Professional Documents
Culture Documents
SPE 62916
Taigbenu and Onyejekwe9 demonstrated how the single- The singular part of the pressure solution given in Equation
phase flow equation could be solved by rewriting it as: (4) satisfies:
ps
ct p 2 psD = D (13)
2 p = lnk p + (2) tD
k t
Subtracting Equation (13) from (12) gives:
Incorporation of Singularity Programming Masuka-
wa and Horne2 and Sato13 applied singularity programming ko pD ps
in conjunction with boundary element methods to compute 2 pD 2 psD = D lnk pD (14)
k tD tD
pressure transients. Unlike the current study Satos solution
was performed in Laplace space. Singularity programming Noting that pD = pns s
D + pD Equation (14) can be simplifed
decomposes the solution into singular and nonsingular com- to:
ponents: s
pD = pns s 2 ns k0 pns ko pD
D + pD (3) pD = D
+ 1 lnk (pns s
D + pD )
k tD k tD
The singular solution used in this work is based on the (15)
pressure response of a well flowing at a given rate in an
The right hand side of Equation (15) includes both the
infinite homogeneous reservoir with uniform permeability,
singular and nonsingular solutions. This does not present a
k0 (Dake14 ):
problem however since the singular solution is known and
can be incorporated easily into the boundary element solu-
q ct r2
ps = pi + Ei (4) tion.
4k0 h 4k0 t
where Ei(x) is the exponential integral: Green Element Method To transform Equation (15)
Z s into an integral equation suitable for solution with GEM
e the right hand side terms, which do not depend on time or
Ei(x) = ds (5)
x s space derivatives of pns
D , are collected together:
M ZZ
X
k0 pns Example 2: Reservoir with Closed Boundaries
= G D
lnk pns
D +f d (21) The value of using singularity programming in combination
e=1 e k tD
with the Green Element Method can be demonstrated by
Equation (21) can be written in a more compact form in comparing well tests simulated using this technique to well
terms of the element matrices R, L, U , V and T : tests simulated using a commercial finite difference simu-
lator. The first example is a well test in a closed square
XM reservoir. The well is located at the center of the reservoir.
k0 p
Rij pj + Lij qj Uijl Vijl lnkj pl + Tij fj = 0 The simulator uses a standard Peaceman15 well index. The
e=1
kj tD reservoir properties are given in Table 1. The reservoir di-
(22) mensions are 3000ft by 3000ft.
Note that in Equation (22) the subscript and superscript The GEM solution used a mesh of 21 by 21 nodes and
have been dropped from pns D for clarity. the simulation used a grid of 21 by 21 cells. No local grid
Z refinement was used in the finite difference simulation. The
e G(r, ri ) reservoir is homogeneous, so an analytical solution could be
Rij = Nj d ij i (23)
e n generated for this problem. The finite difference simulation
Z is compared to this analytical solution in Figure 2. The
Leij = G(r, ri )Nj d (24) GEM simulation is shown in Figure 3.
e The well test response calculated using finite difference
ZZ simulation appears to show some wellbore storage effects,
Uijl = G(r, ri )Nj Nl d (25) however these are artifacts of the finite difference compu-
e tation approach. In the late part of the well test the data
simulated using finite difference simulation did not sense the
ZZ closed boundary until much later than it should have.
e Nj Nl Nj Nl
Vijl = G(r, ri ) + d (26)
e x x y y Example 3: Gridding Study The performance of the
ZZ finite difference method in reproducing the pressure deriva-
tive curve was investigated further by running more finite
Tij = G(r, ri )Nj d (27)
e difference simulations. To assess the role of gridding the
simulation was repeated with both uniform and nonuniform
Equation(22) was solved using a fully implicit treatment i.e: grids. The grid parameters used are shown in Table 2. For
the nonuniform cases the geometric factor is the ratio be-
XM
k0 pm+1 pm tween adjacent grid cell sizes i.e. in the last case the cell
Rij pm+1
j + Lij qjm+1 Uijl l l
containing the well is 5ft by 5ft and the adjacent cells are
e=1
kj t
6.5ft by 6.5ft. The grid used for the first nonuniform case
is shown in Figure 4.
Vijl lnkj pm+1
l + Tij fjm+1 = 0 (28)
The time stepping scheme was the same for all the mod-
els, both GEM and finite difference. The finite difference
Results solution was calculated and reported in the following man-
Example 1: Comparison to an Analytical Solution ner:
- 10 steps of 0.0001 days
Example 1: Well Test in a Closed Reservoir To - 10 steps of 0.001 days
test the combination of singularity programming and GEM - 10 steps of 0.01 days
a pressure transient was computed and compared to a tran- - 20 steps of 0.1 days
sient generated analytically. The reservoir properties are - all subsequent time steps one day
given in Table 1. The reservoir dimensions are 2500ft by
3000ft. The finite difference simulator was able to compute the
The grid used for the GEM solution of the nonsingular solution at intermediate times also if its time step control
component of the problem was a 10 by 10 mesh. The well algorithm required it.
was located in the center of the reservoir. Constant pressure The pressure and pressure derivative curves computed
boundaries were imposed on all sides. The resulting draw- using the uniform and nonuniform grids are shown in Fig-
downs are compared to the analytical solution in Figure 1. ures 5 to 10. The derivative curves for all the uniformly grid-
The agreement between the GEM and analytical solutions ded cases show that the effect of the boundary on the deriva-
is excellent. tive curve is resolved better as the grid is refined. However
even when the grid is refined to 101 by 101 cells the bound-
Examples 2 and 3: Comparison to Finite Difference ary location is still sensed incorrectly. All the uniformly
Solutions gridded cases exhibit a period of infinite-acting radial flow
4 THE GREEN ELEMENT METHOD FOR NUMERICAL WELL TEST ANALYSIS SPE 62916
before sensing the closed reservoir boundary. The nonuni- using GEM without singularity programming are shown in
formly gridded cases performed worse than the uniformly Figures 13 and 14. The analytical versions of these curves
gridded cases in their reproduction of the pressure deriva- are presented for comparison.
tive. The infinite-acting radial flow period did not appear. Figures 13 and 14 show that when singularity program-
Instead of remaining level during the time corresponding to ming is not used the match to the pressure is not as precise
infinite-acting radial flow the derivative maintains a steady as when singularity is applied (compare to Example 4, Fig-
downward slope. ure 12). The early time behavior of the derivative curve
The effects of the errors in the pressure derivative curve is matched poorly. Figure 15 illustrates how this poorly
in this suite of finite difference simulations were quantified matched derivative curve could be interpreted if it were
by treating each one as data in a well test analysis. The treated as true data in a classical well test analysis. The
interpretation was performed using a standard regression match was performed on the derivative, and gave rise to
procedure. The estimates for skin and permeability from the estimates shown in Table 5.
the regression are highly correlated so the skin was set to The values in Table 5 are clearly erroneous, they demon-
zero in each case to ensure the treatment of the permeability strate the artificial wellbore storage effect that occurs in this
was consistent. The initial pressure was also fixed at its method as the well drains the grid block it is located in. Sin-
true value of 2000psi. The wellbore storage coefficient was gularity programming is successful in avoiding this artifical
set to zero. Regression was used to determine the reservoir effect.
permeability and the location of the boundaries. The results
are shown in Table 3.
Recall that the true permeability is 150md and the true Conclusions
boundary location is 1500ft. The uniformly gridded cases The simulated well test examples show that the combina-
all predict permeability values of 154-161md, which is due tion of GEM and singularity programming can reproduce
to the regression trying to fit the early time derivative data pressure transients accurately in comparison to analytical
which appear to show wellbore storage. If the fit is per- models. The proposed method was able to reproduce the
formed manually ignoring this data the true permeability pressure derivative curve much more accurately than con-
of 150md is predicted. ventional finite difference simulation. The use of singular-
ity programming was shown to be a key component of the
Example 4: Reservoir with Constant Pressure proposed method. When singularity programming was re-
Boundaries A well test was simulated in a reservoir with placed by the use of a Peaceman15 well index the accuracy
the properties given in Table 1. The well was located in of the method suffered significantly. The Peaceman well
the center of the reservoir and constant pressure boundary index is not strictly appropriate for modeling pressure tran-
conditions were applied on all sides. Finite difference sim- sient tests because it was derived for single-phase, steady-
ulation was used to simulate the well test. The constant state, radial flow in a homogeneous reservoir. Sato13 showed
presure boundary was simulated by putting injection wells that singularity programming was a useful tool when ap-
in every cell along the boundary. The resulting well test plied in conjunction with the perturbation boundary ele-
response is shown in Figure 11. The effect of the constant ment method. The well test modeled in Example 5 when
pressure boundary was not well matched on the derivative singularity programming was not used was also simulated
curve. The same well test was simulated using GEM com- using finite difference simulation. Neither the GEM scheme
bined with singularity programming. The transient is shown nor the finite difference scheme produced an accurate deriva-
in Figure 12. This simulation captures the effect of the con- tive match, however the GEM simulation accurately repro-
stant pressure boundary accurately also. duced the effect of the reservoir boundaries on the derivative
curve.
Examples 5 and 6: Well Test Simulation without Example 3 shows that the effect of a closed boundary on
Singularity Programming Two wells tests were modeled the pressure derivative could not be reproduced accurately
to study the performance of the GEM simulation scheme by finite difference simulation. Using nonuniform gridding
without singularity programming. The set of reservoir prop- exacerbated this problem and caused the infinite-acting ra-
erties given in Table 4 were common to both models. The dial flow period to be misrepresented in the pressure deriva-
boundary conditions applied were constant pressure bound- tive curve. GEM in conjunction with singularity program-
ary conditions. ming produced accurate matches to every pressure transient
To assess how the strength of the singularity affects the considered without being hampered by numerical artifacts.
computed results two cases were considered. In the first case
(Example 5), k = 100md and qw = 100ST B/d and in the
second case (Example 6) k = 150md and qw = 50ST B/d. Acknowledgments
The singularity in Example 5 is three times stronger than The authors would like to thank the SUPRI-D research Con-
in Example 6. The pressure and pressure derivative curves sortium on Innovation in Reservoir Testing for its financial
computed when Example 5 and Example 6 were simulated support.
SPE 62916 ROSALIND A. ARCHER and ROLAND N. HORNE 5
1061-68
2. Masukawa, J. and Horne, R. N.: Application of the Table 1: Examples 1 to 4 - Reservoir properties
Boundary Integral Method to Immiscible Displacement Prob-
lems, SPE Reservoir Engineering, August 1988, 1069-77 k 150 md
3. Kikani, J. and Horne, R. N.: Application of Boundary 0.3
Element Method to Reservoir Engineering Problems, Jour- 1 cp
nal of Petroleum Science and Engineering, 1989, 229-241 pi 2000 psi
4. Kikani, J. and Horne, R. N.: Pressure-Transient Anal- qw 50 STB/d
ysis of Arbitrarily Shaped Reservoirs With the Boundary ct 10e-06 psi1
Element Method, SPE Formation Evaluation, 1992, 53-60 rw 0.3 f t
5. Sato, K. and Horne, R. N.: Perturbation Boundary h 50 f t
Element Method for Heterogeneous Reservoirs: Part 1 -
Steady-State Flow Problems, SPE Formation Evaluation,
1993, 306-314
6. Sato, K. and Horne, R. N.: Perturbation Bound-
ary Element Method for Heterogeneous Reservoirs: Part
2 - Transient Flow Problems, SPE Formation Evaluation,
1993, 315-322
7. Taigbenu, A. E.: A More Efficient Implementation of
the Boundary Element Theory, Proc. 5th International
Conference on Boundary Element Technology (BETECH
90), Newark, Delaware, 1990, 355-366
6 THE GREEN ELEMENT METHOD FOR NUMERICAL WELL TEST ANALYSIS SPE 62916
102
psi
1
21 by 21 142.85 Uniform N/A
41 by 41 73.17 Uniform N/A
GEM, p
101 by 101 29.70 Uniform N/A GEM, dp
Exact, p
37 by 37 10 to 223.83 Nonuniform 1.2 10-1
Exact, dp
10-2
10-2 10-1 1 10 102 103 104
time, hours
ference simulations
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours
Table 4: Examples 4 and 5 - Reservoir properties
Figure 2: Example 2 - Finite difference simulation of well
0.3 test in closed reservoir
1cp 102
pi 2000 psi
ct 10e-06
aa
rw 0.3ft aa
aaaa b
aaaaaaa
10
aaaaaaaaaaaa b
h 50ft aaaaaa
a a a a a a a aa b
b
a b
b
Length 3000ft b
delta p, pressure derivative
b
b
b
Width 3000ft 1 b
b
bb
b
b b b b b b b bbb bbbbbbbbbbbbbbbbb
b
10-1
a
Table 5: Example 5 - Regression matched parameters b
GEM, p
GEM, dp
-2 Exact, p
10
Exact, dp
102
10
a b
10-1
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours
10
Figure 7: Example 3 - Pressure derivative for uniform 101
by 101 grid case
delta p, pressure derivative
10-1
Finite Difference, p
Finite Difference, dp
10-2 Exact, p
Exact, dp
102
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours
10
10-1
delta p, pressure derivative
1
Finite Difference, p
Finite Difference, dp
10-2 Exact, p
Exact, dp
10-1
10-3
Finite Difference, p 10-3 10-2 10-1 1 10 102 103
Finite Difference, dp Time, hours
10-2 Exact, p
Exact, dp
Figure 8: Example 3 - Pressure derivative for nonuniform
grid case 1
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours
102 10
aaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaa a aa
a a aaaa
a a a aaaaaaa a a aaaaaaaa a a aaaaaaaa
10
1
b b b bbbbbbbb bbb
bb bbbbbbb b bbbbbbbbbbbbbbbbbb bb
delta p, pressure derivative
10-3 10-3 b b bb
10-3 10-2 10-1 1 10 102 103 10-3 10-2 10-1 1 10 102 103
Time, hours Time, hours
Figure 9: Example 3 - Pressure derivative for nonuniform Figure 11: Example 4 - Finite difference simulation of well
grid case 2 test in a reservoir with constant pressure boundaries
102 10
aa a a aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaa a a aaaaa
aaa a a aaaa
a a a aaaa
10
delta p, pressure derivative
1
b bbbbbbbb b b bbbbbbb b b bbbbbbb b b
b bb
b b
b
b
b
b
10-1 b
b
10-1 b
Finite Difference, p a GEM, p b
Finite Difference, dp b GEM, dp b
10-2 Exact, p Exact, p
Exact, dp Exact, dp b
b
b
10-3 10-2 bbbbbbbbbbbbbbbbbbbbbbbbbbbb
10-3 10-2 10-1 1 10 102 103 10-3 10-2 10-1 1 10 102 103
Time, hours Time, hours
Figure 10: Example 3 - Pressure derivative for nonuniform Figure 12: Example 4 - GEM simulation of well test in a
grid case 3 reservoir with constant pressure boundaries
SPE 62916 ROSALIND A. ARCHER and ROLAND N. HORNE 9
102
10
delta p, pressure derivative
10-1
GEM, p
GEM, dp
10-2 Exact, p
Exact, dp
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours
1
delta p, pressure derivative
10-1
10
GEM, p
10-2 GEM, dp
Regression Match, p
Regression Match, dp
1
10-3
delta p, pressure derivative
GEM, p
10-2 GEM, dp
Exact, p
Exact, dp
10-3
10-3 10-2 10-1 1 10 102 103
Time, hours