SPE 121485

Swabbing Test Interpretation Using Nonlinear Regression in San Jorge Gulf

Basin
Euver Naranjo, SPE; José Bravo, SPE; Eugenio Díaz, SPE; YPF, S.A; and José Caldera, SPE, Schlumberger
Oilfield Services

Copyright 2009, Society of Petroleum Engineers

This paper was prepared for presentation at the 2009 SPE Latin American and Caribbean Petroleum Engineering Conference held in Cartagena, Colombia, 31 May–3 June 2009.

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 have not
been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum
Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper 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 SPE copyright.

Abstract A correlation between dynamic permeability from

In the San Jorge Basin in Southern Argentina swabbing
test is the conventional testing technique used by YPF to
evaluate production potential on those reservoirs that do
not flow naturally to surface. During the well completion
phase, a swabbing device is lowered subsequently to
individually test each reservoir layer. These wells, which
produce from multilayer formations, are later completed
for commingled production. Accordingly, the swabbing
test represents the only opportunity to measure dynamic
properties for every individual layer.
So far pressure transient analysis of such test has been
limited to the use of conventional interpretation methods
applied to the Infinite Acting Radial Flow portion, on
those tests where such flow regime is observed. This can
happen if a bottomhole shut in valve is used or if the shut-
in period is given long enough time for the bottom hole
pressure to reach conditions close to initial reservoir
pressure. Unfortunately, both practices substantially
increase the completion cost, which strongly affects well
economics. As a result of this economic limitation, neither
of such practices are applied on a routine basis so most of
the time transient pressure interpretation is not performed.
In an attempt to get value from routine swabbing tests,
where Infinite Acting Radial Flow condition is not
observed, a different methodology of test interpretation
has been introduced. Through the computation of the
instantaneous flow rate from the pressure trend increase
during the swabbing period, it is possible to use nonlinear
numerical regression to make an estimation of the
reservoir permeability even if Infinite Acting Radial Flow
is not present.
With this innovative approach YPF is, for the first time,
constructing representative production models for each
well, improving reserve estimations and production
forecasts.
nonlinear regression and permeability from Nuclear
Magnetic Resonance Tool has also been constructed in
order to reevaluate production response in old wells.
Pressure and Flow Rate during Swabbing
Fig. 1 shows the typical pressure history on a swabbing
test for one layer, in this case on the well EG_638, layer
1233.5_1236.5 mts, eight swabbings were performed with
a final stabilized flowrate of 320 lts/hr (48 bpd) (reported
from rigsite) Initially (point A), the hydrostatic pressure is
controlling the well, the sudden pressure drop (A→B)
indicates the first downhole trip of the piston to remove
(lift) certain amount of fluid, this creates a pressure
differential that allows the reservoir to begin to flow,
gradual pressure increase (B→C) shows layer production
which fills up production pipe.
After a predetermined time has elapsed the piston is
lowered again into the wellbore to lift certain amount of
fluid which lowers the bottom hole pressure (point D)
leading to another wellbore fill up by the layer production,
this is repeated as many times as required until the
dynamic level and fluid recovered is stabilized.
Knowing fluid density and pipe internal diameter, it is
simple to convert pressure change (fluid level change) into
flow rate (Appendix A).
Fig. 2 shows pressure and flow rate for pressure trend of
EG_638 layer 1233.5_1236.5 mts calculated from
Appendix A.
Decreasing flow rate trend is a consequence of the
increasing level of fluid (increasing bottom hole pressure)
and of the transient nature of this flow due the short
duration of the swabbing test.
2 SPE 121485
Test Interpretation
Conventional pressure transient test interpretation theory is
based on pressure analysis during a constant flow rate
period (0 for buildup test or a fixed value for drawdown).
Well known techniques such as the semilog slope or the
log-log horizontal line are fully suitable when flowrate
does not change during the pressure interpretation period.
During swabbing test, this condition is not fulfilled by any
mean, the changing dynamic level of the fluid imposes a
variation of the bottom hole flowing pressure so flow rate
is never constant, as a result of this, it is not possible to
perform any conventional pressure interpretation.
In a attempt to get information from these unconventional
pressure – flowrate data, and based on the fact that quality
and traditional BU or DD data is generally not available,
an alternative approach has been proposed to estimate
formation dynamics properties as k and s.
Solution Approach
The multirate-pressure nature observed in Fig. 1 and Fig. 2
is quite analogous to what happens in a reservoir when a
dynamic formation tester is used. In both cases, pressure
and flowrate history are measured simultaneously and
none of both variables are constant during the test period.
Numerical models have been proposed2 to solve this sort
of problems, with special application in such cases where
an accurate flowrate history is not available. Nonlinear
numerical regression algorithms use a continous
optimization approach of the reservoir parameters and
flowrate history to match pressure history2.
Fig. 3 shows pressure and computed flowrate history
(from the Appendix A) for EG_638 layer 1233.5_1236.5
(as in Fig. 2) as input into the numerical model. Initial
pressure was measured during the open hole logging and
all other reservoir parameters (porosity, fluid viscosity,
total compressibility, etc.) were also available.
Numerical Model calculates flowrate history with the
initial set of optimized reservoir parameters; Fig. 4 shows
both input flowrate (blue line - from pressure data) and
calculated flowrate history by the regression (red line), this
plot helps as a parameter of control on how “good” and
accurate are flowrate history input and optimized reservoir
parameters.
Difference has to do with the change of the fluid produced;
if there is any variation in the density of the fluid this will
impact the pressure change and this will change
transmissibility of the layer, what the NM does is to
estimate a flowrate history based on the optimized k, s and
fluid viscosity, the more change of the fluid during the test
the more difference between both flowrates.
Pressure match (Fig. 5) is not very good, as computed
pressure is calculated with the input flowrate history.
Optimized parameters from this initial optimization run
are:
k: 21.8
Skin Factor, S: -0.049
The key point here is the comparison between input
flowrate and optimized flowrate after the first
optimization run on Fig. 4; as it can be noticed,
optimized flowrate (which is computed by the NM and
has been “cleaned” out of changing fluid effect) is close
to the input flowrate, so we could assume that this new
flowrate is the flowrate that would have existed if no
viscosity change had occurred, which leads us to take it
as valid flowrate history. If the change of fluid is too
severe and the difference between both flowrate histories
is important, it could be impossible to perform a valid
interpretation.
A second optimization run can be performed, in which a
perfect pressure match would be obtained and a more fine
tuned k and s would be calculated. Pressure match is
shown on Fig. 5.a
Optimized parameters from second optimization run are:
k: 22.5
Skin Factor, S: 0.137
Optimized parameters did not change too much from run 1
to run 2 and pressure match is now perfect (due the good
match of optimized and input flowrate after optimization
run 1), obviously match on flowrate will also be perfect on
optimization run 2 (Fig. 5.b).
Permeability and skin values from second optimization run
(22.5 md and 0.137) can be used in Nodal Analysis (using
the transient flow equation) to reproduce flowrate during
swabbing as a function of time and flowing pressure.
IPR curve (Fig. 6 and Fig. 6.a) calculated with the
transient test model after 8 hours (Fig. 5.b) shows good
match with measured pressure and computed flowrate (Q
= 55 bpd, Pwf: 140 psia) on Fig. 2.
No Flow Regime Identification
As stated before, conventional pressure transient test
interpretation theory is based on pressure analysis during a
constant flow rate period (0 for BU test or a fixed value for
DD), analysis technique is based on the recognition of
flow regimes on some kind of pressure vs time plot. As it
is clearly seen, swabbing test is not an application for such
technique due to its variable flow rate nature, which
implies that for such kind of data No Flow Regime
Identification is Possible.
The approach proposed in this work attempts to obtain
reservoir parameters from this uncommon type of data,
however it is necessary to verify how straightforward and
suitable this approach is; to achieve this it would be
SPE 121485
necessary to interpret a test where both, swabbing and
conventional data (BU or DD) is available and compare
the obtained parameters with both interpretation
techniques.
This is possible in two situations: if a conventional BU test
is performed after swabbing the reservoir or if the
reservoir transmissibility is good enough and the wellbore
fillup is let to last long enough for the bottom hole
pressure to reach a value close enough to the initial
reservoir pressure (in this case the flow rate will cease and
a conventional BU analysis could be performed).
Verifying Examples
Fig 7 shows the pressure trend and computed flowrates for
GBk_660 swabbing + BU test, this is a viscous oil well
(120 cp) where the shut in lasted 10 hrs after the swabbing
period of 4.5 hrs.
BU test showed a clean IARF with a k = 3270 md and a s
=6.2 (Fig. 8).
The described methodology with the numerical model was
implemented for this well, Fig. 9 shows pressure and
calculated flowrate as input in the NM.
Fig. 10 shows the optimized and input flowrate after the
first optimization run.
As it can be noticed on Fig. 10, during the first 1.5 hours
both flowrate histories show a poor matching but it gets
better from 2 hours on. This is a result of the changing
fluid effect. At the beginning of the swabbing, the layer
was producing mainly water so flowrate computed with
the 120 cp reservoir model can not match water flowrate,
this changes after 2 hours when oil begins to be produced
and the optimized flowrate better matches the oil flowrate.
This is in concordance with what was explained on Fig. 4,
the more homogeneous the fluid is the more reliable the
interpretation is.
Pressure match (Fig 10.A ) is obtained with k= 3255 md y
s=4.71, obviously this is not a good match since it is done
with the initially water affected flowrate.
Based on good match of both flowrates after 2 hours
(when water effect has disappeared), it could be performed
a second optimization run using the optimized flowrate
from optimization 1.
Optimized parameters k=3299 and s= 5.18 did not change
very much from optimization run 1 and agree acceptably
with the results obtained from the log-log analysis in Fig.
8: k = 3270 and s =6.2.Pressure match is shown on Fig.
11.
A second example is well CG_613 (layer 1563-6 m),
during the swabbing of this layer the last wellbore fill up
was let to last more than 3 hours trying to reach initial
3
pressure to have a conventional BU. Unfortunately, it did
not fully develop into IARF but a good match of the log-
log plot could be obtained using the numerical regression
results. Fig. 12 shows computed flowrate and pressure
trend.
Fig. 13 and 13.a shows optimized flowrate and input
flowrate for CG-613 after first and second optimization
run.
Fig. 14 and 14.a shows pressure history match after first
and second optimization run.
Optimized parameters from NM:
k:20.20
Skin Factor, S -1.5
These were used to match log log plot of the BU (Fig. 15),
it is quite possible that if the test had reached 10 hours of
shut in it would have developed an IARF around the blue
line (20.2 md).
A third example is shown on Fig. 16 where we can
observe the pressure and calculated flowrate trend for a
BU test on GBK_736. The twenty hours swabbing period
was followed by a 25 hrs shut in using a bottom hole shut
in valve.
Fig. 17 shows log-log plot of the BU Log-log plot match
was done with k=4410 md, s=15.1 and Pi=635.4 psia, Fig.
18 shows pressure and flowrate history during swabbings
as input in NM.
Extremely high flowrate was computed at the beginning of
the swabbing ( blue dots > 400 bpd at time <4000 sec in
dotted blue oval on Fig. 18), which is a consequence of
that at this early time the well was producing water (low
viscosity fluid which allows the layer to fill up the
wellbore faster), once reservoir fluid (120 cp oil) began to
flow from the reservoir wellbore fill up speed decreased
(pressure increase slope changed) and flowrate
computations stabilized at 50 bpd (blue dots in the yellow
blue oval on Fig. 18).
Computed flowrate history (optimized) versus input
flowrate history after the first optimization run is shown
on Fig. 19. Discrepancy at the beginning of the swabbing
has to do with water production; fluid viscosity in the
numerical model is 120 cp, which matches pretty well
with the rest of the flowrate history (yellow oval).
It is not possible to match the pressure on the period
affected by initial water production (Fig. 20) because fluid
viscosity in the reservoir model in the NM is 120 cp, the
period with a stabilized flowrate of 50 bpd shows a better
match. Unrealistic negative pressures computed by the
NM at the beginning of the swabbing (red lines in the
dotted blue oval) are caused by the fact that at that early
time a very high flowrate was calculated from the
measured pressure data, so a very low flowing pressure is
4 SPE 121485
required to produce such flowrate with a fluid viscosity of
120 cp (viscosity in the reservoir model).
The Computed parameters from optimization run 1 are as
follow:
K: 4291 md
Skin Factor, S:15.29
Excluding the initial period affected by water production,
both flowrates histories are quite close (Fig. 19), based on
this and in the fact that the initial input is not an accurate
measured flowrate, a second optimization run using the
numerical model flowrate history is performed, results are
shown on Fig. 21 and 21.a.
Computed parameters from optimization run 2 are as
follows:
K:4420 md
Skin Factor, S 14.00
Optimized parameters are fairly close to those obtained
from the conventional BU on Fig. 17 (k = 4410 md, s =
15.1).
Unique Solution
Not reaching IARF implies some concern about
uniqueness of the solution obtained due to the fact that it is
not possible to support the results with graphic evidence as
a straight line on a specialized plot. In this regard
convolution theory could be used to verify how close our
parameters are from the real answer, running a
convolution (superposition) algorithm (Appendix A) with
the swabbing data would give us another set of parameters
to be compared with those obtained through the numerical
regression.
Fig. 22 shows CnE_955 pressure match with the numerical
model, matching optimized parameters are k=44.20 and
s= 0.85.
This data was analyzed again but this time using only data
from the first wellbore fillup (t<4250 sec).
Fig. 23 shows good pressure match with k=40.39 md and
s=0.06 which is acceptably close to results obtained when
the whole test is analyzed on figure 22 (k=44.20 md and
s= 0.85).
When a home made convolution algorithm (model details
in Appendix B ) is used with the same flowrate and
pressure (only for the first wellbore fill up) a straight line
with a slope of 6.10 and a intercept of 34.42 is obtained
(Fig. 24).
Including formation parameters such as reservoir height
and porosity and fluid viscosity and formation volume
factor a k=53.69 md and s=0.97 are obtained.
The discrepancy is drastically reduced if the flowrate
computed from pressure trend (as described in the
Appendix A) is replaced with the regression flowrate
calculated with the optimized parameters. (red line on Fig.
22.a)
On Fig. 25, from the slope of 7.61 and the intercept of
34.91 a k=43.02 md and s=-0.132 are obtained.
Obviously, difference has to do with flowrate history
source, the NM flowrate (red line in Fig. 22.a) is a more
accurate flowrate history, computed with one set of
optimized parameters (k, s and fluid viscosity), than
flowrate from pressure trend (blue line in Fig. 22.a). Both
flowrates are not very different each other (Fig. 26), but
using one or another impacts very much the convolution
results.
The same methodology was repeated but now the second
wellbore fillup was added to the analyzed data. Fig. 27
shows regression pressure match with k=43.89 md and
s=0.39.
Convolution algorithm did not work properly (Fig. 28)
when using flowrate calculated directly from pressure
trend, this as a result of the less accurate flowrate
calculated from the pressure increase history.
When the regression flowrate from the NM (red line in
Fig. 29) was used, a perfect straight line with a slope of
6.65 and an intercept of 30.53 was obtained and k=45.29
md and s=-0.188 were calculated (Fig. 30).
NMR Permeability and Regression Permeability
Correlation
A comparison study between permeability indicator from
NMR log and permeability obtained with the described
methodology was performed. Timur and Coates constants
have been adjusted to fit the dynamic permeability trend,
with better results in some fields than in others. Fig. 31
and Fig. 32 show correlation obtained for Cañadon de la
Escondida and Cerro Grande fields.
Conclusions
1. Because of the changing nature of the dynamic
fluid level during a swabbing test, which implies
a changing flowrate due to the varying bottom
hole flowing pressure, the multirate consideration
is the appropriate approach to evaluate swabbing
test.
2. Results from numerical regression and
convolution showed good agreement, mostly
when the “cleaned” flowrate history from NM
was used as input into to convolution algorithm.
3. Fluid density and viscosity variations in the
wellbore cause changes in the pressure increase
slope and so on the flowrate history. This
strongly affects the reliability of results as the
SPE 121485
NM performs regression using the flowrate and
pressure history. The more homogeneous is the
fluid flowing from the layer, the more reliable is
the set of optimized parameters obtained.
4. Repeating the optimization run using optimized
flowrate history, improves a lot the pressure
history match. This new flowrate history is
computed from one fluid viscosity which
eliminates noise and couples both pressure and
flowrate with one fixed transmisibility value
(optimized parameters). It is author’s believe that
as long as this new flowrate history
(deconvolved) does not differ too much from the
input flowrate, optimized parameters could be
taken as reliable. This procedure although could
be questionable but it is a good approach based
on the quality of the data available.
5. This technique does not replace the well known
pressure transient analysis during a constant
flowrate period, however, comparisons between
different techniques has shown that, in the case of
lack of conventional time-pressure-rate, swabbing
interpretation using a numerical method or
convolution provides a suitable tool to
dynamically characterize reservoirs.
Nomenclature
P=Pressure, psi
q: flowrate, bpd
B: Volume factor, RB/STDB
µ: viscosity, cp
k: permeability, md
h: reservoir thickness, ft
t: time, hrs
Ф: porosity, fraction
c: compressibility, psi-1
r: radius, ft
s: skin factor, dim
BU: Build Up
DD: Draw Down
NM: Numerical Model
Subscripts
i: initial
I:Ith time step
wf: well flowing
o: oil
w: well
References
1. Earlougher, R.:”Advances in Well Test
Analysis”, SPE Monograph Volume 5.
2. Mustafa O., Fikri J.:”Integrated Nonlinear
Regression Analysis of Multiprobe Wireline
Formation Tester Packer and Probe Pressures
and Flow Rate Measurements,” paper SPE
5
56616 presented at the 1999 SPE Annual
Technical Conference and Exhibition held in
Houston, Texas , 3-6 October 1999
3. Sabet, M.A.:”Well Test Analysis”, Gulf
Publishing Company, 1991.
Appendix A
Flowrate Calculation
An Excel worksheet has been designed in order to
calculate flowrate history from pressure trend.
With the tubing capacity and the fluid gradient (density), it
is quite easy to estimate instantaneous flowrate as the
pressure increases (due to wellbore fillup).
Since this expression is based on instantaneous pressure
increase most of the time it yields noisy flowrate history.
The instantaneous pressure change can vary a lot from
delta t to delta t due to the noisy nature of the source data
(real pressure data of low productivity wells), which
delays and complicate the numerical regression.
To overcome this, it was recalled the transient nature of
this flow. Based on this, transient flow equation can be
used to calculate flowrate as a function of pressure drop
(RFT pressure minus P(i)) Instantaneous flowrate is
estimated by:
Level increase generated by Q(i) is:
Pressure increase is:
A pair of values of k and s are assumed to evaluate Eq. 2,
and a trail and error procedure is used until estimated P(I)
from Eq. 4 matches measured P(I).
Eq. 2 is just a tool to calculate flowrate. Dependency
between k and s implies that multiple combinations of
such parameters (permeability and skin) could work as
possible solutions in the described algorithm to estimate
the flowrate, but it is important to have in mind that in this
case this couple of k and s do not have any meaning on the
solution of the swabbing test, the transient flow equation
(Eq. 2) is just used to calculate the flowrate that matches
the pressure increase from time t to time t+1, this flowrate
is unique (is the flowrate that fills the pipe from t to t+1 to
6 SPE 121485

increase the pressure from p(i) to p(i+1)). k and s in that

equation are just one of the multiple combinations that
can be used to estimate the flowrate. At the end, the
important output from this appendix is the flow rate
history, which is used into the transient test analysis. Fig.
a.1 shows measured pressure and pressure obtained from
the trial and error procedure described, it is important to
understand that there is an unique flowrate (for a given
fluid gradient and pipe capacity) that fills up the pipe
(during t(i)-t(I-1)) to the correct level, that matches
measured pressure P(i). Generally, a transmissibility value
is suitable for an entire fill up or for many of them, change
of transmissibility is an indication of fluid viscosity
change during the test, which is possible because this test
is performed just after perforation so drilling and
completion fluids could still be coming out from the
formation.

Fig. 37 shows flow rate computed with Eq. 1 and 2

Appendix B

For a flow rate history of n q and applying superposition:

A plot of is a straight line with a slope of m’ so

kh can be obtained.
SPE 121485 7

Fig. 1-Presure data during swabbing for EG_638 layer 1233.5- Fig. 4-Input and Optimized flowrate history after first
36.5 optimization run.

Fig. 2-Pressure data and computed flowrate for EG_638 layer

Fig. 5-Pressure history match for EG_638 layer 1233.5-36.5
1233.5-36.5.
after first optimization run.

Fig. 3- Pressure data and computed flowrate for EG_638 layer

1233.5-36.5 as input into the regression model. Fig. 5.a- Pressure history match for EG_638 layer 1233.5-36.5
after second optimization run.
8 SPE 121485

Fig. 5.b- Flowrate history match for EG_638 layer 1233.5-36.5 Fig. 7- Pressure trend and computed flowrates for GBk_660
after second optimization run. (swabbing + BU)

Fig. 6-IPR curve for EG_638 layer 1233.5-36.5 with optimized

parametrs. Fig. 8-Log-Log plot of BU test (GBk_660)

Fig .6.a- IPR curve for EG_638 layer 1233.5-36.5 matching Fig. 9- Pressure data and calculated flowrate as input into the
swabbing test data. regression model for GBk_660
SPE 121485 9

Fig. 10- Optimized and input flowrate after the first Fig. 12- computed flowrate and pressure trend for CG_613
optimization run for GBk_660.

Fig. 10.a- Pressure match after the first optimization run for
Fig. 13- Optimized flowrate and input flowrate for CG-613 after
GBk_660.
first optimization run

Fig. 11- Pressure match after the second optimization run for Fig. 13.a- Optimized flowrate and input flowrate for after
GBk_660. second optimization run for CG_613
10 SPE 121485

Fig. 16- Pressure and calculated flowrate GBk_736

Fig. 14- pressure history match after first and optimization run
CG_613

Fig. 14.a- Pressure history match after second optimization Fig. 17- log-log plot of the BU period for GBk_736
run CG_613

Fig. 18: Computed flowrate and pressure trend for GBk_736

Fig. 15- Log-Log plot for CG_613 (last Fill Up)

SPE 121485 11

Fig. 19- Input and computed flowrate for GBk_736 after first Fig. 21.a- Pressure match after second optimization run for
optimization run. GBk_736

Fig. 22-Pressure match for CnE_955

Fig. 20-Pressure match for GBk_736 after first optimization
run

Fig. 22.a- Computed and Optimized flowrate for CnE_955

Fig. 21- Computed and optimizaed flowrate after second
optimization run for GBk_736
 12 SPE 121485

Fig. 26-Calculated and optimized flowrate for CnE_955 (first

Fig. 23-Pressure match for CnE_955 (first Fill Up) Fill Up)

Fig. 24-Convolution straight line for CnE_955 (First Fill Up) Fig. 27-Pressure match for CnE_955 first and second Fill Up

Fig. 25- Convolution straight line for CnE_955 with Optimized Fig. 28-Convolution plot for CnE_955 (calculated flow rate)
flowrate history (First Fill Up)
SPE 121485 13

Permeability from Swabbing vs K indicator from NMR (CG)

40

35

Permeability from Swabbing (md)

30

25

20

15

10

0
0 5 10 15 20 25 30 35
Permeability Indicator from NMR

Fig 29: Optimized and calculated flowrate history (CnE_955) Fig. 32-Permeability from swabbing test vs permeability
indicator from NMR (CG)

Fig. 30- Convolution plot for CnE_955 ( Optimized flow rate)

Permeability from Swabbing vs K indicator from NMR (CnE)

70

60
Permeability from Swabbing (md)

50

40

30

20

10

0
0 5 10 15 20 25 30 35 40
Permeability Indicator from NMR

Fig. 31-Permeability from swabbing test vs permeability

indicator from NMR (CnE)