
is reduced to computing the convolution of the known rate
with the B-splines. To construct the sensitivity matrix we
require:
1. The Laplace transform of the second order B-splines
(over even logarithmically-spaced knots),
2. A simple description of the rate that lends itself to an
analytical (i.e., functional) Laplace transformation;
3. A numerical Laplace inversion algorithm that can provide
results with any required accuracy and is tolerant to the
fact that the B-splines are defined piecewise (in other
words, the inverse Laplace transform (as a function of t)
will contain "jump" features in its higher derivatives).
The Laplace transform of the B-splines is generated by
symbolic algebra manipulation software (see Appendix A). In
cases where rate data are smoothly changing, the rate is
described as a linear combination of terms
q case of smoothly varying drawdown followed by shut-in at tsh.
qk (t ) = 1 − exp(−t / t k ) ...................................................... (14)
That is
r
q(t ) = ∑ d k q k (t ) ................................................................ (15)
k =1
q
Where, t k (knot) values are even logarithmically-distributed
over the interval of interest. Therefore, the Laplace transform
of the rate is easily obtained as
r ⎛1 tq ⎞
q ( s ) = ∑ d k ⎜⎜ − k q ⎟ .................................................. (16)
s ⎟⎠
k =1
⎝ s 1 +t k
In other cases (such as permanent downhole gauge data) we
simply use a piecewise linear approximation or a piecewise
constant rate approximation over a finite interval (segment).
The success or failure of this approach depends primarily on
the numerical Laplace transform inversion. For the previous
three decades fixed precision computing has defined the status
of numerical inversion (see Davies and Martin17 Narayanan
and Beskos18). Recently, several new algorithms were intro-
duced using multi-precision methods (Valko and Abate14 and
Abate and Valkó15).
In this work we use the publicly available GWR algorithm:
http//library.wolfram.com/database/MathSource/4738/
With this algorithm it is possible to invert a large class of
Laplace transforms with essentially any desired accuracy.
Since the algorithm relies on multi-precision computing, it can
be realized only in software systems supporting variable
precision methods for arithmetic and special functions.
Once the sensitivity matrix (X) has been constructed, the
estimate of the unknown parameter vector (c) is obtained from
~ ..............................................................................(17)
ˆc = X + p
Where X+ (X+ = (XTX)-1XT) denotes the pseudoinverse of
matrix X. (For simplicity, here we show the case where all the
weights are unity.). To create the pseudoinverse, small sin-
gular values are removed from the calculations (in our com-
putations the automatic cut-off selection of Mathematica was
employed). Using the c coefficients estimates obtained from
this process, we reconstruct the constant-rate response as
follows:
p u (t ) = ∑ ĉ i Bi2,int (t ) ............................................................(18)
u
i =l
And the logarithmic derivative of the constant-rate response is
given as:
p ud (t ) = t ∑ ĉ i Bi2 (t ) ............................................................(19)
u
i =l
Where the integral of the second order B-spline, Bi2,int (t ) , is
easily obtained in symbolic form (e.g., by Mathematica).
The method described above works for smoothly varying
rates. Whenever the flowrate (or its derivative) undergoes an
abrupt change, the required condition for the successful nu-
merical inversion of the required Laplace transform (in Eq. 13)
is not satisfied. Such a situation emerges, for instance in the
In such a case we need to calculate the sensitivity matrix from
two contributions:
[] [ ]
X j ,i = L−1 ⎡⎢ ∑ q (s )Bi2 ( s )⎤⎥ t j − L−1 ⎡⎢ ∑ qsh (s )Bi2 ( s ) ⎤⎥ t j − tsh ........(20)
u
~ u
~
⎣ i = l ⎦ ⎣ l
i = ⎦
Where q sh (s ) is the Laplace transform of q( ∆t + t sh ) , taken
with respect to the time variable ∆t . For the specific form of
the flowrate given by Eq. 16, the Laplace transformation
results in the following form:
⎛ ⎞
q q
r ⎜ 1 exp(−t sh / t k )t k ⎟
q sh ( s ) = ∑ d k ⎜ − ⎟ .............................(21)
k =1 ⎜ s
⎝ 1 + t kq s ⎟
⎠
As a result of this construction, the "early" B-splines will
appear twice in the reconstruction of the observed pressure. In
general, we consider any flow history consisting of production
and shut-in pairs — however, the shut-in portion of the
sequence can be of zero duration. In this way we can "split"
the flow history into any number of segments — with each
segment having its own flowrate function (and corresponding
two Laplace transforms, one for production and one for the
shut-in part). The selection of "flow segments" is completely
independent from the location of the pressure observations.
For strongly varying flowrates we can use a piecewise con-
stant or a piecewise linear approximation.
SPE 95571 5

Decoupling the constant-rate response and its logarithmic
derivative
As we already mentioned, in general, the earliest part of the
unknown K(t) function is critical, especially if the observed
variable-rate response contains a skin effect (but minimal (or
no) wellbore storage effects). In such cases the K(t) function
contains a Dirac delta component. To reconcile this condition
we add additional "anchor" B-splines to the left of the first ob-
served time point. After considerable experimentation, we es-
tablished that 4 (four) additional B-splines are sufficient for
case, specifically the case of a reservoir model with a strong
characteristic behavior, we selected a dual porosity reservoir
with circular boundary (λ = 1 x10-5, ω = 5 x10-2 and reD = 1600
— see Table 1 for details). To generate the synthetic pressure
histories required for the validation phase, we convolve the
known unit rate pressure solution with the flow history —
which includes two variable rate production and two pressure
buildup periods.
In our validation we use the following variable-rate profile:
−t

virtually any scenarion. q(t ) = 405.23 − 736.79 e 5

These "anchor" B-splines do not affect the reconstructed pud(t)
function, that is the logarithmic derivative at the observation
points, but do affect the reconstructed pu(t) constant-rate res-
ponse at all observation points (i.e., the integral of the pud(t)
function). In fact, these additional B-splines are a convenient
vehicle to represent the Dirac delta component of the function
K(t).
Regularization
We found that using the concept of a pseudoinverse (a cut-off
for small singular values) does not provide a sufficient re-
gularization as the noise level in the data increases. Therefore,
we require an additional regularization which makes sense in a
physical context and ensures the relevance of the spline repre-
sentation.
For the overdetermined system (Eq. 8) to be solved by least
squares, for each spline interval we must append the following
two conditions:
−t
20
+ 368.39 e − 3.684 t KKKK 0 < t ≤ 60 hr
0 KKKKKKKKKKKK 60 < t < 100 hr
(100 − t )
300 e 100 KKKKKKKK100 ≤ t ≤ 200 hr
0 KKKKKKKKKKKK 200 < t < 300 hr
As we ultimately intend to process permanent (pressure) gauge
data, this rate profile was selecteded to mimic typical pressure
drawdown/buildup test sequences. In the validation process
we increase the complexity of the reservoir model step-by-step
to assess, what (if any) influence the underlying reservoir
model may have on the performance of our deconvolution
procedure. We first consider a model without wellbore sto-
rage and skin effects (Fig. 1), we then incorporate a skin factor
of 5 (Fig. 2), and finally we employ a dimensionless wellbore
storage coefficient of 100 and skin factor of 5 (Fig. 3).
Table 1 – Reservoir and fluid properties for Validation Cases 1-4

⎡ ⎤ Reservoir Properties:
α ⎢⎛⎜ t ∑ c i Bi2 (t ) ⎞⎟ − ⎛⎜ t ∑ c i Bi2 (t ) ⎞⎟
u u
⎥=0 Wellbore radius, rw = 0.3 ft
⎣⎝ i =l ⎠ t =t k ⎝ i =l ⎠ t = t k +1 / 2 ⎦ Net pay thickness, h = 30 ft
Formation permeability, k = 3.333334 md
⎡ ⎤
α ⎢⎛⎜ t ∑ c i Bi2 (t ) ⎞⎟ − ⎛⎜ t ∑ c i Bi2 (t ) ⎞⎟
u u
⎥=0 Formation compressiblity, ct = 5×10-6 1/psi
⎣⎝ i =l ⎠ t = t k +1 / 2 ⎝ i =l ⎠ t = t k +1 ⎦ Porosity, φ = 0.15 (fraction)
.............................................................................................. (22) Storativity ratio, ω = 5×10-2 (dim.less)
Interporosity flow parameter, λ = 1×10-5 (dim.less)
In other words, we require that the value of the logarithmic
Outer boundary radius, re = 480 ft
derivative of the constant-rate response differ only "slightly"
Initial reservoir pressure, pi = 8000 psi
between the knot and the middle location between knots. Wellbore storage coefficient, CD = 100 (dim.less)
When α = 0, there is no regularization, and with generated Skin factor, sx = 5 (dim.less)
data with practically no error, α = 0 results in the required
smooth solution for the constant-rate response and its Fluid Properties:
logarithmic derivative. In the presence of random noise and/or Fluid viscosity, µ = 1 cp
other inconsistencies, a positive α is selected based on an Formation volume factor, B = 1 RB/STB
informal interpretation of the discrepancy principle — i.e., we Production Parameters:
increase the value of the regularization parameter until the cal- Nominal production rate, qnominal = 350 STB/D
culated (model) pressure difference begins to deviate from the In the first example, we perform deconvolution with the
observed pressure difference in a specific manner. The mean input data shown in Fig.1. Fig. 4 presents our deconvolu-
and standard deviation of the arithmetic difference of the tion results for this case, and we note that when the decon-
computed and input pressure functions are also computed, but volution results are compared with the exact input response,
algorithmic rules (e.g., L-curve method) for selecting α are not our deconvolution methodology performs very well in this
recommended — for reasons discussed by Schroeter et al.5) case. All of the characteristic features of a dual porosity re-
Validation servoir model are present in the results — including the
boundary dominated flow regime.
In this section we validate our new method by comparing the
results from the deconvolution procedure with the exact res-
ponses obtained from a known reservoir model. As a base
6 SPE 95571

Figure 1 – Input data for Validation Case 1 (no wellbore

storage effects, skin factor (sx) = 0).

Figure 4 – Validation Case 1: B-Spline deconvolution results

for validation case 1 (all data used) (no wellbore
storage or skin effects).

In Fig. 5 we present the deconvolution results for the case

when a skin factor of 5 is included in the input pressure
response — we note an additional increase in the unit rate
drawdown pressure response (from the positive skin factor)
and we see that the well testing derivative function remains
essentially identically to the trends observed in Fig. 4 (as
would be expected).

Figure 2 – Input data for Validation Case 2 (no wellbore

storage effects, skin factor (sx) = +5).

Figure 3 – Input data for Validation Case 3 (dimensionless

wellbore storage coefficient (CD) = 100, and skin
factor (sx) = +5).
Figure 5 – Validation Case 2: B-Spline deconvolution results
for validation case 2 (all data used) (no wellbore
storage effects, skin factor (sx) = +5).
In Fig. 6 we present the deconvolution results when wellbore
storage and skin effects are incorporated into the base model.
SPE 95571
Figure 6 – Validation Case 3: B-Spline deconvolution results
for validation case 3 (all data used)
(dimensionless wellbore storage coefficient (CD)
= 100, and skin factor (sx) = +5).
We note that the deconvolution shown in Fig. 6 is performed
using "surface" rates — which do not reflect wellbore storage
conditions (although wellbore storage effects are (obviously)
included in the observed pressure data).
Fig. 6 is of particular practical interest because we clearly
must recognize that although deconvolution to remove well-
bore storage effects is a worthwhile goal — in current
practice, it is very unlikely that we would be able to obtain the
downhole flowrate data at sufficiently accuracy such that we
could "remove" wellbore storage effects.
Figure 7a – Validation Case 3: B-Spline deconvolution results
for validation case 3 (ONLY final pressure
buildup data are used for deconvolution) (dimen-
sionless wellbore storage coefficient (CD) = 100,
and skin factor (sx) = +5).
7
However, given a general variable-rate/variable pressure data
sequence, we can use deconvolution to reconstruct the
constant rate drawdown response which would include
wellbore storage effects (as shown quite effectively in Fig. 6).
From a theoretical perspective, we can perform deconvolution
using only the pressure buildups (which typically contain
higher quality data than drawdown periods (i.e., less data
noise)). And from a practical standpoint, in some cases only
pressure buildup data are available for analysis (in addition to
the production (flowrate) history). At this point we can not
construct the entire pressure history from a limited portion of
the available data (say from deconvolving only a single pres-
sure buildup test in a larger production/shut-in test sequence).
However, we will illustrate that we can construct a reasonable
portion of the entire pressure history from a single pressure
buildup test. In Fig. 7a we use only the final pressure buildup
test data for deconvolution — in fact we only used the first 60
hours (of the available 100 hours) in this pressure builup test.
We note in Fig. 7a that the entire "early portion" of the test
sequence is effectively deconvolved — but we do not obtain
the complete reservoir pressure signal because we truncated
the deconvolution data set. In particular, we did complete re-
construct the "late-time" boundary effects from the final
pressure buildup test. For clarity, we present the "base data"
for this deconvolution Fig. 7b (i.e., the first 60 hours of the
final pressure buildup test in this sequence (see Fig. 3 for
orientation)).
Figure 7b – Validation Case 3: Illustration of first 60 hours of
the final pressure buildup test sequence (data
used for deconvolution in Fig. 7a).
In the validation cases considered to this point only "ideal"
data have been used as input (i.e noiseless/consistent data) and
regularization was not necessary. As noted, the deconvolved
results are essentially the same as the (input) ideal model
responses. Regularization is not required for synthetic cases,
but for most (if not all) field cases (which usually have
inconsistencies and significant measurement errors) we do
require some sort of regularization procedure.
8 SPE 95571

We provide a synthetic case where we have added random and

systematic data errors (Fig. 8) (we note that we have
employed the same production history and reservoir model
with wellbore storage and skin effects (base performance (no
error) shown in Figs. 3 and 6)). In this particular example, we
have added 10 percent random error to both the pressure and
rate data, and we also seeded the data with a small systematic
error in pressure (see Fig. 8 for a summary of the input data).

Figure 10 – Validation Case 4: Noisy pressure observations

and the B-Spline pwf model response (optimal
regularization (α = 0.0007)).

Fig. 12 presents the results of deconvolution when errors and

inconsistencies are present in the data. Clearly, some artifacts
caused by inconsistent data are seen (particularly in the well
testing derivative), but deconvolution reconstructs the charac-
teristic features of the reservoir model for the duration of the
Figure 8 – Validation Case 4: Input data with systematic entire test sequence. (From another point of view, deconvolu-
error and noise (dimensionless wellbore storage tion without regularization can effectively reveal inconsisten-
coefficient (CD) = 100, and skin factor (sx) = +5).
cies present in the data.)
We can still perform deconvolution on these data, but in this
case, we must use regularization to address the combined
effects of random and systematic errors in the data.
The value of the regularization parameter has a profound
effect the deconvolution solution for this case. Figs. 9, 10,
and 11 illustrate the discrepancy principle by which the
regularization parameter, α is picked. When regularization is
not used (Fig. 9) there is a random error between the observed
response ( ∆p~ ) and its reconstructed value (Xĉ). If a higher
value of regularization parameter is used, over-smoothing is
obvious from the one-sided deviations (Fig. 11). The
optimum value of the regularization parameter (in this case α
= 0.0007) is obtained as the largest value still not causing
systematic discrepancy between observed and reconstructed
responses (Fig. 10). Figure 11 – Validation Case 4: Noisy pressure observations
and the B-Spline pwf model response ("over-
regularization" case (α = 0.005)).

From our validation experiments we can conclude that very

accurate (essentially exact) results can be obtained using any
major event in the test sequence (i.e., a production test, a shut-
in, or the complete history). We also note that when the input
data have errors or inconsistencies, we require some form of
regularization to provide meaningful resolution of these cor-
rupted data. As a comment, we believe that our deconvolution
approach can address relatively high levels of noise and other
data inconsistencies — further proof of this hypothesis is
found in our field examples given later in this work.

Figure 9 – Validation Case 4: Noisy pressure observations

and the B-Spline pwf model response (no
regularization).
SPE 95571
Figure 12 – Validation Case 4: Deconvolved pressure res-
ponse (optimal regularization (α = 0.0007)).
In the deconvolution process the the initial pressure is required
to create the observed pressure response ( ∆p~ ). As it is pointed
by Levitan et al,19 an accurate estimate of the initial pressure is
a critical element of deconvolution. We also confirm this sta-
tement with our experiments — in particular, using a signi-
ficantly erroneous estimate of initial reservoir pressure will
cause a systematic error in the deconvolved, equivalent
constant-rate pressure response. For example, using an initial
pressure larger than the actual initial pressure will result in an
apparent pseudosteady-state flow regime at the end of the
deconvolved test sequence or production data set. Thus, we
note that that an apparent pseudosteady-state flow regime is a
possible artifact of using an incorrect estimate of the initial
reservoir pressure. We will also discuss (later) that a repre-
sentative production pressure, take very early in the product-
ion phase, can serve as a surrogate for the initial reservoir
pressure.
Applications of the New Deconvolution Technique
Having verified our new deconvolution technique, we can now
apply the same procedure to field data. For convenience, we
classify our field examples into four groups:
z Wellbore storage distorted pressure data.
z Well test data including multiple flow sequences.
z Permanent downhole gauge data (several flow sequences).
z Long-term production data (rates and pressures).
These examples illustrate the use of B-spline deconvolution
for analyzing various events (production/shut-in sequences),
and should be considered typical of cases that could be en-
countered in field operations.
Field Case 1: B-Spline deconvolution of pressure transient
data distorted by wellbore storage
The most straightforward application of deconvolution is to
deconvolve pressure buildup data in order to eliminate the
effects of wellbore storage. Fetkovich et. al.20 published data
9
on an oil well which was hydraulically fractured at initial
completion. The early time pressure response is distorted by
wellbore storage effects, but sandface rates are available.
Using this input we perform deconvolution, the results of
which are shown in Fig. 13.
Figure 13 – Field Case 1: Wellbore storage distorted pressure
response, deconvolved response and model
match (Data from SPE 12179 (ref. 20)).
We compare the deconvolved response along with reservoir
model responses (simulation) for consistency. Fig. 13 pre-
sents the distorted pressure data, the deconvolved constant-rate
pressure drawdown and its logarithmic derivative, as well as
model matches for the deconvolved response. We used two
models to match the deconvolved pressure response — the
uniform flux fracture (Fetkovich et. al used this model in their
analysis) and the finite conductivity vertical fracture model.
Our deconvolution results compare extremely well with these
model responses and serves as confirmation of previous
interpretations.
In this case we have used deconvolution to eliminate wellbore
storage effects, and we suggest that this case serves to verify
our contention that our method is sufficiently accurate for use
in eliminating wellbore storage effects (when sandface rates
are available).
Field Case 2: B-Spline deconvolution of conventional well
tests including multiple flow periods (complex
gas reservoir)
In Fig. 14 we present the rate and pressure data obtained
during the four-point test of a gas well, which was followed by
an extended shut-in.
In this case the pressure buildup data seem to be of sufficient
duration for a conventional pressure buildup (or PBU) analysis
— therefore we could simply proceed and analyze the indivi-
dual PBU portion of this data. However, our intention is not
to replace PBU analysis by deconvolution, but rather, to im-
plement our deconvolution algorithm as an option for well test
analysis.
10
Figure 14 – Field Case 2: Moderate pressure gas well in a
complex reservoir.
We choose to analyze all the available data including the
production periods. We perform deconvolution and present
the results in Fig. 15, together with model matches of the
deconvolved pressure drawdown.
Since the flowing fluid is gas we must perform the pseudo-
pressure transform to adhere to theoretical considerations (we
note that pseudotime should also be considered, but to be
theoretically rigorous, pseudotime requires the average
reservoir pressure history (which is never available in prac-
tice). The deconvolved pressure response suggests that a
complex channel type reservoir geometry is possible.
Therefore, we proceed and match the deconvolved response
with two models to verify this possibility. The models consi-
dered include a well in a homogeneous reservoir bounded by
parallel faults and a well in a homogeneous reservoir with
closed rectangular boundaries. The results of our model
matches validate the complexity of the reservoir. We do ob-
serve the familiar "half-slope" trend of the pressure and
pressure derivative functions on the log-log plot, which indi-
cates a channel type reservoir geometry.
On the other hand, we also note that the signature of a closed
reservoir is evident in the pseudopressure drop and
pseudopressure drop derivative functions at late times.
In this application we have used deconvolution to extract
information from a larger time interval than just the pressure
buildup test sequence. Analysis of the entire testing sequence
using deconvolution may reveal reservoir characteristics not
"seen" by a single event in the sequence.
SPE 95571
Figure 15 – Field Case 2: Deconvolved response and model
match (entire history used in this deconvolution).
Field Case 3: B-Spline deconvolution of conventional well
tests including multiple flow periods (low
permeability oil reservoir)
In our third example we analyze only a selected flow period
during a well test sequence as shown for the selected oil well
in Fig. 16. This test sequence includes 170 hours of data
where we have a 50 hour constant-rate production (at 200
STB/D) followed by a shut-in for 16 hours, then production
resumes at a constant rate of 160 STB/D for 24 hours and is
then shut-in for 80 hours.
Figure 16 – Field Case 3: Input data for fractured oil well.
Inconsistencies are observed — in particular, we suspect that
in the first production period the given rate is not correct (we
note fluctuations in the drawdown pressure profile). We have
two options in this case — first we can deconvolve the entire
sequence, but we have to use a high value of regularization
parameter to overcome the effects of inconsistencies. Or we
can focus on the final buildup period, but taking the preceding
flow history into account. We choose the second option.
The deconvolved pressure response and a model match are
shown in Fig. 17. The deconvolved well-testing pressure
derivative function suggests that the well has a finite con-
ductivity vertical fracture that this portion of the reservoir is
bounded by parallel faults. We confirm this supposed reser-
SPE 95571
voir model by matching the deconvolved pressures with just
such a model. This example verifies that we can perform de-
convolution based on a selected flow period.
Figure 17 – Field Case 3: Deconvolved response and model
match (ONLY final PBU data used in this decon-
volution).
Field Case 4: B-Spline deconvolution of permanent down-
hole gauge data
Our next example is taken from an oil well equipped with a
permanent downhole measurement system which is used to
provide continuous (downhole) pressure and rate data. In this
case a tremendous amount of data is available for a very short
time — 10 hours (as shown in Fig. 18).
Nevertheless, pressure buildup analysis or other traditional
type analysis is not a realistic option because the buildups are
very short and, in addition, other phenomena (most likely
temperature change and/or phase resegregation) cause non-
intuitive behavior (e.g., between 2-4 hours the well is shut-in
but the observed pressure is decreasing).
Figure 18 – Field Case 4: Permanent downhole gauge data
for 10 hours.
Fig. 19 presents the results of deconvolution for this case. The
apparent oscillations in deconvolved pressure derivative
function indicate that the data are somewhat inconsistent —
11
but we can proceed and match the deconvolved pressure with
simple models (well in an infinite acting homogenoeuos
reservoir with wellbore storage and skin / well in a homo-
genoeuos reservoir with single sealing fault (with wellbore
storage and skin)).
Figure 19 – Field Case 4: Deconvolved response and model
match (entire data sequence used).
Using analytic simulation, we reproduce the pressure response
relative to the observed rate history for the entire test sequence
as shown in Fig. 20. The pressures generated by the model
fail to honor observed data at some points (especially in the in-
tervals where we already revealed inconsistencies) — but, the-
se simulations do honor the observed pressures at later times,
in agreement with our expectations.
Figure 20 – Pressure history match using the models shown
in Fig. 19.
Field Case 5: B-Spline deconvolution applied to production
data (tight gas reservoir)
Our deconvolution method is not limited to analyzing well
tests — we can also apply this methodology to analyze and
interpret traditional (long-term) production data. Production
data is regarded as a "low frequency" and "low resolution"
type of data. Because of these reasons, rigorous variable
12
rate/pressure analysis of production data is rarely attempted.
Our objective is to "convert" the entire production sequence
(including poor quality data) into an equivalent constant flow-
rate pressure response. We corroborate our results with
material balance deconvolution21 which includes plotting
∆pp/qg (rate normalized pseudopressure) versus material balan-
ce pseudotime, and comparing our deconvolution results and
the material balance deconvolution results with a reservoir
model established from the deconvolved data.
This example considers a tight gas reservoir (Pratikno et al.22).
The data are of good (to excellent) quality, and we note the
effect of shut-ins and daily production fluctuations (Fig. 21).
Figure 21 – Field Case 5: Production data history plot for
East Texas Gas Well.
In Fig. 22 we present the deconvolved responses (our method
and material balance time) compared with an appropriate
reservoir model (well with a finite conductivity vertical frac-
ture in a bounded reservoir), and we note good agreement.
Figure 22 – Field Case 5: Deconvolution response functions,
model match and material balance time normal-
ized data.
We note some anomalies in the well testing pressure
derivative function at early times caused presumably by incon-
sistencies in the data (or simply a sparcity of data at early
time). It is seen from the derivative trend that boundary domi-
SPE 95571
nated flow is in transition to full development, and there does
exist a slight difference in our interpretation of the data and
model for this portion of the analysis.
Field Case 6: B-Spline deconvolution applied to production
data (low productivity gas reservoir)
The second field example is also a gas field case (Palacio et.
al.23). The daily production data are quite erratic (Fig. 23) due
to liquid loading, in spite of these features, the B-spline
deconvolution method performs reasonably well (i.e., the data
deconvolution data functions appear to be reasonable and
well-behaved (see Fig. 24)).
Figure 23 – Field Case 6: Production data history plot for Mid
Continent Gas Well.
In Fig. 24 we again compare our deconvolution results with
traditional production analysis based on material balance time
deconvolution.
Figure 24 – Field Case 6: Deconvolution response functions,
model match and material balance time normal-
ized data.
Artifacts caused by inconsistencies are seen at early times in
the well testing pressure derivative function (again (we
believe) due to sparse/erratic data). The late time well testing
pressure derivative function trend confirms that boundary-
dominated flow regime has been reached. The solution for a
homogeneous reservoir with closed circular boundary matches
SPE 95571 13

all of the deconvolved pressure response functions reasonably pu = Constant (unit) rate pressure response, psi
well, which also confirms our analysis. pud = Well-testing (log) derivative of pu, psi
pi = Initial reservoir pressure, psi
We believe that B-spline deconvolution can be successfully
q(t) = Rate, STB/D or MSCF/D
applied to production data (regularly measured flowrate and re = Reservoir outer boundary radius, ft
pressure data). Processing "low quality" production data with rw = Wellbore radius, ft
deconvolution can provide additional insight into reservoir reD = Outer reservoir boundary radius, dimensionless
performance based analysis. s = Laplace transform variable
It is worth noting that numerous other field cases of long-term sx = Skin factor
flowrate and (surface) pressure data have been successfully S(t) = Spline function
analyzed/interpreted using the proposed B-spline deconvolu- t = Flowing time, hr
tion technique. In fact, we believe that B-spline deconvolution tsh = Shut-in time, hr
may be more utilized for the analysis of production data, com- X = Sensitivity matrix
pared to its potential use for well test data. This hypothesis W = Weighting matrix
arise from the fact that while often poor quality, production Greek Symbols
data are generally available. α = Regularization parameter
Conclusions ∆p = Drawdown (with respect to pi)
φ = Porosity, fraction
Our proposed deconvolution technique is unique because of
µ = Viscosity, cp
the construction of the coefficient matrix of the underlying
ε = Small positive time
least-squares problem using B-splines. Due to the (semi- λ = Interporosity flow parameter
logarithmic) B-spline representation of the unknown impulse ω = Storativity parameter
response and to the high reliability of the numerical Laplace τ = Dummy variable
transform inversion technique used, the elements of the coeffi-
cient matrix are computed accurately and the resulting Superscripts
deconvolution is robust and consistent data. In this work we ^ = Estimated (from least squares)
have demonstrated the robustness of this new technique on ~ = Observed
—
various exaples of increasing complexity (including synthetic = Laplace transform
+
and field examples). = Pseudoinverse
T
= Transpose of a matrix
For the deconvolution of data in practice, it is necessary to
combine implement a physically sound regularization scheme SI Metric Conversion Factors
into our proposed deconvolution method. Regularization cp × 1.0 E-03 = Pa·s
might be seen as an "add-on" to the proposed algorithm, but in ft × 3.048 E-01 = m
a practical sense (i.e., for field applications), our proposed de- md × 9.869 233E-04 = µm2
convolution scheme is likely to perform poorly without the use psi × 6.894 757E+00 = kPa
of regularization methods. bbl × 1.589 873E-01 = m3
*Conversion factor is exact.
We have successfully demonstrated the used of B-spline
deconvolution for the analysis of a wide range of field data — References
from traditional pressure buildup data, to permanent downhole 1. Lamm, P. K.:"A Survey of Regularization Methods for
gauge data, to traditional production data. First-Kind Volterra Equations," in Surveys on Solution
Methods for Inverse Problems, ed. D. Colton, H. W. Engl,
Nomenclature A. Louis, J. R. McLaughlin, W. Rundell, Springer (Vienna,
Variables New York), pp 53-82, 2000.
b = Base of logarithmically evenly distributed knots 2. Stewart, G., Wittmann, M.J., Meunier, D.: "Afterflow
B = Formation volume factor, RB/STB Measurement and Deconvolution in Well Test Analysis,"
paper SPE 12174 presented at the Annual Technical
Bik (t ) = k-th degree B-spline starting at bi Conference and Exhibition, 5-8 October, San Francisco,
Bik,int (t ) = Integral of the k-th degree B-spline California, 1983
3. Coats, K.H., Rapoport, L.A., McCord, J.R., Drews, W.P.:
c = Vector of unknown coefficients "Determination of Aquifer Influence Functions From Field
ct = Total system compressibility, psi-1 Data," J. Pet. Tech. 1964 (December) 1417-1424
CD = Dimensionless wellbore storage coefficient 4. Kuchuk, F. J. Carter, R. G., Ayestaran, L,: "Deconvolution
dk = Coefficient of Wellbore Pressure and Flow Rate," SPEFE 1990
Fc = Fracture conductivity, md-ft (March) 53-59
Gp = Cumulative gas production, MSCF 5. von Schroeter, T., F. Hollaender, and A.C. Gringarten:
h = Pay thickness, ft "Deconvolution of Well-Test Data as a Nonlinear Total
k = Permeability, md Least-Squares Problem," SPE Journal 2004, v 9, n 4,
K(t) = Impulse response ( = pud / t ) 375-390
p(t) = Pressure, psi
14 SPE 95571

6. Levitan, M.M.: "Practical Application of Pressure-Rate Deconvolution," paper SPE 21503 presented at the Gas
Deconvolution to Analysis of Real Well Tests," paper SPE Technology Symposium, 22-24 January, Houston,
84290 presented at the Annual Technical Conference and Texas,1991
Exhibition, 5-8 October, Denver, Colorado 2003. 22. Pratikno, H., Rushing, J.A., and Blasingame, T.A.:
7. Gringarten, A. C. von Schroeter T., Rolfsvaag, T and "Decline Curve Analysis Using Type Curves: Fractured
Bruner, J.: "Use of Downhole Permanent Pressure Gauge Wells," paper SPE 84287 presented at the 2003 Annual
Data to Diagnose Production Problems in a North Sea SPE Technical Conference and Exhibition, Denver, CO.,
Horizontal Well," paper SPE 84470 presented at the 05-08 October 2003.
Annual Technical Conference and Exhibition, 5-8 October, 23. Palacio, J.C. and Blasingame, T.A.: "Decline Curve
Denver, Colorado, 2003. Analysis Using Type Curves — Analysis of Gas Well
8. Cheney, E. W, Kincaid, D. R.: Numerical Mathematics and Production Data," paper SPE 25909 presented at the 1993
Computing, Brooks Cole, 2003. Joint Rocky Mountain Regional/Low Permeability
9. Roumboutsos, A., Stewart, G.: "A Direct Deconvolution or Reservoirs Symposium, Denver, CO, 26-28 April 1993.
Convolution Algorithm for Well Test Analysis," paper SPE
18157 presented at the Annual Technical Conference and Appendix A — B-spline and Its Laplace Transform
Exhibition, 2-5 October, Houston, Texas 1988. The zero-th quadratic B-spline (defined over logarithmically
10. Onur, M., Reynolds, A.C.: "Well Testing Applications of spaced knots with basis b > 1) is given by Eq. A.1, where
Numerical Laplace Transformation of Sampled-Data," SPE θ (x) denotes the Heaviside function. The Laplace transform
Reservoir Evaluation & Engineering 1998. v 1, no 3, 268- of the above B-spline can be written in a compact form (Eq.
277 A.2). In Fig. A.1 we show two B-splines (of order 2, basis b =
11. Cheng, Yueming, Lee, W. J. and McVay, D. A.: "A 2 and indices 0 and 1) — we note that the zero-th index B-
Deconvolution Technique Using Fast-Fourier Transforms," spline is non-zero only in the interval (1,8). An interesting
paper SPE 84471 presented at the Annual Technical property is that the Heaviside function can be represented as
Conference and Exhibition, 5-8 October, Denver, Colorado the sum of all B-splines with indices from negative to positive
2003. infinity (this is the "partition of unity" property).
12. Gilly, P, Horne, R. N.: "A New Method for Analysis of
Long-Term Pressure History," paper SPE 48964 presented ⎛ ( x − 1)(θ ( x − 1)-θ ( x − b)) ⎞
at the Annual Technical Conference and Exhibition, 27-30 ( x − 1) ⎜ ⎟
⎝ b −1 ⎠
September, New Orleans, Louisiana,1998, B02 (t ) =
13. Mendes, L.C.C., Tygel, M., Correa, A.C.F.: "A Decon- b 2 −1
volution Algorithm for Analysis of Variable-Rate Well ⎛ (b 2 − x)(θ ( x − b)-θ ( x − b 2 )) ⎞
( x − 1) ⎜⎜ ⎟
⎟
Test Pressure Data," paper SPE 19815 presented at the b2 − b
⎝ ⎠
Annual Technical Conference and Exhibition, 8-11 +
October, San Antonio, Texas, 1989 b 2 −1
14. Valkó, P.P. and Abate, J. "Comparison of Sequence ⎛ ( x − b)(θ ( x − b)-θ ( x − b 2 )) ⎞
(b 3 − x) ⎜⎜ ⎟
⎟
Accelerators for the Gaver Method of Numerical Laplace b2 − b
⎝ ⎠
Transform Inversion" Computers and Mathematics with +
Application 2004 v.48 pp.629-636. b3 − b
15. Abate, J. and Valkó, P.P.: "Multi-precision Laplace ⎛ (b 3 − x)(θ ( x − b 2 )-θ ( x − b 3 )) ⎞
(b 3 − x) ⎜⎜ ⎟
⎟
transform inversion" International Journal for Numerical b3 − b2
⎝ ⎠
Methods in Engineering, 2004 v. 60, pp 979–993. +
16. Van Everdingen, A.F. and Hurst, W.: "Application of the b3 − b
Laplace transformation to flow problems in reservoirs," .............................................................................................(A.1)
Trans, AIME (1949) v.186, 305-24. 2
−b 3 ) s 2 2
2 2 e ( −1−b −b (b 3 e b (1+ b + b )s
+ (1 + b + b 2 )e b (1+ b + b )s
17. Davies, B. and Martin, B.: "Numerical inversion of the B 0 ( s) =
Laplace transform: A survey and comparison of methods," (b − 1) 2 b 2 (b + 1) s 2
J. of Computational Physics, 1979 (v. 33) 1 -32. 2
−b 3 ) s 2 2
2 e ( −1−b −b (e (1+ b + b )s
+ b (1 + b + b 2 )e (1+ b + b )s
18. Narayanan, G.V. and Beskos, D.E.: "Numerical operational −
methods for time-dependent linear problems," (b − 1) 2 b 2 (b + 1) s 2
International Journal for Numerical Methods in .............................................................................................(A.2)
Engineering, 1982 v. 18 pp.1829 - 1854.
19. Levitan, M.M., Crawford, G.E., Hardwick, A.: "Practical
Considerations for Pressure-Rate Deconvolution of Well
Test Data," paper SPE 90680 presented at the Annual
Technical Conference and Exhibition, 26-29 September,
Houston, Texas 2004.
20. Fetkovich M.J. and Vienot M.E.: "Rate Normalization of
Buildup Pressure By Using Afterflow Data," J. Pet. Tech.
1984 (December) 2211-2224.
21. Johnston, J.L., Lee, W.J.: "Interpreting Short-Term
Buildup Tests From Low-Productivity Gas Wells Using
SPE 95571 15

pressure buildup data. While obvious, it is worth restating that

an accurate production history is absolutely critical for decon-
volution.
Fig. B.1a presents the input data for deconvolution and
pressure buildup analysis.

Figure A.1 – B-splines over logarithmically equidistant knots,

B02 (t ) and B12 (t ) (with base b = 2).

Appendix B — Test of New Deconvolution Method

for only Pressure Buildup Data Figure B.1a – Appendix B Case 1: Input data for the first case
where the duration of the buildup is hundred
In this appendix we will address the issues when there are times larger than the duration of drawdown.
significant differences between the durations of drawdown and
buildup tests. For our purposes we have generated 2 (two) In Fig. B.1b we immediately note artifacts in the very late-
synthetic cases where we select a simple homogeneous time pressure derivative of the pressure buildup case, where
reservoir model with wellbore storage effects for simplicity. these artifacts are caused by fact that the reservoir pressure has
The properties for the base reservoir model are provided in completely built-up to the limiting value of the average reser-
Table B.1: voir pressure.
Table B.1 – Reservoir and fluid properties for Cases B1-B2
Reservoir Properties:
Wellbore radius, rw = 0.3 ft
Net pay thickness, h = 30 ft
Formation permeability, k = 3.333334 md
Formation compressiblity, ct = 1×10-5 1/psi
Porosity, φ = 0.15 (fraction)
Outer boundary radius, re = 480 ft
Initial reservoir pressure, pi = 4800 psi
Wellbore storage coefficient, C = 0.001 RB/psi
Skin factor, sx = 0 (dim.less)
Fluid Properties:
Fluid viscosity, µ = 1 cp
Formation volume factor, B = 1 RB/STB
Production Parameters:
Production rate 1, q1 (PBU-1) = 50 STB/D
Production rate 2, q2 (PBU-2) = 25 STB/D
In the first case, a constant rate production of 50 STB/D is
held for 5 hours and is followed by 500 hours shut-in. We Figure B.1b – Appendix B Case 1: B-Spline deconvolution
results for the first case with conventional pres-
specifically set the duration of the buildup to be one hundred sure buildup analysis and exact results (ONLY
times larger than the duration of the drawdown in order to pressure buildup data are used for deconvolu-
establish the difference between deconvolution and conven- tion).
tional pressure buildup analysis when the durations of product- However, our deconvolution method is not affected by this
ion sequences are not consistent. behavior. Using only the pressure buildup data for this case,
We recognize that this is an extreme test of the algorithm — we perform deconvolution and we then construct the constant
i.e., to use only pressure buildup data (plus the production rate pressure drawdown response for the entire history. We
history) to yield the equivalent constant reate drawdown res- present the results of the deconvolution as well as the conven-
ponse by deconvolution. However, in practice, the primary tional pressure buildup data functions in Fig. B.1b and we
utility of our algorithm is likely to be the resolution of compared these data functions to the correct model responses
16 SPE 95571

for each case.
As shown in Fig. B.2a, for our second case, we add another
production and shut-in period to the well test sequence. The
duration ratio between both drawdown and build-up sequences
is preserved (shut-in duration is hundred times larger than the
duration of the drawdown).
suggest that any reference pressure (other than initial reservoir
pressure) be taken as early as possible during production.
As such, we use the data from the second (final) pressure
buildup and a reference pressure taken a short time into the
initial production period as input for deconvolution in this
case. The results for this case are presented in Fig. B.2b.
It is worth noting that using only a single (early production)
pressure data point for reference and the pressure buildup data
from the second (final) pressure buildup test, we can recover
the equivalent constant rate pressure drawdown response for
the entire well test sequence using deconvolution.

Figure B.2a – Appendix B Case 2: Input data for the second

case where the durations of the buildups are
hundred times larger than the durations of
draw-downs.

In this case we use ONLY the data from the final pressure
buildup for the deconvolution process. As noted previously,
we can construct the constant rate pressure drawdown respon-
se for the duration of the buildup using deconvolution. As a
caveat, we note that all deconvolution methods require direct
knowledge of the initial reservoir pressure (this condition
(obviously) also applies to the process of convolution).

Figure B.2b – Appendix B Case 2: B-Spline deconvolution

results for the second case with conventional
pressure buildup analysis and exact results
(ONLY data from the final pressure buildup are
used for deconvolution).

In our work, we require a reference pressure — where this

pressure can be the initial reservoir pressure or any (single)
pressure taken during production. As a practical point, we