Professional Documents
Culture Documents
Spe 12013 Pa
Spe 12013 Pa
Summary. A new analytic solution, based on an approximate trilinear flow model, is developed to study the
transient behavior of a well intercepted by a finite-conductivity vertical fracture. The solution accounts for the
effects of skin, wellbore storage, and fracture storage. Both constant-pressure and constant-rate cases are consid-
ered. The solution is simple and reliable for short-time analysis. Combining this solution with a semilog
asymptotic solution provides a reliable tool for analysis and formation evaluation of fractured wells. We also
demonstrate that the optimization technique is a convenient means of formation parameter estimation. A set of
early-time asymptotic solutions is also presented. These solutions provide qualitative and quantitative relations of
the simultaneous influences of wellbore storage, fracture storage, and skin damage on early-time wellbore
pressure behavior.
Introduction
The increased activities in exploiting tight reservoirs by fractures. The bilinear model is applied when the influ-
means of hydraulic fracturing techniques 1 have gener- ence of the flow from fracture tip is not felt within the
ated considerable interest in the development of pressure- fracture. Also, while the slope of pressure vs. the fourth
testing procedures for evaluating fracture performance. root of time will provide an estimate of fracture conduc-
Pressure or production analyses of fractured wells have tivity, the fracture length cannot be obtained directly.
been investigated with numerical as well as analytic Therefore, we conclude that no simple, suitable analytic
methods. Cinco-Ley 2 surveyed published numerical and model for formation evaluation of fractured wells, capa-
analytic reservoir flow models for fractured wells. A nu- ble of providing both fracture length and conductivity,
merical approach with a reservoir simulator 3 ,4 can has been developed.
rigorously treat nonlinear fluid/rock properties, as well We present a new analytic mathematical model for flow
as formation heterogeneity and geometry. From the stand- to a fractured well. On the basis of physical and mathe-
point of ease of analysis, however, we prefer analytic matical reasoning, we approximate the flow between the
models, if applicable, over numerical models. formation and the fracture as having a trilinear behavior.
For pressure testing of wells, analytic models can be This model, called the trilinear model, considers the ef-
grouped into two types according to their solution fects of skin, wellbore storage, fracture storage, and
methods: semianalytic and asymptotic analytic models. constant-pressure and constant-rate cases. The solutions
The semi analytic model was first developed by Gringar- are simple and reliable for short-time analysis (the time
ten et at. 5 for infinite-conductivity fractured wells and before semilog straight-line behavior is reached) of a well
was later extended to finite-conductivity fractured wells intercepted by a vertical fracture. Combining our short-
by Cinco-Ley et ai. 6 In these approaches, the govern- time solution with semilog asymptotic solutions S,8 pro-
ing linear partial differential equations were transformed vides a reliable tool for pressure testing of fractured wells.
first to a set of integral equations. Then, these integral We also present a method of formation parameter esti-
equations were discretized in time and space to find the mation by means of an optimization technique. This proce-
unknown variables of pressure and flow in the fracture. dure requires an optimization (or error minimization)
Adapting this algorithm for routine well testing purposes subroutine. We demonstrate that the fracture parameters
would require considerable computer coding and can be determined conveniently with the optimization
storage-comparable to the requirement of a purely nu- technique and the trilinear model.
merical approach. In asym~totic analytic solutions, a Finally, we present early-time asymptotic solutions for
square-root-of-time solution ,8 and recently developed both constant-pressure and constant-rate cases to illustrate
asymptotic bilinear solutions 9 ,10 are used for formation the simultaneous influences of skin, wellbore storage, and
evaluation of fractured wells. The square-root-of-time so- fracture storage at early testing times.
lution is applicable only to short and high-conductivity
Development of the Trilinear Fracture
'Now with Sohio Petroleum Co. Formation Flow Model
"Now with Alcoa Research Center, PittSburgh, PA. Following Cinco-Ley et al., 6 we consider a single ver-
Copyright 1986 Society of Petroleum Engineers tical fracture of finite conductivity intercepting the well-
SPE Formation Evaluation, February 1986 75
In Region 2 (formation flow), O:O;XD :0;1, and
O:O;YD<oo.
FRACTURE FACE
T
h
~
..1-- . ~.
• _________________ • ~
. y
where
WELLBORE I--Xf X_a:> P2D=0 at tD=O .......................... (6)
Fig. 1A-lnfinite reservoir with a vertical fracture.
and
OP2D
PID=P2D-S-- at YD=O, ............... (7)
y 1 (2)
,I
I,
°YD
where
(1). (2). (3) FLOW REGIONS
CONSTRUCTED I
i
~'.::
:. (3)
P2D=0 at YD-+ oo . ........................ (8)
. t-
FRACTURE
J ,. In Region 3 (formation flow), 1 :O;XD < 00 and
(JL"'=(1 - ~-_x
--Xf- O:O;YD<oo.
Fig. 1B-A quadrant of the top view of the fractured well
system and trilinear flow approximation.
. ......................... (9)
OPID
OXD
I _=b(I-CDj OPWD)
xD-O otD
............ (3)
kh[m(Pi)-m(p!)]
Cl.gqT
and
{3kt
tD=----
Op ID
OXD
I x D =!
=0.. ........................ (4)
¢p.ctx/ '
FD =kfbtfk.xf ,
C
C Df = ,
27rc/>c t hx/
1 kc/>jCft 0_ ro - O.2IoEPJ, TRILINEAR
o-fa - O.2"P!, CINCO
C I =-=---, l> - ro - PI, TRILINEAR
.. - fO - PI, CINCO
Tlf ktcf>c t )( - fO -
0_ ro -
lO"PI, TRILINEAR
!Ol,PI, CINCO
2
a=-,
FD
o~~~~~~~~
l~ ,if Id 1~' l~' l~ 1~
and D! MENS I DNLESS T JME
Notice that C Df is the dimensionless wellbore storage Evaluation of the Trilinear Solution
based on fracture half-length, xf'
Applying Laplace transforms to Eqs. 1 through 12 and Cinco-Ley et al. 6 presented a numerical solution to our
solving the resulting ordinary differential equations, we fracture flow problem with tabular results for constant-
obtain the complete solution to the initial boundary value rate drawdown tests for dimensionless pressure vs. dimen-
problem in Laplace space. The derivation is outlined in sionless time for various fracture conductivities. The
Appendix A. dimensionless time varied from 1 X 10 - 3 to 1 x 10 3 and
For constant flow rate at the wellbore, the wellbore included bilinear flow, linear flow, and pseudoradial flow.
pressure in Laplace space is Our solution was inverted with the Stehfest algorithm 12
and compared to Cinco-Ley et al. 's data for a range of
b conductivities (Fig. 2). The trilinear model matches
PwD(S) = ............. (13a) Cinco-Ley et al.'s solution until about tD=l.O. After
s(sbCDf-1/; tanh 1/;) t D = 1.0 the deviation becomes appreciable. The applica-
bility of the trilinear model, however, is not limited to
Similarly, for the constant-pressure case, the wellbore well testing, for two reasons. First, running a well test
flow rate, qD(S), is obtained from until pseudoradial flow often is not practical. This fixes
t D < 1.0 and makes the trilinear model applicable
1 I/; throughout the entire test. Second, however, because the
qD(S)=-2-_-- tanh 1/; •.......... (13b) trilinear flow solution is valid until about tD = 1.0, it can
S P wD(S) sb
be matched to the pseudoradial flow solution to provide
a solution valid for all times. This was done by matching
Note that with a constant wellbore pressure, the wellbore of the trilinear model to pseudoradial flow at the point
storage effect no longer exists in Eq. 13b; that is, the first where the slope ofpwD vs.log(tD) is 1.151 or where the
equality of Eq. 13b, i.e., slope of 11q D vs. log(t D) is 1. 151 for the constant-
pressure case. If this did not occur before t D = 1, the so-
1 lutions were matched at t D = 1 because this point marks
q(s)= 2- , roughly the end of applicability of the trilinear model. This
(s P wD) method of asymptotically matching our trilinear solution
to the pseudoradial solution gave satisfactory results. Fig.
holds only for cases without the wellbore storage effect. 3 shows graphically the trilinear solution with asymptot- .
In Eqs. 13a and 13b, the parameter I/; is defined as ic matching. In accordance with a conventional type-curve
plot,9 Fig. 3 plots 10g(PwDFD) vs. log(tDF2 D) with F D
serving as a parameter. Cinco-Ley et al. 's semianalytic
.......... (14) solutions,6 plotted in dots, is also compared with our so-
lution. Fig. 3 shows that for practical purposes our
trilinear model with asymptotic matching to the pseu-
When C I =S=CDf=O, PwD(S) and qD(S) can be in- doradial flow solution gives a sufficiently accurate
verted analytically to real-time space. Details of the ana- representation of a converged numerical solution for the
lytical inversion and the expression of PwD(t D) in terms problem of interest. In addition, similar plots can be
of a triple integral because of the coupled trilinear model generated easily from the trilinear model for cases con-
are given in Appendix B. The integrands in the integral taining skin, wellbore storage, and fracture storage. In
are well-behaved functions. The triple integral can be Table 1, the trilinear solution with asymptotic matching
computed numerically with a Gaussian quadrature is further compared with Cinco-Ley et al. 's numerical so-
method. The results from the analytical solution agree with lution. For the range of data compared, the deviation is
those from Eq. 13a and Stehfest's inverting algorithm. 12 largest for F D = 0.2 7r at around t D = O. 1 where the rela-
SPE Formation Evaluation, February 1986 77
10
10'~---
I
- THISWORK
~~
10 .- -- -f--
••• CINCO ET AL
....- k:::::;~
~
~ ~
~..--
10-0
-
~
~ ..IlL
t---
.
10 -.-~ ~-
~ -.IL
~
~~
r;.--
.a.
10' ..!!Lll
~
....oL.lJL
~
0 ____ I"""
~
10-" 10-3 10-' 10-1 10' 10' 10' 10' 10' 10· 10' 10'
Fig. 3-Comparison of the type curves generated by the trilinear model (with asymptotic match)
to the numerical solution of Cinco-Ley et al. 9
tive deviation is about 5.5 %. For most of the data com- Eq. 15 was inverted and compared to the trilinear so-
pared, however, the deviations are within about 2 %. lution in Fig. 4. They are practically identical until about
Other analytic models have been proposed. The one that t D = 1 X 10 - 2 . Because the bilinear model does not match
most closely matches the numerical solutions is a bilinear the trilinear solution after t D = 1 X 10 - 2 , it cannot be
flow model of Cinco-Ley et al. 9 They superimpose two matched asymptotically to a pseudoradial solution; that
linear flows-a linear incompressible flow in the fracture is, the bilinear model is two log cycles short of matching
and a linear compressible flow in the formation. It also the entire short-time regime, while the trilinear model is
yields a simple expression in Laplace space without skin, accurate over the entire short-time regime.
wellbore storage, and fracture storage: The solution for constant-pressure testing can be gener-
1r
ated with the trilinear model by exploitation of the sim-
. . .. (15) ple relation q(s) = 1.0/[s2ji(s)] (see Eq. 13b and/or Ref.
ji wD(S) = ( 2 ) V2
14). Type curves for constant-pressure testing were gener-
(2F D) 'h S 1.25 tanh FD S 'A
ated by Agarwal et al. 3 with a numerical simulator that
plotted the reciprocal dimensionless rate vs. dimension-
Under the same condition, Eq. 15 is identical to the less time for a range of conductivities. Dimensionless time
bilinear solution derived previously in Ref. 13. varied from 10 -5 to 1.0. A comparison of our results
The asymptotic limit of this expression for small times with these curves shows good agreement over the range
yields an asymptotic bilinear solution 9 where PwD is of dimensionless time considered by Agarwal et al. To
proportional to t D 'A • The conventional bilinear solution extend the solution beyond the transient regime, an asymp-
that exhibits the above-mentioned behavior will be labeled totic match for flow rate can be made the same as for pres-
the asymptotic bilinear solution. From Eq. 15, the limit sure. Fig. 5 shows an example type curve generated with
for large time yields the linear flow regime where PwD the trilinear model. Laplace inversion with Stehfest's
is proportional to t D Ih • algorithm 12 is convenient and accurate for this problem.
"r---------------
o· fO .. O.2J11Pl, TRILlNE:AR
~s 0- fa .. D.2".PI, BILINEAR
::>..: 6'" F"D .. PI, TRILINEAR
~
"-
+ .. fO .. PI, BJLIN[AR
_ x .. fO .. lOMPI, TRILJNERR
... fa .. lOMPI, BILINEAR
E~ 0" ro ... O.2,..PJ
o-rD .. PI
~ lo" ro .. lOIlP!
~~
~ci
8L----~
°lt~~~~~~~10~·--~~~~~107·,--~~~~~I~
DIMENSIONLESS TlME
Pressure Testing With the well. Also, we will use the parameter optimization (or
Optimization Method error minimization) method to determine the formation
The type-curve method for evaluating formation param- parameters. The method is equally applicable to other
eters is used often in well testing. The method is con- formation evaluation problems.
venient if one or two parameters are to be determined from In the parameter optimization method, an objective
a two-dimensional graphical matching of the measured function is defined that is normally the sum of squares
and the model-predicted data. If more than two param- of the differences between model predictions and meas-
eters must be determined simultaneously, the type-curve ured data. Then an optimization algorithm minimizes the
method becomes cumbersome. objective function and, in the process, determines the
We will demonstrate the applicability of the trilinear model parameters. This procedure can be performed in
model to parameter estimation with data from a fractured either t D space or Laplace (s) space.
SPE Formation Evaluation, February 1986 79
TABLE 1-Continued
Fo = 1011" Fo =2011"
Pwo PwD
to ~ (From Ref. 9) (From Ref. 9)
to ~
0.001 0.0860 0.0866 0.001 0.0713 0.0718
0.002 0.1092 0.1100 0.002 0.0939 0.0946
0.003 0.1266 0.1277 0.003 0.1110 0.1120
0.004 0.1410 0.1424 0.004 0.1254 0.1265
0.005 0.1537 0.1553 0.005 0.1379 0.1392
0.006 0.1650 0.1668 0.006 0.1491 0.1506
0.007 0.1753 0.1773 0.007 0.1594 0.1610
0.008 0.1849 0.1871 0.008 0.1690 0.1706
0.009 0.1938 0.1962 0.009 0.1777 0.1796
0.01 0.2022 0.2047 0.01 0.1863 0.1881
0.02 0.2688 0.2720 0.02 0.2526 0.2549
0.03 0.3183 0.3221 0.03 0.3020 0.3047
0.04 0.3592 0.3633 0.04 0.3428 0.3456
0.05 0.3944 0.3988 0.05 0.3780 0.3809
0.06 0.4257 0.4304 0.06 0.4093 0.4122
0.07 0.4540 0.4588 0.07 0.4376 0.4406
0.08 0.4801 0.4849 0.08 0.4637 0.4665
0.09 0.5043 0.5090 0.09 0.4878 0.4905
0.1 0.5268 0.5316 0.1 0.5104 0.5129
0.2 0.7003 0.7015 0.2 0.6838 0.6820
0.3 0.8240 0.8208 0.3 0.8075 0.8008
0.4 0.9227 0.9143 0.4 0.9062 0.8940
0.5 1.0059 0.9918 0.5 0.9893 0.9712
0.6 1.0783 1.0582 0.6 1.0617 1.0374
(J.7 1.1428 1.1163 0.7 1.1262 1.0955
0.8 1.2010 1.1682 0.8 1.1845 1.1472
0.9 1.2544 1.2150 0.9 1.2378 1.1939
1.0 1.3036 1.2577 1.0 1.2871 1.2365
2.0 1.6499 1.5546 2.0 1.6333 1.5329
3.0 1.8525 1.7394 3.0 1.8359 1.7175
4.0 1.9962 1.8739 4.0 1.9797 1.8519
5.0 2.1077 1.9798 5.0 2.0912 1.9577
6.0 2.1989 2.0671 6.0 2.1823 2.0450
7.0 2.2759 2.1414 7.0 2.2593 2.1193
8.0 2.3427 2.2061 8.0 2.3261 2.1839
9.0 2.4015 2.2634 9.0 2.3850 2.2412
10.0 2.4542 2.3147 10.0 2.4376 2.2925
20.0 2.8006 2.6553 20.0 2.7841 2.6331
30.0 3.0033 2.8561 30.0 2.9868 2.8338
40.0 3.1471 2.9989 40.0 3.1306 2.9766
50.0 3.2587 3.1099 50.0 3.2421 3.0876
60.0 3.3498 3.2007 60.0 3.3332 3.1784
70.0 3.4268 3.2775 70.0 3.4103 3.2551
80.0 3.4936 3.3440 80.0 3.4770 3.3217
90.0 3.5525 3.4027 90.0 3.5359 3.3804
100.0 3.6051 3.4553 100.0 3.5886 3.4330
200.0 3.9516 3.8013 200.0 3.9351 3.7789
300.0 4.1543 4.0038 300.0 4.1377 3.9815
400.0 4.2981 4.1475 400.0 4.2815 4.1252
500.0 4.4096 4.2590 500.0 4.3931 4.2367
600.0 4.5008 4.3502 600.0 4.4842 4.3278
700.0 4.5778 4.4272 700.0 4.5613 4.4049
800.0 4.6446 4.4939 800.0 4.6280 4.4716
900.0 4.7035 4.5528 900.0 4.6869 4.5305
1000.0 4.7561 4.6055 1000.0 4.7396 4.5831
The applicability of the optimization method and the lem, Laplace-space parameter estimation is more difficult
trilinear model to parameter evaluation may be illustrat- to perform than real-time parameter estimation. (Details
ed by the drawdown test of a hydraulically fractured gas of the procedures of Laplace-space parameter estimation
well presented by Agarwal et al. 3 The initial pressure and are listed in Refs. 15 and 16.) We will outline only the
formation properties were determined through a prefrac- procedure of real-time space parameter estimation.
ture pressure buildup test. In addition, the skin effect was 1. The unknown parameters, xf and kfbf , are initial-
assumed to be absent. 3 Our goals were to determine both ized with some reasonable values.
the fracture length included in t D and the fracture con- 2. Given xf and kfbf , we obtain tD and F D .
ductivity included in parameter F D. We conducted pa- 3. We convert the given measured data, q and t, to
rameter evaluation both in the real-time space and in the dimensionless quantities, i.e., q~(i) and tD(i) , where the
La~lace space with the field data given by Agarwal et superscript m denotes the measured data and i = 1 ... N
al. Using their study, we have found that for our prob- is the number of measured data points.
is calculated.
discussing the influence of fracture storage. In other work,
6. The minimization of E t is accomplished with a Cinco-Ley et ai. 19 examined the effect of wellbore
modified Levenberg-Marquadt algorithm.17 New esti- storage and fracture damage (skin) on the transient pres-
mates of xf and kfbf are made, and Steps 2 through 5 sure behavior of vertically fractured wells. Although a
are repeated until a minimum sum of squared errors is fracture storage factor was included in their mathemati-
obtained. cal model, its influence was not investigated.
With the above procedure, a set of parameters was Our purposes in this section are to derive early-time
determined for our example. The optimized fracture par- asymptotic solutions containing wellbore storage, frac-
ameters, xf and kfbf , are given in Table 2 for both real- ture storage, and skin damage and to use the asymptotic
time- and Laplace-space optimizations. Also, the well- solutions to study their simultaneous influences on pres-
bore flow rate calculated through the trilinear model with sure responses at early testing times. The influence of frac-
the optimized parameters is compared to the measured ture storage capacity is discussed through a dimensionless
data in Table 3. Table 3 shows that the optimized param- parameter, CfDf' and is defined as
eters are satisfactory in terms of error deviations of the
predicted data.
For the same set of field data, Agarwal et ai. 3 obtained
xf=727 ft [222 m] and k~f=294 md-ft [0.9 md 'm] with
a type-curve-matching method and their numerical frac-
ture model. The xf value obtained by Agarwal et ai. is In the following, a set of early-time asymptotic solutions
different from our results. The difference could be at- is summarized for cases of interest with constant-rate
tributed to (1) the possibility of numerical convergence boundary condition at the wellbore. Derivations of these
problems associated with the timestep and grid sizes in equations are given in Appendix C.
Agarwal et ai. 's numerical approach, (2) the slight differ-
ences in results predicted by different models, and/or (3) A Case With WeUbore Storage, Fracture Storage, and
different techniques used in determining the parameters. Skin. We begin with
Regarding the third factor, we feel that the optimization
technique is superior to the type-curve-matching method
in fitting a model to measured data.
The matched results are also compared in Fig. 6 for
time up to 100 days along with the prediction of the
asymptotic bilinear model. Fig. 6 shows that the asymp- 4 CfDf 0.5 FO.5
D t D 15
totic bilinear model begins to deviate from the solution --'--'--,,----- . + '" ) . . .... (18)
at about 0.8 of one day, much earlier than the first meas- 3 7r1.5CDf
ured data point. Note that normally the trilinear solution
deviates from the true solution at a dimensionless time As noted in Appendix C, Eq. 18 is derived from the as-
near 1.0. In this example, a dimensionless time of 0.6 sumption that the fracture storage factor, CfDf' is not
is equal to about 10 years of testing time.
In a related example, Hanson 18 applied our trilinear
"
model in conjunction with a nonlinear regression method
and estimated fracture parameters for a tight-oil and a -
_ - -
ASVMPTOTIC IILlNEAR MODEL
TAILlNEA" MODEl
• Measured flow rate data with other field data taken from Agarwal et al.
3 Fig. 8-Comparison of the early-time asymptotic solutions
.. Smoothed data. with the true solutions.
t Real-time parameter estimation.
; Laplace space parameter estimation.
..
2. ASYHPT SOLUTION. EON. ( 18)
cor • 0.005 RSYHPT SOLUTION. EON. (22)
tD C Df 3 7r1.5CD/
..,
3.
erDf - 0.01 RSYHPT SOLUTION, EON • (19)
Q
3
a..
.- Eq. 20 shows that if wellbore and fracture storages are
both operative, then, with the measured early-time pres-
sure data, a plot of (PwDltD) vs. the square root of time
..., should yield a straight line. This linear plot will allow the
simultaneous estimation of the wellbore storage factor and
..... a dimensionless group containing fracture conductivity
and fracture storage.
....,,.... ,.-' ,....
The early-time regime within which Eq. 18 is valid can
be established for individual well-testing cases by com-
tD
parison of numerical results of the general solution and
Fig. 7-Comparison of the early-time asymptotic solutions the asymptotic solution, i.e., Eqs. 13a and 18, respec-
with the true solution_ tively. A necessary condition for the valid-time regime
of Eq. 18, however, can be found by imposing
d(PwD)/dtD >0 for Eq. 18 for a constant-rate drawdown
vanishingly small. This assumption allows us to approxi- case. From Eq. 18,
mate Parameter l/; ofEq. 14 with (aiS +CIs) when tD
is small. Also, although the skin factor was included ini- 47r 3 C D/
tially in the derivation, it does not appear in the early- tD< ......................... (21)
time asymptotic solution of Eq _ 18. This is physically true CfDfF D
because at early testing times the wellbore pressure
response is affected first by the wellbore and fracture This inequality indicates that a necessary bound of the
storages and then by the skin damages existing between early-time regime for which Eq. 18 is valid is directly
the fracture and formation. When wellbore and fracture proportional to wellbore storage squared and inversely
storages are both present, the skin becomes a secondary proportional to the product of fracture storage and con-
factor in affecting the early-time pressure response. ductivity factors. Through numerical testing, we have
Eq. 18 shows that at the beginning of the testing, the found the bound specified by Eq. 21 to be correct qualita-
pressure response is controlled by the wellbore storage tively.
factor, CDf>O; that is, we can obtain from Eq. 18 this As noted earlier, Eq. 18 is derived under the assump-
classic expression of pressure response under the influ- tion that the fracture storage is not vanishingly small. If,
ence of wellbore storage, however, C fDf "," 0, then an appropriate early-time
asymptotic solution,
=~ [exp(~tD) -lJ,
PwD=tDIC D , for tD-+O . .................. (19)
PwD ........... (22)
II C Df
It is only when t D becomes slightly larger that t D 1.5 be-
comes appreciable with respect to t D in Eq. 18 and the where II =FDl/; tanh l/;/7r and l/;=2/(F DS) should be
fracture storage factor appearing in the second term on used. The derivation ofEq. 22 is also given in Appendix
the right side begins to have some influence on the pres- C. Again, as tD approaches zero, Eq. 22 reduces to Eq.
sure response. Therefore, the pressure behavior is no 19.
A Case With Fracture Storage and Skin Damage. The If the second term on the right side is neglected, Eq. 27
early-time pressure behavior is governed by is identical to Cinco-Ley et al. 's fracture linear flow so-
lution. 6 With a two-term approximation, Eq. 27 should
provide a better approximation to the general solution of
. Eq. 13a for a longer time range of validity. As in the
preceding case, however, we can obtain a necessary con-
dition of the early-time regime that validates Eq. 27. In
Eq. 23 shows that when the wellbore storage effect is ab- this respect, the necessary condition is
sent, the fracture storage effect will dominate the early-
time pressure response. As tD becomes slightly larger, (CfDf ) 2
the second term on the right side of Eq. 23 begins to have tD < ........................... (28)
some influence on P wD. Also, the effect of skin damage
between the fracture and the formation is felt at the Eq. 28 shows explicitly that the bound for the valid-time
wellbore. regime is directly proportional to the square of the frac-
Eq. 23 can be rearranged to form ture storage factor. This bound can be prohibitively small.
For example, a medium value of the fracture storage fac-
PwD 211"°·5 tor of 10 -3 (Ref. 6) gives
tDO. 5 (F DCfDf )0.5
tD<0.3XIO- 6 • .....•.•................• (29)
According to Eq. 24, a plot of (PwD/tD 0.5) vs. tD with
early-time pressure data must give a straight line. The The bound of Eq. 29 is about less than a minute in real-
y-axis interception and the slope of this straight line will time scale for normal cases.
allow the estimation of the fracture storage and skin fac-
tor simultaneously, provided that the fracture conductivity Conclusions
factor is determined from late-time data. 1. The analytic solution of the trilinear flow model ac-
As in the preceding example, we can obtain a neces- curately predicts the short-time wellbore flow and pres-
sary condition for the early-time regime within which Eq. sure responses of a hydraulically fractured well.
23 is applicable; that is, PwD of Eq. 23 must satisfy Combining our solution with the semilog asymptotic late-
time solution provides a convenient and reliable tool for
- > 0,
dpwD
- parameter and performance evaluations of fractured wells.
dt D Our model contains the essential physical features required
of a reservoir flow model.
for constant-rate drawdown case. From Eq. 23, 2. The optimization technique is convenient for estimat-
ing reservoir parameters. For the problem considered in
1 this paper, parameter estimation is more easily carried
tD < -SCfDf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (25)
out in real-time space than in Laplace space.
2
3. A set of early-time asymptotic solutions is derived
This inequality indicates that a necessary condition of the for constant-rate and constant-pressure cases. These so-
early-time regime for which Eq. 23 is valid is directly lutions show explicitly the simultaneous influence of well-
proportional to the product skin and fracture storage bore storage, fracture storage, and skin damage on the
factor. responses at the wellbore. In addition, they provide a con-
If the well is produced at constant pressure, the coun- venient means of estimating the above factors, with early-
terpart of Eq. 23 is time response data.
Nomenclature
..... (26) a = parameter defined after Eq. 12
b = parameter defined after Eq. 12
P 3D =p 3D(S,XD) = r
o
exp( -stD)P3(tD,xD)dtD'
.......... (A-IO)
dji -
- 3D I - - - s 'h-
P2D (S,YD,XD=I) . .... (A-4)
Appendix B-Analytical Laplace Inversion
of Trilinear Solution
dx D xD=1
When S=C D =C I =0.0, then for the constant-rate case,
the wellbore pressure in the Laplace space from Eqs. A-9
We assume in Eq. 5 for the formation flow in Region 2 and A-1O is rewritten as
that the x gradient of P2D is small compared to the Y gra-
dient. This assumption 9 makes the approximation that _ b 1 cosh[alh(s+sI/2)V2]
P2D has a weak x dependence in Eq. A-4. In this case, PwD(S)=--1 - - - 1 - - 1- - - - - - -
h
a s (s+sh)\4 sinh[a'h(s+sv2)IA(
applying Laplace transform to Eq. 5 and using Eq. A-4
yields
............................ (B-1)
d2 -
P2D V2 - Instead of finding an inversion of P lD' we consider the
--2-=(s+s )P2D' ................... (A-5)
dYD inversion of (sp wD)/ ( - bla 'h), and attempt to perform
_ PID 'h 'h To determine the inversion of Eq. B-2, we consider first
P2D- 112 IA exp[-(s+s ) YD] .... (A-6)
1+(s+s )2S
-,1 ( cosh aq )
and L sq sinh aq , .................... (B-3)
dji2D (s+s'h)'h _
dYD IYD=O
= - I/;
1+(s+s 2) 2S
I/; P lD· ...... (A-7)
where q=s 112 and a=a 'h. From Ref. 20 we find
dji lD
dYD
IYD=O'
.
+lerfc
(2n + l)a +a ]
lh . . .................. (B-4)
2t
we have
L- 1( cosh qa.)
q sinh qa.
= -a
at
~
[ 2t "12 L..J
n=O
00
.
x [ Ierfc
(2n+ 1)a.-a..
2t I12
+ Ierfc
(2n+ 1)a.+a.
I/;
2t 2
J] Therefore,
00
~ [. (2n+ 1)a.-a.
=t - 1/2 L..J Ierfc V:
I2
n=O 2t 2
=-1 r
1 - tdvv(t-v)-312 exp [ - -v- - ]
21r V2 J 4(t-v)
00 o
+ierfc
(2n+ 1)a.+a. ]
2t~
+ 1/2 t-
1 L:
n=O
[
[a.(2n+ 1) -a.] 2
3 '12 ) 00 duuF(u,a.) exp ( -U- ) ...... (B-ll)
x(411"v)
o 4v
(2n+ 1)a.-a.
xerfc + [a.(2n+ 1)+a.] From Eq. B-ll, we find the Laplace inversion of Eq. B-2
2t I/2
from
(2n+ 1)a.+a.]
xerfc , ................... (B-5b)
2t~
where
r
2
ierfc(x) =
=-1-
1 rtdvv(t-v)-1.5 exp [ --v- ]
oJ
erfc(x ')dx'. . ................. (B-6) 211"12 4(t-0
x
(-U4v2) . .(B-12)
In Eq. B-5b, we determined
X (_1-3) 1/2 )00 duuF(u,a) '12 exp
-1 cosh qa. 411"v 0
L---~
q sinh qa.
Because
We now define
-
f(t,a.)=L - 1[ cosh
. qa. ] . ................ (B-7)
q SInh qa.
we obtain
............................ (B-8)
From Eq. B-8, we obtain
........................... (B-13)
L
-1 ( cosh \4 a. ) \4 3 1,)00
=(411"t)12
S IA sinh s
With the identity
o
rt 1
L v(t')dt'=-L(v), ..................... (B-14)
xduuF(u,a.) exp ( - ::). . ............... (B-9) J
o s
PwD(tD)=-V
-b~tD dw-
1 ~w
-, dvv(w-v)-3/2
I12
a l2 27r
o 0
CDfs bsC Df
For the same input data, the wellbore pressure calcu-
lated by Eq. B-15 agrees with that by numerical inver-
sion of P wD(S) with Stehfest's algorithm.IO Because
Stehfest's algorithm is convenient and accurate for our
problem, the practical utility of Eq. B-15 diminishes.
CDfb s .
:5 + ... ) ..... (C-6)
...................... (C-1)
b
.................. (C-3)
_(a)
1/;- -
1/2
........................... (C-8)
S
1
P wD(S) = , ................... (C-9)
S(CDfS-/I )
when
Eq. C-ll is valid only when C 1 ,.,0 and at early testing =(C1s) 'h( 1 +-C
a
Yo I)V2 .............. (C-14)
times (when s is very large). It also involves fewer ap- 1 s
proximations than Eq. C-7.
Because tanh ~,., 1,
Cases With Fracture Storage and Skin Damage. If
CDf=O, Eq. 13a becomes -b
PwD(S)=-
-b 1 s~
PwD(S)=- , ................ (C-12)
s ~tanh~
where
.. .).
Here we assume that the fracture storage factor, C 1, is Thus, the inversion of the above equation yields
not vanishingly small, so that at early time, S is large and
tanh ~ ,., 1. Then Eq. C-12 becomes
7rtD
b 1 ( a )-V2
PwD(S)=- (C1)'h s1.5 1+ C1Ss
27r°. 5 t D 0.5 tD
- - - - . .. (C-15)
CfDfo.5FDo.5 C 1.5 F D 0.5
fDf