You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/314775559

Numerical Well Testing: A Method To Use Transient Testing Results in Reservoir


Simulation

Conference Paper · October 2005


DOI: 10.2523/95905-MS

CITATIONS READS
7 731

4 authors, including:

Medhat Kamal Yan Pan

44 PUBLICATIONS   467 CITATIONS   
The University of Nottingham Malaysia Campus
96 PUBLICATIONS   600 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Questionnaire study about knowledge on colorectal cancer (CRC) View project

Pressure Transient Analysis of Polymer Flooding With Coexistence of Non-Newtonian and Newtonian Fluids View project

All content following this page was uploaded by Medhat Kamal on 29 August 2019.

The user has requested enhancement of the downloaded file.


SPE 95905

Numerical Well Testing - A Method to Use Transient Testing Results in Reservoir


Simulation
Medhat M. Kamal, Yan Pan, and Jorge L. Landa, Chevron Corporation; and Olubusola O. Thomas, Stanford University

Copyright 2005, Society of Petroleum Engineers

This paper was prepared for presentation at the 2005 SPE Annual Technical Conference and NWT was applied to field cases to demonstrate the
Exhibition held in Dallas, Texas, U.S.A., 9 – 12 October 2005.
applicability of the method and proved that incorporating
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as results from the proposed NWT procedure improved
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to reservoir description. Descriptions enhanced by NWT
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 matched not only well test data but also field production
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
data, and predicted different reservoir performances
for commercial purposes without the written consent of the Society of Petroleum Engineers is from those obtained via matching only production data.
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous The paper concludes by describing the software
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.
developments needed to make NWT a routine analysis
method.
Abstract
Transient testing is facing a challenge to have the Background
information it provides incorporated in numerical Transient well testing has been a core competency of
simulators used to predict reservoir performance. Of the oil industry for a long time because it provides
greater importance, the lack of using the large number of petroleum engineers with valuable information about
dynamic in-situ reservoir parameters that can be reservoirs. From the early days when the technology
determined from transient tests deprives operators from started in groundwater hydrology and was quickly
valuable data they can use to improve their reservoir adopted in petroleum engineering, it has been used to
management efforts. The main reason for the problem is determine formation permeability, wellbore conditions,
that transient testing technology, developed largely via and reservoir pressure. Gradually we learned how to
analytical solutions, provides average reservoir determine, among other things, effective fracture lengths
parameters not suited to the numerically discretized and conductivities, distances to boundaries, and the
environment of current reservoir simulators. Previous most applicable model for reservoir behavior.
efforts to include transient testing data in numerical Information obtained from transient well testing was
reservoir simulation studies focused on history matching being used routinely by drilling, completion, production,
the pressure behavior during the test on a Cartesian- and reservoir engineers. Developments in transient well
type plot just like matching stabilized production rates testing were achieved mostly using analytical solutions
and pressures. Such approach neglects the wealth of which necessitated assuming uniform reservoir
information contained in the transient behavior of the properties throughout the area investigated by the
tests. transient test. Determining a single value of effective oil
A method called Numerical Well Testing (NWT) is permeability, for example, was sufficient when the
proposed in this paper to preserve the information reservoir performance was being predicted using
obtainable from traditional well test analysis and deliver material balance and decline curve analysis. With the
it in a form suitable for direct use in numerical reservoir introduction of numerical reservoir simulation and its
simulation. By using this method petroleum engineers ability to better predict the reservoir performance,
will not shy away from using transient testing results just providing an answer from transient well testing assuming
because they do not know how to incorporate averaged uniform properties is no longer sufficient. It is important
values in their discretized reservoir model. to note that some, may be most, of the information
NWT is a systematic method that consists of five steps obtained from transient well testing’s analytical solutions
starting with traditional well test analysis, locally refining are still sufficient. For example values of wellbore skin,
grids based on the analysis results, modifying full field fracture length, and reservoir pressure can be used in
numerical models, upscaling, validating the updated numerical reservoir simulation quite effectively and are
model with production data, and predicting future not expected to be affected significantly by using NWT.
reservoir performance. Details of how to accomplish But for parameters such as formation permeability
each step are presented in the paper. distribution and distances to boundaries, NWT is needed
2 SPE 95905

10-12
to provide the reservoir engineer with the necessary not captured. Landa, et al developed a procedure to
information to develop a more accurate reservoir integrate well test data (both pressures and pressure
description and performance prediction. Absent this type derivatives with respect to time), reservoir performance
of information, reservoir engineers may find it simply history and 4-D seismic information into reservoir
easier to use only the transient testing results that can description. He also assessed the uncertainty associated
13
be readily incorporated in the simulation model (like skin with property estimations using the Gradient Method .
14
and fracture length) and ignore those that cannot (like Raghavan, et al demonstrated an approach to
permeability) in which case valuable information about incorporate geologic and geophysical data in the
the reservoir description would be lost and the best analysis of pressure transient tests in a fluvial reservoir.
possible performance prediction would not be obtained.
One of the consequences for the lack of appropriate Proposed Solution
methods to incorporate the results of transient testing in In this paper, we present a step-by-step approach of
numerical simulation models, is the tendency among incorporating well test data into full-field simulation
reservoir engineers to “history match” the measurements model using transient analysis standard.
from transient tests in the same manner they “history The proposed Numerical Well Testing (NWT) method
match” production data. This method usually consists of consists of five steps:
supplying the rate history as input information and 1. Analysis of test data using state-of-the-art
changing the reservoir description until the measured analytical analysis methods
and calculated pressures are visually matched on a 2. Extracting the area of influence of the well test
Cartesian plot. As most of the value and information from the available numerical simulation model
from a transient tests are contained in early time (this is and modifying the grid to allow for transient-level
why we often use logarithmic scale for the time axis), analysis
such information would be lost if data were matched on 3. Modifying the properties in the extracted area
Cartesian plots. The wealth of information developed until measured pressure and derivative are
over the years to analyze transient tests allows us to matched with model response
identify various flow regimes, know which reservoir / well 4. Upscaling the fine grid to the level that existed in
parameters affect each flow regime and use specific the original numerical simulation model and
regions from the well test data to calculate certain replacing the modified area in the simulator.
parameters. Using these techniques reduces the 5. Verify the production history of all wells using
uncertainties in the values of the reservoir parameters. the new full-field model. Repeat steps 3-4 if
Again, using a simple “history match” on a Cartesian plot necessary to obtain consistent results.
deprives us from the valuable information we know how The full-field simulation model resulted from this NWT
to extract from transient tests. procedure will be ready for reservoir performance
Therefore, it behooves us to develop an analysis prediction.
method that allows for obtaining and conserving all the
information we know how to deduce from transient well Step 1: Analytical Analysis
testing, removing the restrictive assumptions of The data from the transient test is analyzed using the
uniformity needed for analytical solutions and providing current available analytical analysis methods. The
the reservoir engineer with results in a format ready for constant fluid PVT properties used as input for the
input in numerical simulation models. Such an analysis analysis should be the same as those used in the
method is termed Numerical Well Testing (NWT) and is simulation reservoir model at the time of the test. The
presented in this paper. results from this analysis will be used as initial values in
the third step. Some of the results probably will not
Previous Work change in a significant way during the final analysis
Since well tests provide important dynamic information (e.g., skin, fracture length) while the values of other
about reservoirs, numerous efforts have been made to properties will change (e.g., permeabilities, distances to
use transient pressure data to improve reservoir boundaries). As we are assuming that a numerical
1
description. Kamal proposed a systematic method to simulation model for the field exists (and this is why we
interpret and apply the results from multiple-well tests in are performing NWT), it is safe to envision that some
conjunction with single-well tests to obtain quantitative knowledge exist about the reservoir behavior (e.g.,
heterogeneous reservoir descriptions. The issues related double porosity, layered, existence of boundaries, etc).
to numerical simulation of well tests have been This information would be helpful in reducing the number
2-3 4-9
discussed . Many authors have presented different of potential models that may be used to analyze the test
approaches to generate reservoir models conditioned to data in this step. The results from this step will also
well production and transient test data while honoring provide the area of influence for the test.
geostatistic information. However, they only used
pressure data in Cartesian plots for matching the well Step 2: Extracting and Preparing the Grid
tests, the same common practice of production history The area of influence from the previous step is extracted
matching in reservoir simulation. The reservoir from the numerical simulation model. The model
characteristics contained in pressure derivatives were probably is for a multi-layered heterogeneous reservoir
SPE 95905 3

with spatial distribution of permeability, porosity, and boundaries, approximate level of effective permeability
fluids’ saturations. The reservoir pressure and fluids as well as other pertinent properties. The sensitivity
saturation distributions in the influence area of the test coefficients of the selected parameters are calculated
13
are taken from the original full-field simulation model at using the Gradient Method . Then applying automatic
10
the time of the beginning of the well test. The boundary history match techniques in a regression program ,
conditions for the extracted section should be set these properties are modified until an acceptable match
carefully to take into account the effects of surrounding is obtained between the numerically simulated results
wells ans select an area where the boundaries can be from the model and measured pressure and pressure
readily modeled. The grid size will most probably be too derivative on a log-log scale. The modified section
coarse for transient test analysis. The extracted model models with fine grids around the well and geological
should be fine gridded, especially around the well and features for each well test are then put back into the full-
any geological feature, to the level needed to be able to field simulation model. If the section model has complex
match the measured pressure and the pressure geologic features, which involve difficult simulation input
derivative. adjustment and intensive calculation, a multistep
The optimal grid size for local grid refinement (LGR) procedure may be necessary to progress from fine grid
is selected based on obtained formation transmissivity, model to intermediatete grid model before being
skin factor and wellbore storage. To illustrate this point, integrated back to the full-field model.
a drawdown test in a well without wellbore storage in a
single-layer closed homogeneous reservoir with uniform Step 4: Upscaling to the Original Model
grid size was numerically simulated to investigate the To use the information obtained from the transient tests
grid size necessary to generate the exact analytical and is now reflected in the description of the fine gridded
pressure transient response during radial flow and full-field model, the reservoir properties must be
pseudosteady state for different transmissivity values. upscaled to the dimensions of the original model using
From this sensitivity study, the correlation between the any of the upscaling methods that have been discussed
grid-size required for transient test simulation and in the literature. It is recommended that all major
formation transmissivity was obtained. Figure 1 shows features resulting from the transient test analysis
the relationship between the required grid size and performed in the first step be retained in the models that
transmissivity when accurate radial flow response can we end up with after the third and fourth steps.
be reproduced at 0.01 hour, 0.1 hour and 1 hour after
each flow rate change. The simulation time steps also Step 5: Validating the New Full-Field Model
have to be selected according to the starting time of a The modified full-field simulation model needs to be
characteristic reservoir response, such as radial flow, verified using the available production history of all the
shown in transient analysis, and the grid size after wells in the reservoir, not only the tested wells. When it
refinement, in order to regenerate the transient test with is necessary repeat steps 3 to 4 until consistent results
numerical stability. Most likely, the time steps used are are obtained. If new production history matching efforts
much smaller compared to time steps usually seen when are made, it is also recommended to apply NWT
predictions of reservoir performance are being made. procedure to verify whether the reservoir model honor
40 the well test data.
35
dt = 1.0 hr
dt = 0.1 hr
Field Examples
30
dt = 0.01 hr The effectiveness of NWT was proved using well test
data from two fields, Field C and Hibernia West Field.
gridsize (ft)

25

20 The numerical analyses show that incorporation of well


test data improves the reservoir model by matching not
15
only the transient data but also the field production data.
10

5
Field C
Field C is a large oil field. It is divided into sections. The
0 central area is relatively homogeneous. The starting
1000 10000 100000
simulation model, which is subsequently referred to as
transmissivity (md-ft/cp) the coarse-grid full-field model, contains 600,000 cells
Figure 1: Grid size as a function of transmissivity.
and has been history matched to some static pressure
data. A number of wells have been tested in this field.
Step 3: Modifying the Model Parameters (History Match) The transient tests with single-phase oil production at
From the first step, select the main properties that are three vertical wells, C-1, C-2 and C-3 in the central area
important in affecting the reservoir transient behavior. (Figure 2) were chosen to be analyzed and incorporated
Values of the parameters obtained from the analytical into the full-field reservoir model.
analysis step should be used as the initial values in the
regression process. These values include information
about wellbore skin, fractures (if present), distances to
4 SPE 95905

1000

C-1
100
C-3

1E-3 0.01 0.1 1

Log-Log plot: dp and dp' [psi] vs dt [hr]


Figure 4: Analysis of the well test data from C-1.
C-2

Figure 2: Field C full-field simulation model with 3 tested wells. 2- Determination of test influence area and fine grid size
Since the total test duration is around 220 hours, it leads
Since the three wells are located in the relatively to an investigation radius of 790 ft. This influence area
homogeneous central area, and the transient tests at the lies within 3x3 coarse-grid blocks. The properties of this
three wells did not provide any layering geological region, such as the permeability and porosity
information, the relative layering of the full-field model distribution, the pressure and saturation fields at the time
were kept intact, and the layer-averaged horizontal of the test, were then extracted from the current full-field
permeabilities in the test influence area of each well simulation model.
were modified by applying permeability multipliers The estimated transmissivity from step (1) is 9660
throughout all the layers in the model. An automatic md-ft/cp. Based on the relationship between reservoir
10
history-matching program was used to search for the transmissivity and the grid size necessary for transient
optimal permeability multipliers while their sensitivity test simulation (Figure 1), and considering the early time
coefficients with respect to layer-averaged permeabilities wellbore storage and skin effects, the coarse-grid well
being calculated. The well skin factor and initial reservoir block was refined to 3.3 ft x 3.3 ft. The surrounding grid
pressure near each well were also modified when blocks were refined as well for the stability of numerical
necessary. simulation. The LGR of the extracted section model is
shown in Figure 5.
Well C-1
1- Analytical pressure transient analysis
The total duration of flow rate changes at Well C-1 was
220 hours with the last 16 hours consisting of 8-hr
drawdown and 8-hr buildup (Figure 3). From the well test
analysis (Figure 4), it is concluded that the pressure
response is from a homogenous system, and the
estimation of the average permeability is 14 md for 1384
ft total net pay. Possible flow barrier might exist 46 ft
away from the well. The estimated skin factor is 14.4.
Figure 5: LGR for C-1 numerical transients test simulation

8500
3- Modification of section model to match well test data
The well test at C-1 is a short test with very small
influence area compared to the reservoir size. There
6500
was no confirmative pressure behavior to prove the
presence of a flow barrier near the well. To match the
well test data, three ways were used to modify the layer-
averaged permeability distribution in the area.
(a) Modifying properties in only the coarse grid block
containing the well, effectively 410 ft away from the well.
One permeability multiplier was applied to the well block.
History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr]) (b) Modifying properties within the locally refined area
around the well, as well as modifying properties within
Figure 3: Overview of C-1 well test.
the entire area of influence. Three permeability
multipliers were defined: one for the inner LGR, one for
the coarse well block and one for the entire area of
influence.
SPE 95905 5

(c) Modifying the properties in the entire region of


influence. One permeability multiplier was applied to the
entire area of influence.
The results after history matching to the well test data
are shown in the Cartesian plot of pressure versus time
(Figure 6) and the log-log pressure derivative plot
(Figure 7), which is used as transient analysis standard.
Figure 7 shows that the pressure derivative curve
generated from the simulated test data using the original
full-field model is lower than the measured data. This
indicates that the effective average kh in the model is
larger than the reservoir value. After the automatic
history-matching process, the simulated pressure data
match the gauge data. The results show that modifying
properties within the region with LGR is not the best way
to match this test. Essentially, the area contained in the
LGR is too small to account for the pressure response
observed at the well, and a larger area must be
Figure 8: Layer-averaged permeability of all models
modified. Permeability multipliers of 0.406 and 0.437 (a) Original model (b) reduction of well block permeability
were applied to the coarse well block only (case a) and (c) reduction of permeability within LGR and area of influence
the area of influence alone (case b), respectively. Case (d) reduction of permeability within area of influence
(a) appears to be the better of the two choices, as
suggested by Figure 7. Figure 8 shows the layer- Well C-2
averaged horizontal permeability profiles of the history- 1- Analytical pressure transient analysis
matched models. The total duration of flow rate changes at Well C-2 was
73 days with the last 27.5 days for buildup test (Figure
9000 9). The well test analysis (Figure 10) suggests that the
pressure response is from a homogenous system, and
gauge data (ref)
original model
the estimation of the average permeability is 1.1 md for
7000
modified within wellblock 407 ft total net pay. Possible flow barrier might exist 670
modified within influence area ft away from the well. The estimated skin factor is 4.9.
modified within LGR
5000 10000

8000

6000

History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])


Figure 6: History match of buildup pressure in Cartesian plot

gauge data (ref)


original model
modified within wellblock
modified within influence area History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])
modified within LGR
1000
Figure 9: Overview of C-2 well test.

100

1000

10
1E-4 1E-3 0.01 0.1 1

Log-Log plot: dp and dp' [psi] vs dt [hr] 100

Figure 7: History match of buildup pressure in log-log plot 0.01 0.1 1 10 100

Log-Log plot: dp and dp' [psi] vs dt [hr]


Figure 10: Analysis of the well test data from C-2.
6 SPE 95905

2- Determination of test influence area and fine grid size 10000


Since the total test duration is around 1750 hours, it
leads to an investigation radius of 2240 ft. This influence
area lies within 7x7 coarse-grid blocks. The properties of 8000 gague data (ref)
original model
this region were then extracted from the current full-field uniform modification
simulation model. modified based on proximity to well
modified based on sensitivity
The estimated transmissivity from step (1) is 2004 6000

md-ft/cp. Based on the relationship between reservoir


transmissivity and the grid size necessary for transient
test simulation (Figure 1), and considering the early time
wellbore storage and skin effects ended around 1 hour
after shut-in, the coarse-grid well block was refined to 10
History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])
ft x 10 ft. The surrounding grid blocks were refined as
well for numerical stability. The LGR of the section Figure 12: History match of buildup pressure in Cartesian plot.
model is shown in Figure 11.
gague data (ref)
original model
uniform modification
10000 modified based on proximity to well
modified based on sensitivity

1000

100

Figure 11: LGR for C-2 numerical transients test simulation.


0.1 1 10 100 1000
3- Modification of section model to match well test data
The well test at C-2 has a medium duration. There was Log-Log plot: dp and dp' [psi] vs dt [hr]
no confirmative pressure behavior to prove the presence Figure 13: History match of buildup pressure in log-log plot.
of a flow barrier near the well. To match the well test
data, three ways were used to modify the permeability
distribution in this area.
(a) Uniform modification of properties within the area
of influence. One permeability multiplier was applied to
the entire region.
(b) Non-uniform modification of properties within the
area of influence of the test, based on proximity of the
coarse-grid block to the well. Three permeability
multipliers were defined as follows: the first multiplier for
the well block, the second multiplier for the four coarse-
grid blocks adjacent to the well, and the third multiplier
for all the other blocks in the area of influence.
(c) Non-uniform modification of properties within the
area of influence, based on the sensitivity map. Here,
four permeability multipliers were defined, based on the
magnitude of the sensitivity coefficient for each grid
block during the buildup period of the test.
The results after history matching to the well test data
are shown in the Cartesian plot of pressure versus time Figure 14: Layer-averaged permeability of all models.
(a) Original model (b) reduction of permeability in entire area of
(Figure 12) and the log-log pressure derivative plot influence (c) reduction of permeability in area of influence based
(Figure 13). Figure 13 shows that the effective average on proximity to well (d) reduction of permeability in area of
kh in the original full-field model is larger than the influence based on sensitivity coefficients
reservoir value. After the automatic history-matching
process, the simulated pressure data of all three cases Well C-3
match the gauge data. The layer-averaged permeability 1- Analytical pressure transient analysis
fields for the history-matched models are shown in The transient test at Well C-3 includes 12-day buildup
Figure 14. In general, the permeability multipliers were (Figure 15). From the well test analysis (Figure 16), it
in the range of 0.4 to 0.5, with a skin factor of 4.7. may be concluded that the pressure response is from a
SPE 95905 7

homogenous system, and the estimation of the average


permeability is 0.6 md for 197 ft total net pay. Apparent 3- Modification of section model to match well test data
flow barriers, possibly two intersecting faults, are present The well test at C-3 has a long duration. There was
at 130 ft and 320 ft away from the well. The estimated apparent pressure behavior due to flow barriers near the
skin factor is -1.6. well, which was confirmed by geological information. In
the original full-field simulation model, the flow barriers
close to the well are not captured in the coarse grids.
11000
Therefore, two small intersecting no-flow faults were
added to the model. Many other flow barrier
10000
configurations are also possible.
The results after matching the well test data are
shown in the Cartesian plot of pressure versus time
9000 (Figure 18) and the log-log pressure derivative plot
(Figure 19). Figure 19 shows that the original full-field
model could not reproduce the pressure behavior (radial
flow followed by boundary effects) observed in
measured data. After the automatic history-matching
process, the permeability in the coarse-grid well block
History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr]) was increased by a factor of 5 and the permeabilities
Figure 15: Overview of C-3 well test. away from the well were decreased. The simulated
pressure data from the modified model match the
measured data much better. The layer-averaged
permeability fields for the original model and the history-
1000
matched models are shown in Figure 20. As indicated
earlier, many other flow barrier configurations (different
location and length) are possible. Because of the
relatively short duration of the well test when compared
100
with the total history of production from the well, the
extension of these flow barriers in the reservoir could be
0.01 0.1 1 10 100 much longer than the ones obtained by the history match
algorithm, and thus it is important to apply geological
Log-Log plot: dp and dp' [psi] vs dt [hr] insight at the time of translating this information to the
Figure 16: Analysis of the well test data from C-3. full-field model. What is important is that a flow barrier
close to the well was identified via well test analysis.
2- Determination of test influence area and fine grid size 11000
The total test duration is around 316 hours, which leads
to an investigation radius of 2700 ft. Since some of the
layers pinch out 5300 ft away from the well, a larger 9000
gauge data (ref)
influence area which lies within 13x13 coarse-grid blocks original model
was extracted from the current full-field simulation modified model

model. 7000

The estimated transmissivity from step (1) is 536 md-


ft/cp. Based on the relationship between reservoir
transmissivity and the grid size necessary for transient
test simulation (Figure 1), and considering that the early
time wellbore storage and skin effects ended about 1
hour after shut-in, and the presence of two intersecting History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])
faults, the coarse-grid well block was refined to 6 ft x 6 ft. Figure 18: History match of buildup pressure in Cartesian plot.
The LGR of the section model is shown in Figure 17.

Figure 17: LGR for C-3 numerical transients test simulation


8 SPE 95905

gauge data (ref) C-1 and C-2, the modified simulation model predicts
original model
modified model lower pressure during the maximum oil rate constrained
production period and lower oil rate during the minimum
well pressure constrained production period. For Well C-
1000
3, both models predicted similar performance, which
may not be a surprise considering that the flow barriers
introduced in the model are relatively small. However,
the fact that the barriers could be more extensive and
100
0.01 0.1 1 10 100 still honor the well test data raises a red flag on the
future performance of this well.
Log-Log plot: dp and dp' [psi] vs dt [hr]
Figure 19: History match of buildup pressure in log-log plot.

(a) (b)

(a) (b)
Figure 20: Layer-averaged x-permeability for
(a) original model (b) history-matched model

Depending on the characteristics of each well test,


different techniques had to be applied to modify the
permeability distribution in the test influence area to (c)
Figure 21: Performance predictions for C-1.
match the pressure transient data. There is no universal (a) Oil production rate (b) Bottom-hole pressure
method for the entire full-field model. (c) Average reservoir pressure

4- Assembling full-field simulation model


The modified section models near the three wells were
assembled back and the original coarse-grid full-field
simulation model was updated. This was achieved by
applying the corresponding multipliers within the test
influence regions and in the case of well C-3 including
the shortest possible flow barriers identified by well test
analysis and calibrated by history matching.
(a) (b)
In this field example, the tested areas are relatively
homogeneous, and therefore, it was possible to achieve
matching transient data by modifying the properties in
each coarse grid uniformly. The grid refinement was
necessary to numerically reproduce transient response.
However, property adjustment in the fine-grid level was
not required. Therefore, no upscaling was needed for
this field example.
(c)
5- Validating the Modified Full-Field Model Figure 22: Performance predictions for C-2.
(a) Oil production rate (b) Bottom-hole pressure
The updated full-field simulation model containing well- (c) Average reservoir pressure
test information was used to compare with available field
production data and to predict the performance of the
three wells. The results were also compared with those
generated from the original coarse-grid simulation
model. Ideally, the production history at all the wells in
the reservoir should be checked for consistency.
Figures 21-23 show the production history and
performance predictions for the three wells. The average
reservoir pressures near each well calculated from both
the original and modified models match the few static
pressure measurements in a similar fashion. For wells
SPE 95905 9

the full-field model. The practical difference between the


models is that the simplest model took seconds to
compute the test response while the intermediatete one
took minutes and the most complex model took hours.
Complexity was thus incorporated in the reservoir model
step by step, and this rendered efficiency and
practicality.
The initial full-field simulation model was constructed
(a) (b) using mainly seismic, well logs, cores, and PVT data.
The model had also been partially history matched to
some extent using traditional production history data.
The grids and simulation time steps were adjusted to
simulate transient test response following the same
procedure described before. Then, the quality of the
match between the simulated test data at Well B16_33
using this model and the observed data was
(c) investigated. As the Cartesian plot Figure 25 shows, the
Figure 23: Performance predictions for C-3. calculated pressure did somehow match the last buildup
(a) Oil production rate (b) Bottom-hole pressure
(c) Average reservoir pressure
pressure in each shut-in period, but the pressure buildup
profiles were quite different from the measured data.
West Hibernia Using the pressure transient analysis standard, the log-
The Hibernia field is located in the Jeanne d'Arc Basin, log diagnosis plot, it was observed that the simulation
315km east of St John's, Newfoundland, in water 80m model did not capture the transient behavior due to flow
deep. The western area of Hibernia has complex boundary conditions (Figure 26). The effective average
faulting. The West Block is divided into two major fault kh value near the well in the model is very close to what
blocks each with four sub-blocks, namely J, K, L, M and the measured data observed, which is the flat part
F, H, I, N (Figure 24). Well B16_33 is located in Block K. between 0.2 hr and 0.4 hr after shut-in in the pressure
The multiple buildup test data collected during a four- derivative curve of buildup #3 shown in Figure 26. Due
month period at this well (Figure 25) were used to assist to possible gauge problem during buildup #6, this part of
in determining the fault configuration and flow radial flow was missed. Since buildup #6 is the longest
communication between the fault blocks in order to (more than 1000 hrs), it was used to calibrate faults far
improve the reservoir performance prediction. away from the well. In the simulation model, the well skin
factor is constant throughout all the tests. However, the
2-D
15000

gauge data indicate higher skin in the early buildup #3


and lower value in buildup #6, and the values are all
10000 H higher than that in the original model. Starting from 1 hr
I to 20 hrs after shut-in, the transient behavior shows the
5000 boundary conditions near the well in JKLM fault block.
J F ? After 20 hrs, the pressure response is from the long
0
? faults separating fault blocks J, I and H. Based on the
K Tested w ell

L analysis of the data in the log-log plot, the permeability


M N distribution in the model was unchanged, and the well
-5000
skin factor was increased to the value obtained from
transient analysis of buildup #6. Time-dependent skin
-10000 could also be used. However, the focus of numerical well
-15000 -10000 -5000 0 5000 10000 15000 20000 25000
test analysis in this example was to modify the fault
Length [ft] vs Length [ft] configuration.
Figure 24: West Hibernia areal map

The integration of well test information into the full-


field reservoir simulation model for this field case was
accomplished by using three numerical reservoir models
of increasing complexity. Starting with the simplest one,
the information obtained by transient analysis of
generated well test from one model was incorporated
into the next level model until reaching the full-field
model. This approach made it possible to reduce the
computational efforts of incorporating the well test data,
especially those related to geological features, directly in
10 SPE 95905

fault configuration as in the original full-field model, the


B16_33_Gauge (ref)
5100 B16_33_Simulation 0 fault locations and lengths were modified to match the
pressure derivative behavior observed in the log-log plot.
BU #3
4900 BU #6

4700

d w ell
Tes te

History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr]) 0


2000
Figure 27: First-step model with Voronoi grids near Well B16-33
Figure 25: Well B16-33 test gauge data and simulated data using 1000
0
original model in Cartesian plot. -1000
0 [f t]
The 0
obtained fault ngth from
configuration
s Le numerical
t] v
transient analysis using the[fsimple but computationally
B16_33_Gauge (ref)
B16_33_Simulation 0 BU #3 fast section model is shown as the bold black lines in
1000
Figure 27. The analysis identified the existence of an
internal no-flow barrier near the well that was absent in
the original reservoir model built using mostly static data.
100
The analysis on this simple model also confirmed the
finding from conventional full-field history match that
there was flow communication between blocks J, I and
10
H. The simulated buildup #3 and #6 pressures using the
simple recalibrated model are shown as blue lines in the
log-log plot in Figure 28. Figure 29 shows the results in
1
1E-3 0.01 0.1 1 10 100 1000 Cartesian plot. Comparing Figure 28 with Figure 26, the
new model captured the transient behavior due to flow
Log-Log plot: dp and dp' [psi] vs dt [hr]
barriers much better than the original model.
B16_33_Gauge (ref)
B16_33_Simulation 0
BU #6 build-up #3 (ref)
build-up #6

100

100

10

10

1
1E-3 0.01 0.1 1 10 100 1000

1
Log-Log plot: dp and dp' [psi] vs dt [hr] 1E-3 0.01 0.1 1 10 100 1000

Figure 26: Well B16-33 pressure buildup gauge data and Log-Log plot: dp and dp' normalized [psi] vs dt
simulated data using original model in log-log plot.
Figure 28: Numerical analysis of buildup #3 & #6 in log-log plot

Since changing fault properties (location, length and


connectivity) in a full-field simulation model requires
substantial computational efforts, a relatively small and
simple section of the model (Figure 27) was selected to 4950

determine the approximate fault configuration within


significantly reduced CPU time. This model is a single-
4750
well, oil phase with constant PVT properties at the
reservoir conditions at the time of the beginning of the
test in the full-field simulation model. It is 1-layer,
homogeneous with the same kh value obtained from
analytical transient analysis. For practicality and
efficiency, higher level of details, such as heterogeneity History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])
and 3D modeling was deferred to the intermediatete
Figure 29: Numerical analysis of tests at B16-33 in Cartesian plot
model in the next step. At this stage, Voronoi grids were
used for transient test simulation. Starting from the same
SPE 95905 11

The next logical step would have been to incorporate


the results obtained from the simple model directly in the B16_33_Gauge (ref)
B16_33_Simulation final
full-field model and then to proceed to fine tune the 5100

model using automatic history matching techniques, but


this process would be not only very time consuming
BU #3 BU #6
because of the computational efforts involved, but also 4900

not practical since the full-field model was still in the


process of being history matched using data from the 4700
other wells in the same reservoir. Thus, a model of
intermediatete complexity was constructed by extracting
a portion of the full-field model. This model was still a
single-well model but incorporated the permeability and
porosity heterogeneity in three dimensions, rock and
History plot (Pressure [psia], Liquid Rate [STB/D] vs Time [hr])
multiphase fluids properties, and geometrical complexity
of the full-field model. One of the key elements in this Figure 31: Well B16-33 test gauge data and simulated data using
modified full-field model in Cartesian plot.
step was the identification of interference with other
wells in the reservoir. Then the fault configurations
obtained using the simple model were incorporated in
this intermediatete model and were recalibrated using
automatic history matching. Figure 30 shows part of the
intermediatete reservoir model near Well B16_33 in
three-dimensional view. B16_33_Gauge (ref)
B16_33_Simulation Final BU #3
Once a good match was obtained, the results were 1000

returned to the full-field reservoir simulation model. A


simulation run was then made with the full-field reservoir 100
model to verify that the model is able to replicate the well
test data within acceptable tolerance. The simulation
results from the final full-field model are shown in 10

Cartesian plot (Figure 31) and log-log plot (Figure 32).


By improving the quality of the match over the original 1
model, an improved reservoir description was obtained 1E-3 0.01 0.1 1 10 100 1000

by incorporating pressure transient information.


Log-Log plot: dp and dp' [psi] vs dt [hr]

B16_33_Gauge (ref)
B16_33_Simulation Final
BU # 6

B16-33 100

10

Fault
1
1E-3 0.01 0.1 1 10 100 1000
Flow Barrier
Log-Log plot: dp and dp' [psi] vs dt [hr]
Figure 32: Well B16-33 pressure buildup gauge data and
simulated data using modified full-field model in log-log plot.

Figure 30: Intermediate reservoir section model near Well B16-33 Discussion
NWT was used to help describe the two fields presented
In this process three reservoir models of different in the examples in the previous sections. In Field C
complexity were generated. These models were example, the method was demonstrated step by step.
consistent with the transient pressure data. Because of Use of analytical analysis results and different
the linkage between the models, any future intended techniques to modify permeability distribution were
update of the full-field model arising from history match presented. The results showed that reservoir models
can be easily verified by probing the simple and modified via matching well test data might provide
intermediatete reservoir models to preserve the pressure different performance predication from that provided by
transient data. Ideally, the production history at all the the models matching only production data. For West
wells in the reservoir should also be checked once any Hibernia, an approach using increasilngly complex
change in the full-field simulation model is made to reservoir models to incorporate the boundary conditions
maintain the consistency. due to multiple faults observed in transient data into full-
12 SPE 95905

field simulation model was discussed and proved that References


numerical transient analysis improved the efficiency and 1. Kamal, M.M.: “Use of Pressure Transients to
accuracy of simulation study. Describe Reservoir Heterogeneity,” JPT, (Aug.
As in any process that uses mathematical inversion 1979) 1061, Trans., AIME, Vol. 257, SPE Reprint
algorithms for model parameter estimation, caution Series No. 14.
should be taken by incorporating information from other 2. Blanc, G., Noetinger, B. and Placentino, L.:
reservoir description tools before assigning a geological "Contribution of Pressure Moments to the
interpretation in the reservoir model. Interpretation of Numerical Simulations of Well
th
Since the process described in this work involves the Tests,” presented at the 5 European Conf. on the
transfer of information among different models of the Mathematics of Oil Recovery, Leoben, Sept. 3-6,
same physical entity, it is important that the model be 1996.
constructed and modified in a way that ensures a 3. Abdelmawla, A. and Heinemann, Z.: “Numerical
smooth transfer, and this should be planned and Well Test Modeling in A Full-Field Simulator Offers
managed at the very early stage of any reservoir New Opportunities for Reservoir Characterization,”
th
characterization effort. presented at the 6 European Conf. on the
Moving between the five steps of NWT requires using Mathematics of Oil Recovery, Peebles, Scotland,
the results from given software into another. To facilitate Sept. 8-11, 1998.
this process software that can perform these functions 4. Chu, L., Reynolds, A.C., and Oliver, D.S.:
smoothly should be developed. Since there is no “Reservoir Description from Static and Well-Test
standard format for data input or output in reservoir Data Using Efficient Gradient Methods,” paper SPE
management software (e.g., numerical simulators, well 29999 presented at the 1995 SPE International
test analysis) developing such software may be case Meeting on Petroleum Engineering, Beijing, PR
specific (i.e., a special software may be needed to link China, Nov. 1995.
well test analysis package X to numerical simulator Y). 5. Oliver, D.S., He, N. and, Reynolds, A.C.:
Conditioning Permeability Fields to Pressure Data,”
Conclusions paper presented at the 5th European Conference on
1. A method (NWT) that enables the use of the the Mathematics of Oil Recovery, Leoben, Austria,
information deduced from transient well tests on (1996), Sep. 3-6.
regular basis in reservoir description and 6. He, N., Reynolds, A.C., and Oliver, D.S.: “Three-
performance predictions using numerical simulation Dimensional Reservoir Description from Multiwell
is presented in this study. Pressure Data,” paper SPE 36509 presented at the
2. NWT preserves the wealth of information usually 1996 SPE Annual Technical Conference and
obtained from analytical analysis of well tests at Exhibition, Denver, CO, Oct. 6-9.
least as initial information for the final model. 7. Wen, X., C. Deutsch, and A. Cullick: “High-
3. NWT presents a practical process to introduce the Resolution Reservoir Models Integrating Multiple-
results of well tests as they become available into Well Production Data,” SPE Journal, December
the reservoir numerical simulator. 1998, Vol. 3, No. 4, 344-355.
4. Using a number of numerical reservoir models of 8. He, N., and Chambers, K.: “Calibrate Flow
increasing complexity in a sequential process was Simulation Models with Well-Test Data to Improve
shown to be an efficient and practical method to History Matching,” paper SPE 56681 presented at
integrate transient testing data in reservoir the 1999 SPE Annual Technical Conference and
simulation models. Exhibition, Houston, TX, Oct. 3-6.
5. NWT was successfully used in two field examples. 9. He, N., Oliver, D. and Reynolds, A.: “Conditioning
The results showed that incorporating pressure Stochastic Reservoir Models to Well-Test Data,”
transient data improved the efficiency and accuracy SPE Reservoir Evaluation & Engineering Journal,
of the simulation study. February 2000, Vol. 3, No. 1, 74-79,
6. To facilitate the use of NWT, software linking the 10. Landa, J. L., Kamal, M. M., Jenkins, C. D., and
results from analytical analysis to numerical Horne, R. L.: “Reservoir Characterization
reservoir simulators should be developed. Constrained to Well Test Data: A Field Example,”
SPEREE, August 2000, 325-334.
Acknowledgement 11. Landa, J.L.: “Reservoir Parameter Estimation
The authors thank the managements of Chevron Constrained to Pressure Transients, Performance
Corporation and Chevron Canada Resources for History and Distributed Saturation Data,” Ph.D.
permission to publish this work. We also thank our dissertation, Stanford University, Stanford, CA,
colleagues Kaveh Dehghani, Akshay Sahni, Mark 1997.
Langston, Irene Gullapali, Muu Hoang, Jill Matieshin, 12. Landa, J. and Horne, R.: ‘‘A Procedure to Integrate
Terence Moynihan for providing field data and making Well-Test Data, Reservoir Performance History and
valuable suggestions. 4D Seismic Information Into a Reservoir
Description,’’ paper SPE 38653 presented at the
SPE 95905 13

1997 SPE Annual Technical Conference and


Exhibition, San Antonio, Texas, 5–8 October.
13. Anterion, F., Eymard, R., and Karcher, B.: “Use of
Parameter Gradients for Reservoir History
Matching,” paper SPE 18433 presented at the 1989
SPE Symposium on Reservoir Simulation, Houston,
TX, Feb. 6-8.
14. Raghavan, R., Dixon, T.N., Phan, V.Q., and
Robinson, S.W.: “Integration of Geology,
Geophysics, and Numerical Simulation in the
Interpretation of a Well Test in a Fluvial Reservoir,”
SPE Reservoir Evaluation & Engineering Journal.
(June 2001) 201.

View publication stats

You might also like