Spe 55225 Productivityindezgas

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Use of Analytical Solutions To Improve

Simulator Accuracy
J.K. Jasti, Mobil E&P Technical Center, V.R. Penmatcha,* SPE, Stanford U., and D.K. Babu, SPE,
Mobil E&P Technical Center

Summary ing arbitrary trajectories. A sequence of analytical and numerical


Analytical solutions previously developed for wellbore design and procedures is outlined here to better estimate the relationship be-
scoping analysis were utilized in this study to benchmark and tween the wellbore and wellblock pressures.
modify the existing wellbore models in the reservoir simulator.
The analytical solutions were developed for single phase oil or gas Methodology
flow problems, and they are currently being used for well perfor- The rate-pressure relationship can be summarized using a
mance studies covering a wide range of well configurations: ver- wellbore-to-gridblock connection factor F ci as
tical, horizontal, deviated, and arbitrary well geometry, and frac-
tured completions in single or multilayered reservoirs. In this q i 5F ci~ p block,i 2p w f ,i ! for wellblock i. ~1!
paper, some of the analytical solutions are utilized for benchmark- Explicit formulas for the connection factors for homogenous me-
ing the existing well models in the reservoir simulator. Alternative dia with uniform grids can be found in previous publications by
well models are proposed for improving the accuracy of simulator Peaceman1,2 for vertical wells, and Babu3 et al. for horizontal
predictions. Two methods of calculating gridblock-wellbore- wells. Whereas the previous formulas covered essentially two-
connection factors are described. Results show that the new meth- dimensional flow, the present work is applicable to fully three-
ods impact the simulator accuracy significantly for three- dimensional problems, and is designed to handle arbitrary well
dimensional flow cases and for highly deviated wells. trajectories in three-dimensional space.
Two methods are presented for computing the connection fac-
tors. The first method applies when a uniform grid is employed in
the simulator. In this approach, the connection factors are deter-
Introduction mined from direct analytical solutions. For locally refined grids,
Flowing bottomhole pressure and flow rate are the two important and for arbitrarily structured grids, a second method is developed.
parameters computed by reservoir simulators. It is essential that Both methods are applicable for homogenous and anisotropic me-
these quantities are estimated accurately, especially in view of dia with either one well or multiple wells being active. The res-
increased application of complex well trajectories for exploiting ervoir region is assumed to be rectangular box shaped, with all six
oil and gas reservoirs. A relationship between wellblock pressure faces being closed to flow. The connection factors are determined
and wellbore pressure is utilized in the reservoir simulators to in a two step procedure: in the first step an analytical single phase
determine the wellbore pressure if the flow rate is specified, or to solution is used to determine production rates q i and wellbore
determine the wellblock flow rate, if the wellbore pressure is pressures p wf,i ; the second step involves computation of the
given. Accurate evaluation of ‘‘connection factors,’’ defining the p block,i .
relationship between the simulator-generated wellblock pressure
and the wellbore flowing pressure, therefore becomes an impor- Step 1: Determination of Flow Rates and Wellbore
tant item. Peaceman, through his landmark publications,1,2 estab- Pressures. All expressions are derived as solutions to the general
lished a mathematical relationship between wellblock pressure diffusion equation
and wellbore pressure for a fully penetrating vertical well. Babu3
et al. extended this work for the case of a fully penetrating hori- ]2p ]2p ]2p
kx 1k y 1k z
zontal well in a slablike drainage area. This topic continues to ]x2 ]y2 ]z2
generate interest and attention by various investigators:4–11 Kim’s
work on nonsquare gridblocks; the work of Morita et al. on fine ]p
5 fm c t , where p represents pressure ~psi!. ~2!
mesh and finite element three-dimensional ~3D! simulators; ]t
Peaceman’s work on off-centered and multiple wells in a single
For purposes of completeness, we indicate briefly a procedure to
gridblock; Lin’s work on partially penetrating vertical wells, het-
solve Eq. ~2!. The methods used are standard, and combine ana-
erogeneous media, and nonuniform grids; Ding’s work on double
lytical and numerical techniques to obtain the solution, as indi-
layer potentials with transmissibility adjustments; Sharpe and
cated in Refs. 12–17.
Ramesh’s publication on orthogonal grid generation, nonuniform
The pressure change at location (x i ,y i ,z i ) due to unit rate of
grids and 3D flow aspects; the work of Chen et al. on productivity
production at location (x j ,y j ,z j ) at time t is represented by
index calculation in reservoir simulators, etc.
C i j (t). The production at location (x j ,y j ,z j ) could be due to any
In this paper, we present results from our attempts to increase
of the following types of completions: a line sink ~representing
simulator accuracy when solving problems connected with the
horizontal, vertical, or inclined wells!, or a planar surface sink
production from a system of wells. The wells can be partially
~representing hydraulic fractures!.
penetrating ~allowing for 3D flow around the wellbore!, and hav-
Thus, C i j 5p initial2p ~~ x i ,y i ,z i ! ; ~ x j ,y j ,z j ! ;t ! , ~3a!
*Now with Mobil E&P Technical Center. where p((x i ,y i ,z i );(x j ,y j ,z j );t) denotes the pressure being ob-
Copyright © 1999 Society of Petroleum Engineers served at time t, location (x i ,y i ,z i ) due to unit rate of production
This paper (SPE 55225) was revised for publication from paper SPE 38887, first presented at location (x j ,y j ,z j ).
at the 1997 SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, By using the principle of superposition in space, the pressure
5–8 October. Original manuscript received for review 21 October 1997. Revised manu-
script received 23 November 1998. Revised manuscript approved 11 December 1998. drop observed at location (x i ,y i ,z i ) due to simultaneous produc-

SPE Journal 4 ~1!, March 1999 1086-055X/99/4~1!/47/10/$3.5010.15 47


tion from several completions or wells located at (x j ,y j ,z j ), each We choose equal time steps (Dt), and separating the contributions
producing at time-independent rates q j , j51,2,3,...,n w , can be at time level t k , we rewrite Eq. ~9! to obtain
expressed as
nw
D p i 5 @ p initial2p @~ x i ,y i ,z i ! ; ~ x j ,y j ,z j ! , j51,2,3,...,n w ;t ## D p o~ t k ! 5 ( q ~ t !@ C
j51
j
k
i j~ t
k
2t k21 ! 2C i j ~ t k 2t k !#
nw

5 (C i j~ t !q j , ~3b!
k21 nw
j51 1 ( ( q ~t
m51 j51
j
m
!@ C i j ~ t k 2t m21 ! 2C i j ~ t k 2t m !#
where n w is the total number of sinks ~wells or completions!.
For the case of variable ~transient! production rates from each for i51,2,...,n w ; k51,2,3, . . . . ~10!
completion, q j (t), the pressure drop at any location (x i ,y i ,z i ) can
be obtained by applying Duhamel’s theorem Noting that C i j (0)50, C i j (t k 2t k21 )5C i j (Dt), and D p o
5p initial2p(t k ), Eq. ~10! can be written as

E( F G
nw
t dC i j ~ t2 t !
D p i5 q j~ t ! 2 dt, i51,2,...,n w . ~4! nw
0 j51 dt
(qC
j51
k
j i j ~ Dt ! 1p
k

Equation ~4! can be solved when the total production rate is speci-
k21 nw
fied, or when the wellbore pressure is specified as a function of
time. The solution for the former case is described below. This 5p initial2 ( ( q ~t
m51 j51
j
m
!
total rate can be constant, or a known function of time t:
nw 3 @ C i j ~~ k2m11 ! Dt ! 2C i j ~~ k2m ! Dt !#
q t5 ( q ~ t !.
j51
j ~5! for i51,2,...,n w , and k51,2,3, . . . ~11!

The unknowns are: Dp j (t), and q j (t), j51,2,...,n w ; resulting in a In matrix notation, Eqs. ~11! and ~8! can be written as

4F G F G
total of (2n w ) unknowns.

3
C 11 C 12 ... C 1n w 1
The total number of unknowns in Eqs. ~4! and ~5! is (2n w ), q k1 r k21
1
whereas the number of equations is only (n w 11). An additional C 21 C 22 ... C 2n w 1 k21
(n w 21) relationships are required to achieve the closure of the q k2 2
system of equations. If frictional losses are neglected within and ... ... ... ...
... 5 ...
between wells, then the following (n w 21) relationships can be ... ... ... ...
q nk w r nk21
used for closure of the system: C nw1 C n w 2 ... C nwnw
w
1
Pk q kt
D p 1 ~ t ! 5Dp 2 ~ t ! 5...Dp nw~ t ! 5Dp 0 ~ t ! 5p initial2p ~ t ! , say. 1 1... 1 0
~6!
Equation ~6! represents the case wherein all wells are producing at for time levels k51,2,3, . . . , ~12!
the same flowing bottomhole pressure p(t). Thus the set of equa-
where r k21
i represents the terms on the right-hand side of Eq. ~11!.
tions can be rewritten as
We have described the basic theoretical framework in which

E( F G
nw
t dC i j ~ t2 t ! the general well performance problem can be tackled. A variety of
D p o5 q j~ t ! 2 dt for i51,2,...,n w , ~7! well geometries, well groups, fractured and unfractured wells,
0 j51 dt
single and multilayered formations, etc., can be handled by con-
nw structing the appropriate coefficient matrix C i j (t). The coeffi-
q t5 ( q ~ t !,
j51
j ~8! cients are derived by integrating the Green function solution along
the surface of the production elements: wells are approximated as
where the unknowns are Dp o ;q 1 ,q 2 ,...,qn w . Equation ~7! is dis- line sources, fractures are approximated as planar sources. The
cretized at time level k as expressions for the coefficients are not included here for the sake
of brevity and they can be found in the literature.16–20 However,
k nw
as an illustration, a procedure for the construction of C i j ’s for a
Dp o ~ t k ! 5 ( ( q ~t
m51 j51
j
m
!@ C i j ~ t k 2t m21 ! 2C i j ~ t k 2t m ! , horizontal well segment is presented in Appendix A.
By setting (n w ) as the number of gridblocks containing well-
i51,2,...,n w . ~9! bores, we can solve the system of Eqs. ~12! to determine the

TABLE 1– EFFECT OF PARTIAL PENETRATION ON THE PRODUCTIVITY INDEX ERROR BETWEEN THE TWO WELL MODELS

Well Length/ Drawdown


Reservoir Pressure
Thickness p wf (psi) p wf (psi) p̄ 2 p wf % J Error
Ratio from 2D from 3D Well from 3D Model Between the Two
Case (l/h) Well Model Model (psi) Well Models

1a 1 3628.0 3627.89 319.7 0.03


1b 0.5 3400.1 3440.94 506.7 28.1
1c 0.2 2718.3 2930.39 1017.2 220.85
1d 0.1 1584.5 2216.45 1731.2 236.5

48 Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999
common wellbore pressure p wf , and individual block production
TABLE 2– FLOW RATE DISTRIBUTION ALONG THE WELL AS
rates q i at any given time. However, we still need to determine the
CALCULATED BY THE TWO WELL MODELS „REPRESENTS
p block,i to find the wellbore connection factors.
CASE 1b AS SHOWN IN TABLE 1…
Note: The system of Eqs. ~12! needs to be solved speedily.
Much of the computational effort in solving Eqs. ~12! is spent in Flow Rate Calculated Flow Rate Calculated
generating the coefficient matrix C i j . Algorithms are available for Grid by the 2D Model (STB/D) by the 3D Model (STB/D)
generating these coefficients extremely rapidly, requiring negli-
(5,5,1) 998 944.6
gible amounts of time on PC machines. (5,5,2) 999 950.8
(5,5,3) 1000 965.9
Step 2: Determination of Wellblock Pressures. Two methods (5,5,4) 1001 999.5
are proposed for determining the gridblock pressures. The first (5,5,5) 1002 1139.1
method uses an analytical solution that is applicable to the uni-
form grid system with a central finite difference scheme. The
second method applies to general grid systems—uniform and non-
uniform. ~13a! needs to be modified to accommodate such geometries. In
such cases, analytical solutions to the modified finite difference
Method 1. For the case of a uniform grid system (Dx,Dy,Dz) schemes cannot be derived. Therefore simulators are used directly
using a central difference scheme, the original partial differential to determine the wellblock pressure p i jk . Once again, the analyti-
equation can be written in the discretized form as ~large time, cal solution is used to determine the p wf ,q i . The connection fac-
post-pseudo state behavior; subscript k represents here vertical tors are then determined using Eq. ~1!.
coordinate!

S D S D
Results and Discussion. Several simulations were performed to
kx ky
~ p i21,j,k 22p i, j,k 1p i11,j,k ! 1 ~ p i, j21,k 22 p i, j,k quantify the differences in simulator predictions generated by us-
Dx 2 Dy 2 ing the existing models, and those resulting from use of the new

S D
revised well models. The existing well models are based on the
kz
1p i, j11,k ! 1 ~ p i, j,k21 22p i, j,k 1p i, j,k11 ! earlier work of Peaceman1,2 for vertical wells, and Babu3 and
Dz 2 co-workers for the horizontal wells. Since the major difference

S D
between the earlier well models and the new model introduced in
887B 0 m 0
5 ~ N x N y N z •q i jk 2q t ! this paper is the ability to handle 3D flow, and complex wells,
abh comparisons are made only for these two cases. For each case, the
simulator was run until pseudosteady state was achieved, and the
for i50,1,...,~ N x 21 ! ; j50,1,...,~ N y 21 ! ;
wellbore pressure ( p wf), gridblock flow rate q i , and average res-
k50,1,...,~ N z 21 ! ; ~13a! ervoir pressure for the drainage area p̄ were recorded at a given
time. The discrepancy between the earlier well model and the new
the grid dimensions are (N x ,N y ,N z ) so that model was quantified using the productivity index J. J is defined
a[N x Dx; b[N y Dy; h[N z Dz. ~13b! as the ratio of total well flow rate to the drawdown pressure. The
new model described in this paper is identified as the 3D model,
Here q t is the total production rate, and q i jk is the production from whereas the existing well models in the simulator are labeled as
the gridblock ~i,j,k!. Obviously, q i jk 50 for all gridblocks wherein the two-dimensional ~2D! model.
no active wells are present. It is possible to derive an exact solu-
tion to the system Eq. ~13a!. In this sense, the simulator can be 3D Flow Effects. Production from a partially penetrating vertical
bypassed and yet the solutions that would have been gener- well was simulated to quantify the differences in J between the
ated by the simulator can be obtained. The above finite differ- two well models. All cases are simulated under single phase flow
ence equations are solved by adopting the same techniques as conditions with no gravity effects. The common parameters used
described in the Appendices A and B ~SPE No. 23525! to Babu are: Bo51 RB/STB; oil viscosity51 cp; porosity50.2; initial res-
et al.’s paper on wellblock radius.3 However, here the methodol- ervoir pressure54000 psi; total compressibility53.0e-05/psi.
ogy is generalized with the formulas relevent to handling 3D Drainage area dimensions are: 6000 ft39000 ft3200 ft. The well
problems. was located at x52700 ft, y54050 ft; permeability in the three
The solution to the system of finite difference equations ~FDE! principal directions was: kx5ky5kz5100 md. Total well pro-
@Eqs. ~13!# is written in terms of the average reservoir pressure p̄, duction rate was set at 5000 STB/D; Uniform grid of 10310
and the block pressure p i jk for any gridblock as 310 was utilized. The extent of well penetration ~well length/
reservoir thickness! was varied by anchoring the well at the top of
~ 2 p !~ 141.2! B 0 m 0 the reservoir. The results are presented in Table 1. As expected,
p i jk [p̄2
abh when the well is fully penetrating, the discrepency between the
two models is negligible, since the flow is two-dimensional. How-
3 ~ S i 1S j 1S k 1S i j 1S jk 1S ki 1S i jk ! . ~14!
ever, as the well is shortened relative to the reservoir thickness,
Explicit formulas for the S i ..S i jk are given in Appendix B. By the error in J increases to as high as 36.5% at 10% penetration.
setting the (i, j,k)[(i w , j w ,k w ) in Eq. ~14!, the wellblock pres- Results also show that when a well is completed in more than
sure p i jk can be determined. The procedures for determining the one gridblock, discrepancy between the two well models was ob-
connection factors are: solve Eq. ~12! to find p wf ,q i ; substitute served in flow distribution along the well. Production profiles
q i 5q i jk values in Eq. ~14! to determine p i jk ; finally use Eq. ~1! to computed by using the two well models for case 1b ~as shown in
find the connection factors (F ci). ~Index i in Eqs. ~1! and ~12! is Table 1! are presented in Table 2. Results clearly show that the
identical to the block index (i jk) in Eqs. ~13! and ~14!!. simulation done using the 2D well model did not adequately re-
solve the 3D flow near the wellbore.
Method 2. If the grids used by the simulator are nonuniform or Three-dimensional flow around the wellbore can be more ac-
irregular, as in the case of local grid refinement schemes, Eq. curately captured by grid refinement. A limited number of simu-

Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999 49
TABLE 3– THE EFFECT OF GRID REFINEMENT ON THE J ERROR BETWEEN THE TWO MODELS

p wf (psi) % J Error Simulation


from 2D P wf (psi) from 3D Between the Two Time
Case Nx Ny Nz Well Model Well Model Well Models (hr)

1d 10 10 10 1584.5 2216.5 236.5 0.16


1e 10 10 20 1584.6 2216.5 236.5 0.2
1f 30 30 20 1904.5 2216.5 218.0 0.7
1g 30 90 20 2013.1 2216.5 211.7 2.0

lations was carried out to quantify the effect of grid refinement on have higher productivity than vertical wells since high angle wells
the simulations with the 2D well model and the results are pre- can have longer contact through the pay zone. Fig. 1 shows the
sented in Table 3. The case with well penetration of 10% ~case 1d effect of the angle of deviation on wellbore productivity. In order
in Table 1! was used as the base case over which additional re- to separate the length effects from the angle of deviation effects,
finements are made. Since the base case had high error, due to the well length was fixed at 150 ft in Fig. 1. The pseudosteady
predominantly 3D flow near the wellbore, intuitively it may be state productivity index was normalized by the vertical well pro-
desirable to put more grids in the vertical direction. Results in ductivity index. Fig. 1 shows that for isotropic reservoirs, the
Table 2 show that increasing the number of grids in the vertical angle of deviation has no effect on well productivity. As the ver-
direction had no impact on accuracy. This is mainly due to the tical permeability is reduced, well productivity decreases with the
isotropic permeability. The length scales as @li/sqrt~ki!# and, due angle of deviation. Productivity is the lowest for horizontal wells
to the large aspect ratio, more grids are needed in the lateral since their performance is most susceptible to vertical permeabil-
direction to capture the 3D effects and to minimize the error in J. ity. At the other extreme, for a vertical well, the vertical perme-
Table 3 shows that error indeed decreases with increased grid ability does not effect the productivity unless the penetration ratio
resolution but at an expense of computational time. This is not a is small.
problem for simulations with the 3D model, since it can still be Fig. 2 shows the results for the fully penetrating well case. The
used only with a 10310310 grid system. well length increases as inverse cosine of the angle of deviation
Slant Well Model Results. Highly deviated wells or wells with except when the well is horizontal, at which orientation the well
principal direction not parallel to any of the grids are generally length was arbitrarily set at 2000 ft. The productivity results pre-
approximated as staircase of horizontal or vertical segments or a sented in Fig. 2 represent the combined effects of wellbore devia-
combination of the horizontal and vertical well components. tion and changing well length. Results clearly show that perme-
Among these methods, staircase of horizontal well segments is the ability anisotropy plays a dominant role. Errors resulting in
most popular one. The validity of such ad hoc well modeling is staircase approximation are evaluated using the analytical model.
examined using analytical models. Several runs were performed The error is quantified as the discreprency in J between the slant
using the following parameters: well and the staircased well. Figs. 3 and 4 show the errors for
drainage area dimensions: isotropic and anisotropic drainage areas. Results show that, as
3000 ft33000 ft3150 ft; expected, the error in staircase approximation is negligible if the
permeability: kx5ky5kz5100 md; permeability is isotropic. On the other hand, large errors in J
and kx5ky5100 md with kz510 md; could result by using the staircase approximation for anisotropic
Bo51.2 RB/STB; viscosity51.0 cp, porosity50.2; media. The error does not rapidly diminish as the number of stair-
total compressibility51.0e-05 psi; case segments is increased. Conversely, we can conclude that
production rate: 1000 STB/D. fairly large numbers of vertical gridblocks are needed to accom-
Slant wells or highly deviated wells are increasingly coming modate large number of staircase segments and to minimize the
into use worldwide. It is generally perceived that high angle wells error in staircase approximation.

Fig. 1–Effect of angle of deviation on productivity ratio, constant well length.

50 Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999
Fig. 2–Effect of angle of deviation on productivity of a slant well, fully penetrating well.

Fig. 3–Productivity index error in staircase approximation for isotropic reservoir.

Fig. 4–Productivity index error in staircase approximation for anisotropic reservoir.

Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999 51
Conclusions t 5 time, days
1. A procedure for the accurate calculation of pressure and rate t k 5 time, at level k(t k [k.Dt), days
distribution along the wellbore, in single phase flow problems, is Dt 5 uniform time step, for computing transient val-
outlined in this work. ues, days
2. Numerical simulators often use rectangular grids to compute (x i ,y i ,z i ) 5 general observation point coordinates at which
the wellblock pressures and wellblock rates. For the special case pressures are measured, ~ft, ft, ft!
of a homogeneous reservoir with uniform rectangular grids, it is x i ,y j ,z j 5 general point coordinate at which production
possible to solve analytically the finite difference system of equa- takes place, ~ft, ft, ft!
tions. Analytical formulas for the general wellblock pressures, f 5 porosity
p i jk , are derived in this paper. m 5 oil viscosity, cp
3. Utilizing analytical solutions to single phase oil flow prob-
lems, this paper identifies errors in computing p wf and q by simu- Subscripts
lators that routinely employ 2D flow based formulas. Illustrative
examples indicate that errors increase as the partial penetration i 5 block index
ratio decreases, confirming that existing 2D formulas in simula- (i, j,k) 5 block index
tors are in need of modification when 3D flows dominate. i, j,k,l,m,n 5 summation indices
4. This work suggests alternative approaches to rectify such 0<i,l,l 8 <L21;0< j,m<M 21;0<k,n<N21
errors. The suggested techniques attempt to modify the way the v 5 vertical
connection factors (F ci) are hardwired into simulators. h 5 horizontal
5. Staircase approximations of deviated wells are shown to
lead to errors. Errors are seen to increase with increasing perme- Acknowledgments
ability anisotropy. Local grid refinement does not necessarily im- The authors thank Mobil management for permission to publish
prove accuracy; global refinement does reduce errors, but at a this material. Helpful comments from Mobil colleagues Ilhan
price of increased computational time. Sener, Myung Hwang, Pramod Bansal, ~late! Jack Harper, Hong
Chan, Marty Cohen, Usuf Middya, Bob Jones, and Peter Dzurman
Nomenclature are gratefully acknowledged.
a 5 length of reservoir, ft
A lmn 5 general coefficient in the finite Fourier expansion References
of the variable f i jk , associated with the node 1. Peaceman, D.W.: ‘‘Interpretaton of Wellblock Pressure in Numerical
Reservior Simulation,’’ SPE J. 18, 183 ~June 1978! Trans., AIME
~i,j,k!
265.
b 5 width of reservoir, ft 2. Peaceman, D.W.: ‘‘Interpretation of Wellblock Pressures in Numeri-
B o 5 formation volume factor for oil, rb/stb cal Reservior Simulation With Nonsquare Grid Blocks and Aniso-
c t 5 total compressibility of formation, 1/psi tropic Permeability,’’ SPE J. 23, 531 ~June 1983!.
C i j 5 matrix element: pressure drop observed at loca- 3. Babu, D.K., Odeh, A.S., Al-Khalifa, A.J., and McCann R.C.: ‘‘The
tion i due to unit production at location j, psi/ Relationship Between Wellblock and Wellbore Pressures in Numeri-
bbld cal Simulation of Horizontal Wells,’’ SPE Reservoir Eng. 6, 324
F ci 5 connection factor for block i, bbld/psi ~1991!.
h 5 thickness of reservoir, ft 4. Kim, D.: ‘‘Well Correction Factor for Three-Dimensional Reservoir
(i w , j w ,k w ) 5 nodal coordinates of a typical wellblock Simulation with Non-Square Grid Blocks, and Anistropic Permeabil-
ity,’’ MS thesis, Texas A&M U., College Station, Texas ~May 1988!.
J 5 productivity index, @ q/(p̄2p wf) # , bbld/psi
5. Morita, N. et al.: ‘‘Three-Dimensional Well Model Preprocessors for
(k x ,k y ,k z ) 5 permeability in three spatial directions, md Reservoir Simulation With Horizontal and Curved Inclined Wells,’’
l 5 well length, ft paper SPE 20718, Presented at the 1990 SPE Annual Technical Con-
L 5 gridblocks in I direction ference and Exhibition, New Orleans, 23–26 September.
M 5 gridblocks in j direction 6. Peaceman, D.W.: ‘‘Interpretation of Wellblock Pressures in Numeri-
n w 5 total number of wells, completions, etc. cal Reservoir Simulation: Part 3—Off-Center and Multiple Wells
N 5 gridblocks in k direction Within a Wellblock,’’ SPE Reservoir Eng. 5, 227 ~1990!; Trans.,
(N x ,N y ,N z ) 5 grid dimensions; number of gridblocks in each AIME 289.
direction 7. Lin, C.Y.: ‘‘New Well Models for Partially Penetrating Wells in Het-
p 5 pressure, psi erogeneous Reservoirs Using Non-Uniform Grids,’’ paper SPE
29122, Presented at the 1995 SPE Symposium on Reservoir Simula-
p̄ 5 average reservoir pressure, psi
tion, San Antonio, Texas, 12–15 February.
p initial 5 initial pressure, psi 8. Ding, Y.: ‘‘A Generalized 3D Well Model for Reservoir Simulation,’’
p block,i 5 block pressures, psi SPE J. 1, 437 ~December 1996!.
p wf 5 wellbore pressure, psi 9. Peaceman, D.W.: ‘‘Representation of a Horizontal Well in Numerical
Dp i 5 pressure drop observed at location i, psi Reservoir Simulation,’’ SPE Advanced Technology Series 7 ~April
Dp o 5 common pressure drop at the wellbore, in the 1993!.
absence of frictional losses, psi 10. Sharpe, H.N. and Ramesh, B.A.: ‘‘Development and Validation of a
p i jk 5 pressure measured at block ~i,j,k!, psi Modified Well Model Equation for Nonuniform Grids With Applica-
q i 5 production rate at location i, psi tion to Horizontal Well and Coning Problems,’’ paper SPE 24896,
production from wellblock i, psi Presented at the 1992 SPE Annual Technical Conference and Exhibi-
tion, Washington, DC, 4–7 October.
q t 5 total production from all completions, wells,
11. Chen, G., Tehrani, D.H., and Peden, J.M.: ‘‘Calculation of Well Pro-
bbld ductivity in a Reservoir Simulator,’’ paper SPE 29932, Presented at
r ki 5 right-hand side terms in the matrix equation, at the 1995 SPE International Meeting on Petroleum Engineering,
location ~block! i, and time level k, psi Beijing, 11–14 November.
S i ,S i j ,S i jk 5 partial trigonometric sums: single, double, and 12. Muskat, M.: ‘‘The Flow of Homogeneous Fluids through Porous Me-
triple sums dia,’’ I.H.R.D.C., Boston ~1982! 209.

52 Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999
13. Gringarten, A.C. and Ramey, H.J. Jr.: ‘‘An Approximate Infinite a lmn cos~ p lx i /a ! cos~ p lx j /a !
Conductivity Solution for a Partially Penetrating Line-Source Well,’’
SPE J. 15, 140 ~April 1975!; Trans., AIME 259.
Pi j[ (
l,m,nÞ0 b lmn

S D S D S D
14. Gringarten, A.C., Ramey, H.J. Jr., and Raghavan, R.: ‘‘Unsteady-
p my i p nz i p nz j
State Pressure Distributions Created by a Well With a Single Infinite- 3cos cos cos
Conductivity Vertical Fracture,’’ SPE J. 14, 347 ~August 1974!. b h h
15. Rosa, A.J. and Carvalho, R.S.: ‘‘A Mathematical Model for Pressure
Evaluation in an Infinite Conductivity Horizontal Well,’’ SPE Form. ~ sin~ p my 2 /b ! 2sin~ p my 1 /b !!
3 ~A-4!
Eval. 4, 559 ~December 1989!, Trans., AIME 257. p m/b
16. Gringarten, A.C. and Ramey, H.J.: ‘‘The Use of Source and Green’s
and the decaying transient terms T i j are given by
Functions in Solving Unsteady-Flow Problems in Reservoirs,’’ SPE
J. 13, 285 ~October 1973!; Trans., AIME 255.
17. Ozkan: ‘‘Performance of Horizontal Wells,’’ PhD dissertation, U. of Ti j[ (
lmnÞ0
Tulsa, Tulsa, Oklahoma ~1988!.
18. Babu, D.K. and Odeh, A.S.: ‘‘Productivity of a Horizontal Well,’’
a lmn cos~ p lx i /a ! ¯ ~ sin~ p my 2 /b ! 2sin~ p my 1 /b !!
SPE Reservoir Eng. 4, 417 ~1989!. 3 E lmn ,
19. Babu, D.K. and Odeh, A.S.: ‘‘Appendices A and B of SPE 18298: b lmn ~ p m/b !
Productivity of a Horizontal Well,’’ paper SPE 18334 ~November ~A-5!
1989!.
20. Carslaw, H.S. and Jaeger, J.C.: Conduction of Heat in Solids, second where
edition, Clarendon Press, Oxford, U.K. ~1959!, p. 361.
E lmn [exp@ 2 b lmn ~ t2 t ! / a # , ~A-6a!

Appendix A: Construction of Cij „ t … for a Horizontal


b lmn [ p 2 S k xl 2
a2
1
k ym2
b2
1
k zn 2
h2
D ~ l,m,n ! 5 ~ 0,1,2,3,...! ,
Well Segment ~A-6b!
We outline briefly a procedure for constructing the matrix ele- a lmn 51 for single series,
ments C i j (t) mentioned in the text. A single horizontal well seg-
ment, running parallel to the Y axis, and with the wellbore axis 51/2 for double series, ~A-6c!
defined by x5x j ; y 1 <y j <y 2 ; z5z j , is used here for illustrative 51/4 for triple series.
purposes. For arbitrarily slanted wells, and for fracture surfaces,
etc., the integration ranges would be modified appropriately. We Note: it is understood that when m50 in the series, the term
start with a continuous point source solution ~Green’s function @ sin(pmy2/b)2sin(pmy1/b)#/(pm/b) is absent from the series.
format!, and derive formulas for the pressure drop D p at any
arbitrary point located at (x i ,y i ,z i ) in the reservoir. The well Computational Aspects
segment, located at (x j ,y j ,z j ), is producing at the rate q j bbl/d. Transient terms T i j are computed directly as they are: evaluation
Thus, at time t ~days! of single, double and triple series to a prescribed accuracy. Rapid
exponential decay of the terms assures fast convergence.
D p i j ~ t ! [ ~ p initial2p i j ! Time-independent components Pij present computational chal-

S DE E
lenges. However, as indicated in the publication by Babu and
887B m q j t y2
Odeh,19 double and triple series can be reduced to rapidly con-
[C i j ~ t ! q j [ S x S y S z dy o d t , ~A-1!
abh a 0 y1 verging single and double series, along with evaluation of certain
simple analytical expressions. In this way, the construction of the
where matrix element @ C i j (t) # becomes relatively easy and straightfor-
ward.
S x [S ~ x i ,x j ;a;k x ; ~ t2 t ! ; a !

F
For a Fracture Element. Integrals in Eqs. ~A-1! and ~A-2! are

S D S D
`
p nx i p nx j
(
extended to two dimensions along the plane of the fracture. Ob-
[ 112 cos cos viously, the computational difficulties are less severe for these
n51 a a

DG
cases.

3exp S 2 p 2 n 2 k x ~ t2 t !
aa2
. ~A-2! For a Point Source/Sink Problem. The spatial integration with
respect to y o in Eq. ~A-1! is eliminated. The resulting series are
again handled by algebraic reduction techniques used in Ref. 19.
Similar expressions are used to define S y and S z :
Appendix B: Analytical Formulas for Gridblock
S y [S ~ y i ,y o ;b;k y ; ~ t2 t ! ; a ! , Pressures—General Exact Solution of FDE
~A-2a! The analytical solution to Eq. ~13! is derived along the same lines
S z [S ~ z i ,z j ;h;k z ; ~ t2 t ! ; a ! . as the 2D solutions presented in the paper of Babu3 et al. We
begin by listing a few standard identities ~m, M are integers!
Here, the dimensional constant a is given by a 5158 fm c t , days.

S D
j5i
After multiplying out, and integrating the resultant infinite series pm 1 sin~ p m ~ i11 ! /M !
sums, we obtain (j50
cos
M
j1 5
2 2 sin~ p m/2M !
. ~B-1!

C i j~ t ![ S 887B m
abh DF t
a
1 P i j 2T i j . G ~A-3!
Using product-sum formulas, and Eq. ~B-1!, we get, for 0<l,l 8
<(M 21),

S D
M 21
p l ~ i11/2! p l 8 ~ i11/2! M
Here the time-independent, pseudosteady state components P i j
are given by
(
i50
cos
M
cos
M
5
2
d ~ l,l 8 ! , ~B-2!

Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999 53
where the kronecker delta d (l,l 8 )51, l5l 8 Þ0: Simplifying Eq. ~B-10!, and writing out all the terms, we have

S D S D S D
50, lÞl 8 , pl 1 pm 1 pn 1
52, l5l 8 50.
~B-2a! (
lmnÞ0
A lmn cos
L
i1
2
cos
M
j1
2
cos
N
k1
2
Next, consider a general relationship
L21,M 21,N21
p l ~ i11/2!
3 F S DS
Dx
kx
2 cos
pl
L
21 1 D S DS ky
Dy 2 cos
pm
M
21 D
f i jk 5 (
S DS DG
A lmn cos
l,m,n50 L kz pn
1 cos 21 [ ~ F i jk /2 ! . ~B-12!
p m ~ j11/2! p n ~ k11/2! Dz 2 N
3cos cos
M N
We rewrite Eq. ~B-12!, and solve it by applying Eqs. ~B-4!–~B-6!.
for i50,1,...,~ L21 ! ; j50,1,...,~ M 21 ! ; The solution to Eq. ~B-12! is therefore given by

k50,1,...,~ N21 ! .
By applying Eq. ~B-2! to Eq. ~B-3! in all the indices ~l,m,n! and
~B-3!
A lmn [ S 887B 0 m 0
abh DS D ( S D
22
l lmn i, j,k50
q i jk cos
pl
L
i1
1
2
~i,j,k! it can be shown that

S D
LM N
A lmn 5 f i jk cos(
pl
i1
1
S D 3cos
pm
M
1
2 S D S D
j1 cos
pn
N
1
k1 .
2
~B-13!
8 i jk50 L 2
Substituting Eq. ~B-13! into Eq. ~B-8!, we obtain the exact solu-
3cos
pm
M
1
j1 cos
2 NS D S D
pn 1
k1 ,
2
~B-4!
tion that a numerical simulator would generate for large times
after pseudosteady state is attained. Thus

S D
LM N
4
A lm0 5
i jk50
f i jk cos(
pl
L
1
i1 cos
2
pm
M
1
j1 ,
2 S D S D p i jk [p̄2
~ 2 !~ 2 p !~ 141.2! B 0 m 0
abh
~B-5!
pl pn pn
S D
LM N
A l005 f i jk cos
pl
(
1
i1 . S D ~B-6!
3 (
cos
L
~ i11/2! cos
M
~ j11/2! cos
N
~ k11/2!

S D
2 L 2
i jk50
l,m,nÞ0 l lmn
Similar expressions and formulas hold for A lon , A oon ; and A omn , L21,M 21,N21
A omo , etc.
Finally, we also have (
i 8 j 8 k 8 50
q i8 j 8k8

S D
LM N
A 0005 f i jk . ( ~B-7!
3 cos
pl
L
1
S D
i 8 1 cos
2
pm
M
j 81
1
2 S D , ~B-14!

S D
1 i jk50
pn 1
We next assume a solution to Eq. ~13a! in the form cos k 81
N 2

S D
L21,M 21,N21
pl 1
p i jk [p̄1 (
l,m,nÞ0
A lmn cos
L
i1
2 where l lmn [
kx
sin2
pl
1
ky
sin2
pm
1
kz
sin2
pn
.
Dx 2
2L Dy 2
2M Dz 2
2N

3cos
pm
M
1
j1 cos
2 S D
pn
N
1
k1 ,
2 S D ~B-8!
Note: We have to attach factors ~1/2! and ~1/4! in the above sum-
~B-15!

where the volumetric average pressure p̄ is defined as mations when dealing with double sums and single sums.
In summary, the solution to the FDE can be written down as
~ LM N !~ p̄ ! [ (
i jk50
p i jk . ~B-9!
~ 2 p !~ 141.2! B 0 m 0
p i jk [p̄2
Substituting Eq. ~B-8! into Eq. ~13a! of the main text, we obtain abh
the following: 3 ~ S i 1S j 1S k 1S i j 1S jk 1S ki 1S i jk ! . ~B-16!

( F S DS S D S D
A lmn
Dx
kx
2 cos
pl
L
i1
3
2
1cos
pl
L
i2
1
2
We set here N x [L, N y [M , N z [N, and obtain

S D(
lmnÞ0
N x 21
cos~ p l ~ i1 21 ! /N x !

S DD S D S D
1
pl 1 pm 1 pn 1 S i[
22 cos i1 cos j1 cos k1 2 l51 l loo

S( S DD
L 2 M 2 N 2

S S D S DD G
pl 1
ky kz 3 q i w j w k w cos i 1 , ~B-17!
1 similar terms with , 5F i jk , ~B-10! iw j wkw Nx w 2
Dy 2
Dz 2

S D(
N y 21
1 cos~ p m ~ j1 21 ! /N y !
where we have written, in place of the right-hand side of Eq. S j[
2 l omo
~13a!: m51

F i jk [ S 887B 0 m 0
abh D
~ LM N•q i jk 2q t ! . ~B-11! 3 S( iw j wkw
q i w j w k w cos
pm
j 1
Ny w 2 S DD
1
, ~B-18!

54 Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999
S D( pn
N z 21
1 cos~ p n ~ k1 21 ! /N z !
cos~ p m ~ j1 2 ! /N y ! cos
1
S k[ ~ k11/2!
Nz
2 n51 l lmn S jk [ ( l omn

S( S DD
m,n51
pn

S( S D S DD
1
3 q i w j w k w cos k 1 , ~B-19! pm 1 pn 1
iw j wkw Nz w 2 3 q i w j w k w cos j w 1 cos k w1 ,
iw j wkw Ny 2 Nz 2
~B-21!
cos~ p l ~ i1 2 ! /N x ! cos~ p m ~ j1 2 ! /N y !
1 1

Si j[ ( l lmo cos~ p l ~ i1 21 ! /N x ! cos~ p n ~ k1 21 ! /N z !


l,m51
S ki [ ( l lon

S( S D S DD
l,n51
pl 1 pn 1
3
iw j wkw
q i w j w k w cos i 1 cos
Nx w 2
j 1
Ny w 2
,
3 S( iw j wkw
q i w j w k w cos
pl
S D
1
i 1 cos
Nx w 2
pn
k 1
Nz w 2
1
S DD ,

~B-20! ~B-22!

cos~ p l ~ i1 21 ! /N x ! cos~ p m ~ j1 21 ! /N y ! cos~ p n ~ k1 21 ! /N z !


S i jk [2 (
l,m,n51 l lmn

3 S(iw j wkw
q i w j w k w cos
pl
S D
1
i 1 cos
Nx w 2
pm 1
j 1 cos
Ny w 2 S D
pn
k 1
Nz w 2
1
S DD . ~B-23!

Here, l lmn is defined in Eqs. ~B-15!. The indices (i w , j w ,k w ) de- Dp w ~ x w ,y w ,z w !


note wellblocks in which active wells are located.

;
q
E s5L ds
, ~C-3!
Appendix C: Location of the Observation Point on
Wellbore Surface „see Fig. C1…
We generalize the results and formulas given in Refs. 18 and 19
2L s52L
A ~ x w 2x s ! 2
kx
1
~ y w 2y s ! 2
ky
1
~ z w 2z s ! 2
kz
for the location of the point on the well perimeter at which pres-
sure is calculated. This pressure will be identified as the flowing where (q/2L) represents production per unit length of well axis,
bottomhole pressure p wf . We start with a point sink solution in and ds is a small element on well axis. Integrating Eq. ~C-3! and
three dimensional space. By considering the well to be a finite line retaining only the dominant quantities, we get
sink, and integrating along the well axis, we arrive at appropriate
formulas for the well pressure p wf .
We assume that the center of well axis (x c ,y c ,z c ) is known. D p w ;ln@ r 2w M ~ u , f !# , ~C-4!
Polar coordinate representation of a general point Q(x s ,y s ,z s ) on
the well axis is given by
where
x s 5x c 1s sin u A cos f A ,
y s 5y c 1s sin u A sin f A ,
z s 5z c 1s cos u A ,
~C-1!
M ~ u , f ! 5A sin2 u F S cos2 f A sin2 f A
kx
1
ky
1
cos2 u
kz D G
where, s measures the distance of Q from well center (x c ,y c ,z c ).
The well axis is represented by polar coordinates ( u A , f A ), where
u A is the angle between vertical Z axis and the well axis, and f A
F S
2 ~ sin u sin u A !
cos f cos f A sin f sin f A
kx
1
ky D
is the angle between X axis and the well axis projection onto the
XY plane.
1 S DG
cos u cos u A
kz
2
, ~C-5!
An arbitrary observation point (x w ,y w ,z w ) is selected on the
well perimeter in a normal cross section through the center
(x c ,y c ,z c ). We assume that the observation point coordinates are where
given by ~u,f!, so that
x w 5x c 1r w sin u cos f , ~ sin u A cos f A ! 2 ~ sin u A sin f A ! 2 ~ cos u A ! 2
A[ 1 1 . ~C-6!
kx ky kz
y w 5y c 1r w sin u sin f , ~C-2!
z w 5z c 1r w cos u .
In addition to the above relations, we have the orthogonality
From potential theory, we know that the pressure drop experi- property of ‘‘well axis’’ and ‘‘radius vector to the observation
enced at the point P(x w ,y w ,z w ) due to a line sink is given by the point.’’ Thus, vectors (sin u cos f,sin u sin f,cos u) and
dominant component, in anisotropic medium, according to the fol- (sin uA cos fA ,sin uA sin fA ,cos uA) are orthogonal to each other.
lowing rule: Hence, taking their dot product, we get

Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999 55
M ~ u ! 5A S sin2 u cos2 u
kh
1
kv D
2cos2 u cos2 u A
1
2
1
kv kh S D 2
.
~C-10!
From geometric considerations ~see Fig. C1…, we can show that
u has the range ( p /22 u A )< u <( p /21 u A ).
In Eq. ~C-10!, the Min.Max values (M 1 ,M 2 ) occur at ( p /2
6 u A ) and p/2. By solving the equation M ( u )5 AM 1 M 2 , and
after some algebraic simplifications, we obtain the coordinates of
the observation point ~u,f!:

cosec2 u A
tan2 u 5cot2 u A 1 ~C-11a!
Ak v ~ sin2 u A /k h 1cos2 u A /k v !
Fig. C1–Location of observation point on an inclined well.
and

sec u A
tan~ f 2 f A ! 5 4 . ~C-11b!
cos~ f 2 f A ! 1cot u cot u A 50. ~C-7! Ak v ~ sin2 u A /k h 1cos2 u A /k v !
Procedure. In order to locate the points at which the maximum Jay K. Jasti is an Engineering Advisor with Mobil E&P Tech-
and minimum pressure values occur on the well bore, we elimi- nical Center. His primary interests are in the area of Well
Performance. Jasti holds a PhD in chemical engineering
nate one variable, say f, from Eqs. ~C-5! and ~C-7!, and then
from the U. of Michigan. V.R. Penmatcha is a senior engi-
equate to zero the derivative of M with respect to u. Let the
neer with Mobil E&P Technical Center in Dallas. He holds a
maximum and minimum values of M be given by M 1 and M 2 . Btech degree in chemical engineering from Andhra U., In-
Then, the observation point u at which the average value of Dp dia, and a MS degree from the U. of Tulsa and a PhD de-
occurs is given by gree from Stanford U., both in petroleum engineering. He

S D
was 1995–96 President of the Stanford U. Student Chapter.
Dp 1 1Dp 2 1
Dp5 ; @ ln M 1 1ln M 2 # D. Krishna Babu is an Associate Engineering Advisor in the
2 2 Reservoir Engineering Analysis Group at Mobil E&P Techni-
cal Center in Dallas. His interests are in the application of
5ln AM 1 M 2 5ln~ M ~ u !! ~C-8! quasianalytical and numerical techniques to problems in
so that u is determined by solving M ( u )5 AM 1 M 2 . porous media, with special reference to horizontal wells. He
previously was a research engineer in the Civil Engineering
Example. For simplicity, we consider k x 5k y 5k h ; k z 5k v . For Dept. at Princeton U. and on the faculty of City College of
New York. Babu holds a MA degree in applied mathemat-
this case,
ics from Osmania U., India, and a PhD degree in applied

A[ S sin2 u A cos2 u A
kh
1
kv D ~C-9!
analysis from the State U. of New York at Stony Brook. He
serves on the Editorial Review Committee and was a
1991–93 member of the Reservoir Simulation Symposium
and Program Committee.

56 Jasti, Penmatcha, and Babu: Improve Simulator Accuracy SPE Journal, Vol. 4, No. 1, March 1999

You might also like