Q
SPE 50946
Numerical and Analytical Well Test Analysis: A Case History
W. John Lee and John P. Spivey, SPE, S. A. Holeltch & Associates, Inc.
‘multilayer, multiphase reservoir. Some engineers believe that
recent improvements in analytical models have made
‘mumerical modeling for well test interpretation all but
‘obsolete. Others believe just as strongly that sumerical
‘models should be used for routine well test interpretation.
Still others believe that both tools are useful, and that a
‘combination of the two approaches provides the best general
strategy for reservoir characterization.
‘This field case focuses on the Gas Research Institute (GRI)
Staged Field Experiment (SFE) No. 2. This well was
‘hydraulically fractured. A variety of different analysis
‘techniques were applied in order to evaluate the consistency
between reservoir and fracture properties estimated by the
different methods of analysis.
‘Analysis methods applied to the flow and buildup tests
performed on this well include type curve matching using
analytical solutions, numerical modeling with single-phase,
‘multi-layer radial and Cartesian grids, and merical modeling
with multiphase, multi-layer radial and Cartesian grids. The
simpler models were able to match segments of the production
‘and pressure history; only the full multiphase, multilayer
mode! was able to satisfactorily match the complete
production and pressure history.
‘We have found that an approach combining analytical and
‘mumerical models has been the most successful strategy in our
‘work. As in this field case, analysis begins with the simple
‘models and proceeds through more complex models as
‘necessary to match the reservoir behavior in the desired detail
153
Introduction
‘At the May, 1998, SPE Pacific-Asia Forum on “Integrating.
Well Testing and Reservoir Characterization,” there was an
catire session devoted to use of numerical models in well est
analysis. The reason for the session was that some engineers
fool that, given the increasing sophistication of analytical
models available in commercial software, we can obtain a
reservoir description consistent with available data in most if
not all cases. That is not the same as saying itis the “corect”
reservoir description: that, we will never know becaute we
will never have sufficient data. However, some feel that a
sufficient analytical model can almost always be found.
Other engineers believe that mmerical models are all we
need today, and that we can use them to history match test
data to obtain the reservoir description we desire. This camp
believes implicitly that real reservoirs have complexities than
‘cannot be captured with the most comprebensive analytical
models. Therefore, we should include in our model all the
complexives of which we are aware, and proceed to adjust
reservoir parameters until we match the dats, completely
‘bypassing the step of test interpretation with analytical
‘models.
To still other engineers, a combination of the two
approaches is of considerable value in certain situations with
complex reservoir geometry or multiphase fluid flow. This
paper discusses a case history in which the combination
approach proved valuable. The case history summarized in
this paper is an analysis of a hydraulically fractured, low-
permeability gas reservoir. The analysis was done by Peter
‘Bastian and Jan Sherman and is summarized in reference 1
‘and discussed in more detail in reference 2. I have selected
this particular case history to illustrate how analytical and
‘mumerical methods can be combined because the case is
typical of reservoir characterization needs in tight case
formations; i¢., the formation characteristics in this example
arise repeatedly in practice.
‘The Situation
‘The well characterized in this example was the Staged Field
Experiment (SFE) Well No. 2, located in
County, Texas. This well, drilled as part of the Tight Gas
Sands program funded by the Gas Research Institute (GRD,2 W. J. LEE and J.P. SPIVEY
SPESO0S \
was hydraulically faced in the lower Travis Peak
formation. The fracture was modeled with a three-
dimensional fracture propagation model, and it was of great
interest in the research program to determine the consistency
of fracture properties determined from fracture models and
from post-fracture production and transient test analysis.
In the lower Travis Peak formation in SFE No. 2, there
‘were two mini-fracs followed by two main fracture treatments
totaling 331,000 gallons of el and 188,000 pounds of
intermediate strength proppant.
The Travis Peak formation is a massively thick, deltaic
deposit composed of stacked channel belts and flood plain
deposits. Multiple gas and water-bearing sands are present in
the vertical sequence. ‘Thick shale beds are absent; thus,
hydraulic fractures are expected to grow vertically. without
comainment, placing numervus gas and water sands in
communication with each other through the fracture and with
te wellbore.
‘The properties of the multiple layers in the Travis Peak
formation vary significantly. The layers tested in the SFE No.
2 produced fracture fluid followed by formation water
‘throughout ‘their flow histories; thus, two-phase, gas-water
flow bad to be modeled. Further, the hydraulic fracture
changed the geometry of the flow system near the wellbore
‘and altered the fluid distributions near the wellbore. By any
‘measure, tis is a complex flow situation.
‘Post-fracture production performance of the well is shown
in Fig. 1. Notable on this figure are (1) gas breakthrough at
about 22 days after the fracture treatment; (2) the beginning of
bottom-hole pressure measurements after about 31 days; (3)
shut-in of the well for a four-point test at sbout 41 days; (4)
recovery of a water volume equal to the load volume in about
46 days, and (5) shut-in for a pressure buildup test after about
83 days. A 30-day shut-in period followed. Note that the
‘water production rate never dropped below about 100 STB/D,
‘even long after the load volume had been recovered.
Searreacsanna, 0
‘Note also a linear decline in pressure during the flow
period, suggesting a flow regime similar to pseudo-steady-
state and, and more importantly, a very limited reservoir.
Performance Analysis
Bastian and Sherman analyzed the production and test data
‘with successively more complex models. Fist, they attempred
analysis with conventional type curves for hydraulically
fractured wells in infinite-acting formations, which assume
single-phase flow and a single-layer formation. Given the
complexity of the situation, we might ask: Why bother? The
answer is that a single phase and a single layer might dominate
the test performance; we need to determine whether this is the
‘case and also determine fracture and formation properties that
should be directionally correct. This philosophy is not unlike
‘that of using a tanktype material balance as a first step in a
‘complex multi-dimensional simulation. Frequently, much can
be teamed.
Fig. 2 shows the shape of the pressure and pressure
erivative graphs. Plotting functions include normalized
pscudo-pressures and radial equivalent shut-in times. Fig. 3
shows @ match with an infinie-conductivity type curve
(Gncluding wellbore storage) and Fig. 4 shows a match with a
finite-conductivity fractare type curve (but with high
conductivity, so there was nothing new here except to provide
‘more confidence in the infinite conductivity curve match).
‘The result of these matches was a gus permeability of 0.0076
im and a fracture half-length of about 280 - 290 ft. (See
Table | near the end of the paper.) The matches are good, end
if the good match were our sole criterion, we could say that we
hhave found an adequate reservoir description. However, the
significant and varying water production rates, the delayed gas
production rates, and the observed bottomhole flowing
‘pressures have to be consistent with the proper reservoir
Gescription. Regardless of the type curve match, we must
check consistency with other data,
gett rem cen
:
154‘nt Const Wats Sa
4— Finke conduetty facture snatysis of
‘ar pressure boi, Lower Tava Pak ca
Numerical Modeling
‘Two types of gridding were used for numerical modeling, as
ilustrted in Fig. 5. A two-dimensional radial model was
‘used for rapid screening of alternatives, and a three-
‘dimensional Cartesian model was used to model the expected
flow geometry more rigorously. This approach is similar in
principle to following tank-type material balance in a
reservoir study with first « simplified one or two-dimensional
model to identify major reservoir features, followed by a more
rigorous (but more time-consuming) three-dimensional model.
155
NUMERICAL AND ANALYTICAL WELL TEST ANALYEIC: AGASE HISTORY 3
‘TWO-DIMENSIONAL RADIAL MODEL
(MODEL TYPE | AND It
Layer 1
Fig. 5 Schematics of reservoir model geometries used In Travis.
oak history matching.
Bastian and Sherman noted that use of the radial model in
history matching required transformation of fracture properties
from a Cartesian coordinate system to a radial coordinate
system, Details of the transformation are provided in
reference 1.
Variables compared in the history matching procedure
‘included (1) the post-facture cleanup period during which the
well produced only water; (2) the time of initial gas
‘Production following the cleanup period; (3) the post-fracure
‘Production performance; and (4) the post-fracture buildup test.
Bastian and Sherman wed several different combinations
‘of models in the process of obtaining a final history match.
‘They started with simple, single-phase models and proceeded
‘to more complex, multi-layer, two-phase models.‘4 W.J. LEE and JP. SPIVEY
SPE sone
Single-Phase Modeling
Single-phase modeling was done first to provide initial
estimates of reservoir and fracture properties. Both radial and
Cartesian coordinate systems were employed numerous times.
The primary limitation of this work, of course, was the
analysts’ inability to model fracture fluid cleanup and gas
breakthrough, Therefore, the history matching began at the
start of gas breakthrough; water production before and after
this time was ignored.
Tike multi-layer system. Fig. 1 indicates an almost linear
ecline in pressure during the flow period, similar to pseudo-
steady-state flow. and characteristic of a limited system. The
flow period and shut-in period pressures could not be matched
simultaneously with a single-layer model of limited areal
‘extent. Fig. 6 shows a match of the drawdown pressures with
1 single-layer model; Fig. 7 shows an attempted match of the |
‘shut-in pressures, with the same reservoir description. The
limited reservoir size used to match the flowing pressures also
affects the calculated shut-in pressures and causes the shape of
the graph of the simulated pressures to be different from that
of the observed pressures.
Fig. 6— History match of Lower Travis Peak drawdown pressures
ting single-phase, single-layer areal mol - Example Match No.
156
Tie Bees Bate urs
mgmt et
Sera pec ae ae
Figs. 8 and 9 show different matches of both data sets.
The shape of the shut-in pressure graph is matched more
closely. ‘In this case, reservoir permeability-thickness product
‘was reduced and reservoir size was increased to improve the
shut-in pressure match — but the linear tread in flowing
pressure dectine is not reproduced. Thus, there is a need for a
reservoir description that bebaves somewhat like a limited
system during the flow period and somewhat like a larger
system during shut‘in. Single-layer, multi-phase flow systems
failed to reproduce the observed reservoir performance
characteristics, so subsequent modeling efforts included
‘multiple layers and multi-phase flow.
ewig Stam Prema a
1NUMERICAL AND ANALYTICAL WELL TEST ANALYSIS: A CASE HISTORY 5
‘roe rautont sT80
‘Slcoecornn a
Fig. 0 — History match of Lower Travie Pesk bulldup prosoures
sig engl phase sigleayerarel mode —| ‘Example Match No.
‘Simulating Cleanup and Gas Breakthrough
To sinmlate the long time to gas breakthrough, Bastian and
Sherman used two techniques. The first was to place a region
of increased water saturation around the fracture, with the
amount equal to the net load volume remaining to be produced
at the start of production. The second technique was to
employ relative permeability hysteresis, using imbibition
relative permeability curves constructed from the only
laboratory data available, drainage relative permeability
‘curves. Residual gis saturation on the imbibition relative
permeability curves was an important history-matching
parameter.
Multitayer, Multiphase Modeling
‘The history match was improved considerably with a two-
layer reservoir description. One layer produced mostly gas
and was limited in size. The second layer contained mobile
‘water and was essentially infinite acting. Both layers
communicated with the wellbore and with each other through
the hydraulic fracrure. Imporcanty, this reservoir description
is consistent with the complex nature of the Travis Peak
formation and with all known well and completion data. Of
course, there are doubtless more than two layers that
‘contribute flow, but the field performance is at leat consistent
with the simplified two-layer system.
Figs. 10 and 11 present the history match ofthe flow and
shut-in data with the radial model. The match captures many
aspects of actual well performance, especially shut-in
‘pressures. Gas breakthrough time and water rates match wel.
Flowing pressures are lower than observed, but follow the
‘correct trend,
157
‘am tes et are
1 — Hor maton of tor Tavs Peak bul pressures
(Shag twophaas mutiayr scsi mocet os NM
‘These results from the radial model are not the final match.
The radial model has the advantage of speed; it allows
relatively rapid examination of a umber of sltematives
However, it does not capture the detail of flow near the well
correctly. For this reason, a final three-dimensional Cartesian
model was also developed. History matches are shown in
Figs. 12 and 13. With the exception of flowing pressures,
(rend is correct, although the pressures are low), this reservoir
description leads to a match of the key characteristics of
‘observed well performance.
‘A summary of reservoir descriptions from conventions!
analysis and reservoir simulation is given in Table 1.as Prteion a a
‘Scrreocton tm ee?
W.J. LEE and J.P. SPIVEY ‘SPE 0048
Fig. 12 — History match of
performance
Lower Travie Peak production
i.
I
Fig. 12 — History match of Lower Travie Peak bulldup pressures
sing two-phase multlayer areal model.
"TABLE 1 - SUMMARY OF CONVENTIONAL ANALYSIS AND RESERVOIR SIMULATION RESULTS
LOWER TRAVIS PEAK
Siuulation History Matching
Single-Phase ‘Two-Phase
Single-Layer Two-Layer
‘Conventional Analysis ‘Cartesian Radial Caress
Tafinite Finite
Conductivity | Conductivity
Fracture | Fracture
Type Curve | TypeCuve | Caset_| Case? | avert | Laver? | tavert | Laver:
Gas permeability, md 0.0076 0076 [0.012] 0.0070] 0.0014] 0.0019 | a.0014 | 0.0019
Net pay, a a 4 4 35 a 35 a
Fracture hal-lengt, ft 289 280 160 160] 310 310} 300 300
‘Oa porosity, % 45 45 29 20 29 20
Area, 26 24 160 WwW 40 26 160
Formation permeabitity, oor oos| 001] 0.08
nd
‘Water permeabiy, md 0.0003 02 | 0.0003] 0.02
Total porosity, % 45 49 45 49
Water saturation, % 35, oo 2» oo.
158SPE 2096 NUMERICAL AND ANALYTICAL WELL TEST ANALYSIS: A GASE HISTORY
Discussion and Conclusions
‘Wo must ack: Is the match chown in Figs, 12 and 13 corrcot?
In response, we say that the match is sufficient; it appears to
be adequate for forecasting, and it is also consistent with the
results of three-dimensional fracture modeling. In the longer
term, the model might need to be modified to provide a match
‘of production after several months or years, perhaps caused by
other layers becoming dominant. No dats available at the time
of the match, though, indicated that the reservoir description
is not adequate. The same cannot be said of the single-phase,
single-layer matches; thus, the value of numerical modeling in
this case.
‘But is the match unique? The short answer: no. Many
possibilities were eliminated during the history-matching
process; sill, there are other possibilities. For example, here
is the possiblity that two layers are insufficient for a proper
reservoir description that will lead to a “correct” 20-year
forecast.
‘Was the match with the analytical solution helpful?
Certainly. It provided a valuable starting point. Results in
Table 1 show a remarkably (and perhaps coincidentally)
accurate” fracture half-length estimate and order-of-
smagnitude-comrect gas permeability estimate. This model
cannot, however, lead to 2 useful forecast of future well
performance. Multi-layer, multi-phase analytical models
‘could be expected to do better, but must allow different
‘drainage areas in different layers and different gas and water
sanurations and relative permeablliy characteristics in the
layers. Further, they must model fractures in some layers and
cleanup of fractare fluids in case like the present one.
References
1, Bastian, P. A. and Sherman, J. B.: “Analysis of Hydraulically
Fracture, Low-Permeability Gas Reservoir Using, Numerical
Simulation,” paper SPE 21511, presented at the SPE Gas
‘Technology Symposium, Hoosten, Texas, January 73-25, 1991
2, “Staged Field Experiment No. 2: Applicaton of Advanced
Geological, Petophysical and Engineering. Technologies to
Evaluate and Improve Gas Recovery from Low Permeability
Sandstone Reservoirs ~ Vol. I.” CER Corporation and S. A.
Holditch & Associates, Inc, April 1990, GRI Contract No.
5088-211-1682.
159