Anda di halaman 1dari 40

Modeling Conventional and

Non-Conventional Wells
Christian Wolfsteiner,
1
Khalid Aziz
1
and Louis J. Durlofsky
1,2
1
Stanford University,
2
Chevron Petroleum Technology Company
Sixth International Forum on Reservoir Simulation

Hof/Salzburg, Austria, September 3-7, 2001.


Abstract
This paper presents state-of-the-art techniques for predicting the performance of wells.
Our main focus is on non-conventional wells (NCWs), which include horizontal and
multilateral wells and wells with downhole control devices. Techniques used in nite
difference simulators as well as more recently developed semi-analytical methods are
presented. The advantages and disadvantages of each approach are discussed.
Limitations of conventional approaches for modeling NCWs are discussed and recent
research on the development of well models is presented. A procedure is presented for
calculating the well index WI for any type of well in a nite difference simulator. Several
examples are presented demonstrating the application of semi-analytical techniques for
quick productivity calculations and for the estimation of WI.
Introduction
Because of the importance of well models in reservoir simulators, this topic has been ex-
tensively discussed in the petroleum literature and new modeling techniques are continu-
ally appearing. Our objective is to improve the ability of simulators and simpler analytical
models to predict well performance. We are particularly interested in non-conventional
wells (NCWs).
Performance of wells in oil and gas reservoirs can be predicted through a variety of tools:
inow performance relations or IPRs (Vogel, 1968; Bendakhlia and Aziz, 1989), con-
ing correlations (de Souza et al., 1998), analytical solutions (Kuchuk et al., 1988) and
reservoir simulators (Nolan, 1990). While reservoir simulators can handle the greatest
complexity, other tools have certain advantages in specic situations. As a matter of fact,
the representation of wells within a simulator requires a well model (analytical or nu-
merical) to relate the well ow rate to the difference in the block pressure and the well
pressure. Analytical models provide exact solutions to problems that have been suf-
ciently simplied to be amenable to such techniques. Typically analytical solutions are
possible for single-phase ow in homogeneous reservoirs of simple shape and with sim-
ple boundary conditions. On the other hand, reservoir simulators can be used to obtain
approximate solutions for far more complex problems. For example, reservoir simulators
can be used for multiphase, multi-component ow in heterogeneous reservoirs of irregu-
lar shape. IPRs can be obtained directly from eld data or by numerical experiments with

Also presented at the IEA Symposium on Enhanced Oil Recovery, Vienna, Austria, September 9-12,
2001.
1
a simulator. They provide a quick and simple way to predict well performance and have
many applications in the petroleum industry. Many of the practical aspects of horizontal
well productivity are discussed by Joshi (1988, 1991).
Well modeling techniques were originally developed for vertical or near vertical wells
(i.e., conventional wells). The primary interest of the petroleum industry has now shifted
to horizontal or multilateral wells, here referred to as NCWs. Such wells are also termed
advanced wells in the petroleum industry (Holmes, 2001). Modern NCWs can have very
complex architectures with downhole control devices used to optimize well performance
during the life of the well. Wells with downhole sensors and control devices are often
called smart wells. The design and operation of NCWs is more strongly affected by near
well heterogeneities and frictional pressure drop in the well than is the case for conven-
tional wells. As a result, research on the modeling of NCWs has intensied over the past
ten years. A NCW has the potential to increase productivity by many fold over what is
possible by a vertical well in the same location. But NCWs are more expensive to drill
and complete. Economic deployment of NCWs requires reliable predictive techniques.
Aziz et al. (1999) have discussed many of the problems associated with the prediction
of horizontal well performance. For more advanced wells these problems become even
more serious.
In this paper we will present an overview of the current state of techniques for modeling
wells, particularly NCWs. Our emphasis will be on semi-analytical solution methods
and the coupling of these techniques with nite difference (or nite volume) reservoir
simulators. Most of the examples and recent results in this paper are from work done in
the Petroleum Engineering Department at Stanford University. While high quality work
is also being conducted elsewhere, it is not accessible to the authors as easily as their
own work. However, references to a wide variety of other important published work are
provided.
Modeling with Simulators
Cartesian or logically Cartesian grids are commonly used for reservoir simulation. A
well simulated with such a grid may intersect several grid blocks. While the well terms
in reservoir simulation ow equations for a grid block appear as component sources or
sinks for that grid block, the well is controlled by either specifying the ow rate of one
or more phases (at reservoir or some other conditions, i.e., separator) or pressure at some
location (wellhead or bottomhole). The well constraint equation is based on the specied
conditions and has to be solved for each well, resulting in additional unknowns (well
pressure or well rate) in the system. In discretized form the ow equation for a typical
block can be written as:

p
m
c,p
i,l

p
m
w
c,p
i
=
1
t

p
_
M
n+1
c,p
M
n
c,p
_
i
(1)
where m designates mass ow rate, c = 1 . . . n
c
is the component index, and p = 1 . . . n
p
is the phase index. The summation l is over all connections to other gridblocks from
gridblock i. For the multiphase ow case, the mass rate of ow across the interface
2
between blocks i and l takes the form:
m
c,p
i,l
=
_

p
y
c,p
Ak
x
k
r,p
x
_
i,l
_
(p
p
l
p
p
i
)
p
i,l
g
g
c
(H
l
H
i
)
_
. (2)
The last term in the above equation accounts for inow due to gravity. The equations are
written in consistent units with and as conversion factors. The accumulation term can
be expressed as:
M
n
c,p
i
=
_
V
_
S
p
y
c,p
_
n
_
i
, (3)
and the well term can be expressed as:
m
w
c,p
i
=
_
WI y
c,p

p
k
r,p

p
(p
p
p
w
)
_
i
= T
w
c,p
i
(p
p
p
w
)
i
. (4)
It is clear from the last equation that the key to modeling wells is the determination of the
correct WI for each block i the well is passing through.
It is possible to increase the accuracy of predictions related to the well by conforming the
grid to the well. The earliest attempts to do this involved cylindrical, local grid renement
(Aziz and Settari, 1979) and hybrid grids (Pedrosa and Aziz, 1986), and Voronoi grids for
horizontal and vertical wells (Palagi and Aziz, 1994).
The well index accounts for the geometry of the gridblock, location and orientation of the
well segment in the gridblock and rock properties. It is generally based on the assumption
of single phase, single component ow around the well (Peaceman, 1978), where the
reservoir ow equation simplies to:

l
T
i,l
(
l

i
)

w
m
w
i
= c
t
1
t
_

n+1
i

n
i
_
. (5)
The gravity term has been included with pressure in the denition of . The well equation
becomes:
m
w
i
= WI

(p
i
p
w
i
) = T
w
i
(p
i
p
w
i ). (6)
The single phase equations can be written in differential form by taking appropriate space
and time limits of Eqs. 5 and 6:
(k) = c
t

t
+ q
w
, (7)
where q
w
is the source term per unit volume.
The discrete single phase well equation given in Eq. 5 shows that the segment well index
or transmissibility can be calculated provided that, for a given ow rate, the correct well-
block and well pressures are known. In many cases the single-phase assumption makes it
possible to calculate the exact well pressure analytically or semi-analytically. The grid-
block pressure can then be computed via a single phase ow simulation by specifying the
ow rates in all wellblocks. From the exact well pressures and the numerically computed
wellblock pressures, accurate well indices can be determined. A detailed procedure for
accomplishing this will be described later in this paper.
3
(Peaceman, 1978, 1983, 1990) has provided expressions for WI for vertical and horizon-
tal wells. An expression for wells in non-square gridblocks and anisotropic permeability
in his 1983 paper has become the standard default (indicated by superscript d) procedure
in virtually all commercial simulators:
WI
d
i
=
_
_
2
_
k
x
k
y
h
ln
r
w
r
o
_
_
i
, (8)
with
r
0
= 0.28
_
_
k
y
k
x
x
2
+
_
k
x
k
y
y
2
4
_
k
y
k
x
+
4
_
k
x
k
y
. (9)
The above equation works well for vertical wells. With appropriate changes to account for
well orientation Eq. 9 can also be used to model horizontal wells, provided assumptions
made by Peaceman (1983) are satised. However, often these assumptions are violated
for horizontal and other NCWs. Other techniques have been presented in the literature for
calculating the well index for conventional wells, but their usefulness is limited. In a later
section we will discuss well index computations for NCWs.
In the simulation of most vertical or near vertical wells the frictional pressure drop is
ignored and the hydrostatic component is computed by assuming no holdup of uids in
the wellbore. In horizontal or near horizontal wells the pressure drop in the wellbore may
inuence inow into the well when the drawdown is of the same order of magnitude as the
pressure drop in the well. Novy (1995) suggests that when the wellbore pressure drop is
less than 10% of the drawdown it can be ignored. The wellbore pressure drop is generally
only important in long, high production rate wells in very high permeability reservoirs.
A well must be divided into segments when it is necessary to model uid ow in the well.
A segmented well model makes it possible to account for pressure drop and uid holdup
in the well. Also, a segmented well model is required for handling downhole devices like
control valves, chokes, separators, pumps, sensors, and multiple laterals. Fig. 1 shows
a multilateral well with control devices in a three dimensional grid and corresponding
segments.
In well models the ow is assumed to be one dimensional. The variables for segments
forming the well grid are added to the reservoir variables. The full system of equations
can be solved by a variety of techniques (Byer et al., 1999). These calculations result
in the prediction of inow/outow for each segment, pressure drop over the segment and
composition of the uids in each segment. In addition, in situ fraction of each phase and
the ow pattern can also be predicted. A detailed discussion of segmented well models
is available in papers by Almehaideb et al. (1989), Nolan (1990), Coats et al. (1995) and
Holmes (2001).
While the impact of wellbore hydraulics on productivity may be signicant in some
cases, near well heterogeneities often dominate well performance (Ding, 1995; Wolf-
steiner et al., 2000a). For practical reasons, it is generally impossible to make the grid
around the well sufciently ne to directly represent small scale heterogeneities. Thus
the problem of upscaling in the near well region arises. This problem has been addressed
by several researchers. We review some approaches briey here, though techniques for
near well upscaling will not be considered in detail in this paper (we will however present
4
Figure 1: Multilateral well control devices (Holmes, 2001).
approaches for capturing the effects of near well heterogeneity within a semi-analytical
context). For a more complete discussion of near well upscaling within a reservoir simu-
lation context, see Wolfsteiner (2002).
Ding (1995) pointed out that conventional upscaling techniques based on linear ow were
inappropriate for radial near well ow. In addition to coarse grid well indices, Ding also
computed upscaled well block transmissibilities to match the ne grid uxes. Soeriaw-
inata et al. (1997) presented an analytical method for obtaining effective permeabilities
of well blocks using an incomplete layer concept. Durlofsky et al. (2000) and Mug-
geridge et al. (1999) extended Dings work and conned the upscaling region to a local-
ized volume around the well of interest with generic boundary conditions. Mascarenhas
and Durlofsky (2000) presented an upscaling technique for horizontal wells featuring an
optimization approach that forces agreement of ne and coarse grid well block uxes by
simultaneously adjusting well block indices and transmissibilities of blocks adjacent to
the well.
5
Semi-Analytical Well Models
As we saw earlier, even when a well is modeled with simulators the calculation of the well
index WI requires the exact pressure at the well for a model single-phase ow problem
that includes essential well and near well reservoir features. These computations are
generally performed on one or more representative wells with suitable assumptions to
make the calculations possible. Furthermore, in many situations it is desirable to have
simple formulae to quickly predict well productivity without detailed simulations. In
recent years semi-analytical solutions have been developed that greatly expand the use of
such techniques over traditional analytical techniques.
In this section we give an overview of basic analytical forms for horizontal wells with
complex trajectories. We will also discuss a new method that is capable of approximately
accounting for reservoir heterogeneity, and an approach for modeling downhole inow
control devices. Common to all analytical and semi-analytical well modeling approaches
are the simplifying assumptions about well and reservoir geometries and properties. The
assumptions of the uid being single phase and the reservoir having homogeneous pa-
rameters (with the exception of models for layered reservoirs (Lee and Milliken, 1993;
Basquet et al., 1998)) applies generally to all models discussed in this section.
Analytical Solutions for Horizontal Wells Early models were developed using analo-
gies to vertical wells and were focused on well test applications. These analytical solu-
tions form the basis for handling more complicated geometries.
Goode and Thambynayagam (1987) modeled the transient of a horizontal well in a semi-
innite reservoir. They used successive integral transforms (Fourier and Laplace). The
well was modeled as a strip source with uniform ux boundary conditions. The solution
in Laplace space was inverted numerically. Daviau et al. (1988) demonstrated an ana-
lytical modeling approach for horizontal wells in either a semi-innite or nite reservoir
with constant pressure and/or no ow boundary conditions. Gravitational effects were ne-
glected and the well was a line source parallel to the boundaries. They used Greens func-
tions (Gringarten and Ramey, Jr., 1973) and the method of images (Carslaw and Jaeger,
1959). Clonts and Ramey, Jr. (1986) and Ozkan et al. (1989) obtained transient solutions
for a horizontal well in an anisotropic, semi-innite reservoir using Greens functions.
Kuchuk et al. (1988) used Laplace transforms for the same problem and modeled a gas
cap and/or water aquifer with constant pressure boundaries. Babu and Odeh (1989) were
the rst to solve the problem of a horizontal well parallel to one of the coordinate direc-
tions, arbitrarily located in a box shaped reservoir with no-ow boundaries. They used
separation of variables and Fourier series. They also developed a simple approximate
form for pseudo steady-state ow by using a shape factor, a skin for partial penetration,
and an aspect ratio to account for anisotropy.
These important investigations have shown that the productivity of horizontal wells is
much more sensitive to placement within the reservoir (well length, partial penetration)
and reservoir anisotropy. All these solutions were obtained by ignoring gravity and pres-
sure drop in the well.
Dikken (1990) presented the rst model that coupled the reservoir to wellbore ow. For
a single phase uid he assumed all laminar or all turbulent ow in the pipe and constant
inow (productivity) along the wellbore. Landman (1994) reformulated Dikkens model
6
and expressed the analytical solution as an integral that was solved numerically. More
important, he allowed for varying productivity along the well by dividing the well into
segments. He also gave a result for the optimal perforation density given a production
strategy. Novy (1995) was interested in the circumstances under which pressure drop in
the wellbore becomes important. He assumed a cylindrical reservoir with a fully pen-
etrating horizontal well along its axis. Based on Dikkens work, Novy provided charts
that plot the maximum well length (at which friction becomes important, i.e., reduces the
inow by 10 %) over the total well inow.
Semi-analytical Forms for NCWs While the solutions for basic well congurations
have been long known, their use has been limited because of difculties in the evalua-
tion of these solutions. Hence much effort has been spent on simplifying and nding
asymptotic forms of such solutions. Improved computational capabilities made it possi-
ble to not only compute basic analytical solutions but to also apply simple paradigms such
as the principle of superposition to handle complex geometries. The technique that nu-
merically combines analytical results; i.e., evaluates and approximates innite sums, and
performs continuous integrals over space and time, is referred to as the semi-analytical ap-
proach. This expressions stems from the fact that, apart from numerical truncation errors,
these methods are approximations to the true analytical form and can be made arbitrarily
accurate by increasing the level of discretization along the well.
Penmatcha and Aziz (1998) presented a technique for ow around a horizontal well in
three dimensions. In their semi-analytical approach, the horizontal well was split into
segments. For each segment the formulation of Babu and Odeh (1989) was employed
and superposition in space and time was applied. The resulting system was solved using
the Newton-Raphson method. One new aspect of this work was the use of a general well
hydraulics equation that accounted for both friction and uid acceleration. Ozkan et al.
(1999) presented a semi-analytical model that coupled wellbore hydraulics and reservoir
ow for horizontal wells in an innite reservoir.
Economides et al. (1996) used the instantaneous point sink/source solution (IPSS) to
solve the transient ow problem of an arbitrary well trajectory in a closed box shaped
reservoir. The anisotropic reservoir was rst rescaled to its isotropic equivalent (Besson,
1990). Special attention must be paid to the interpretation of well solutions in transformed
space, as the well cross section becomes elliptical in the transformed space (Kuchuk and
Brigham, 1979; Brigham, 1990). An equivalent wellbore radius has to be used in such
cases. Ouyang et al. (1998) coupled the IPSS approach with a general single phase hy-
draulics model (Ouyang and Aziz, 1998) that accounted for frictional and accelerational
effects in the wellbore. They decoupled the reservoir and wellbore equations and applied
a successive substitution scheme. Boundary conditions for each reservoir face could now
be either constant pressure or no-ow. They recognized the importance of gravity and
obtained their solution in terms of a potential.
Besides reservoir anisotropy and wellbore hydraulics, which have been implemented into
these solutions successfully, other reservoir effects can have a signicant impact on the
performance of NCWs.
Lee and Milliken (1993) provided solutions for the productivity of a well of arbitrary
orientation in a three dimensional, layered anisotropic formation. Their aim was to im-
7
1
st
lateral
3
rd
lateral
2
nd
lateral
M (i
w
, 1,
4)
M (i
w
, 3, 5)
M (i
w
, 3, 1)
M (i
w
, 3,
2)
M (i
w
, 3, 3)
M (i
w
, 3,
4)
M (i
w
, 3, 6) M (i
w
, 3, 7)
) 6 ; 5 , 3 , (
i
p
w

) 1 , 3 , (
iw

M (i
w
, 1, 1)
M
(
i
w
, 1, 2)
M
(
i
w
, 1, 3)
) 4 ; 3 , 1 , (i L
w

q (i
w
)
Segment
(i
w
, 2, 2)
Figure 2: Arbitrary well conguration (Ouyang and Aziz, 1998).
plement this solution as a well model into a nite difference reservoir simulator. They
applied slender body theory and the method of reections to capture effects of a layered
medium (Muskat, 1937). Classical two dimensional well models (Peaceman, 1978) were
special cases of their approach. Basquet et al. (1998) used point sources distributed along
the well trajectory in an anisotropic, layered reservoir. They used Fourier and Laplace
transforms for space and time variables respectively.
Reservoir heterogeneity, particularly in the well vicinity, plays a key role in determin-
ing well performance. Recent research at Stanford University on the development of
semi-analytical techniques that approximately account for reservoir heterogeneities will
be discussed in the following sections.
Semi-Analytical Solution Technique Ouyang and Aziz (1998) have presented a semi-
analytical approach for the solution of the coupled wellbore-reservoir problem. They
model the reservoir as a parallelepiped with any combination of constant potential or no-
ow boundary conditions on the six bounding faces. Isothermal ow in the reservoir is
described by the single phase pressure equation for slightly compressible uids,
(k) = c
t

t
, (10)
with appropriate initial and boundary conditions. Equation 10 is formulated using poten-
tial = p + (g/g
c
)z rather than pressure p to account for gravity effects, which can be
signicant when the well is not strictly horizontal. The permeability k is assumed to be a
diagonal tensor. In the actual numerical solution, the problem is rescaled to the isotropic
equivalent.
Wells can have an arbitrary conguration and trajectory (see Fig. 2). Each well i
w
(of a
total of n
w
wells) can have n
L
(i
w
) laterals which in turn consist of n
S
(i
w
, i
L
) segments.
8
The total number of segments N
S
is then given by
N
S
=
n
w

i
w
=1
n
L
(i
w
)

i
L
=1
n
S
(i
w
, i
L
). (11)
The inner boundary condition is either constant rate or constant pressure for each individ-
ual well. Due to the nature of the Greens functions used, the constant pressure condition
requires integration in time by taking a series of constant rate steps. This basic method
does not account for skin around the well.
Method of Solution. The method of solution for this coupled reservoir-well problem is
referred to as instantaneous point source/sink solution or IPSS (Economides et al., 1996;
Maizeret, 1996; Ouyang and Aziz, 1998; Ouyang, 1998). Ouyang and Aziz (1998) solve
for the dimensionless potential drawdown
D
in any point of the bounded reservoir
given an arbitrarily oriented well sink/source segment S. The approach entails the use of
Greens functions, with superposition of image wells, application of Neumanns product
rule (Gringarten and Ramey, Jr., 1973), and integration over space and time. Free-space
Greens functions for instantaneous sinks/sources satisfying Eq. 10 in one dimension are
well known (Carslawand Jaeger, 1959). An innite series of image sinks/sources is super-
imposed to model a sink/source point between two boundaries, each with either constant
pressure or no-ow boundary conditions.
The coupled reservoir-wellbore system has 2 N
S
unknowns: the dimensionless in-
ow (or outow) q
ID
(i
w
, i
L
, i
S
) and potential drawdown
wD
(i
w
, i
L
, i
S
) at the midpoint
M(i
w
, i
L
, i
S
) of each segment (see Fig. 2). The method yields N
S
potential drawdown
equations, N
S
n
w
equations for wellbore hydraulics (of the form of Eq. 13) relating the
potential of a segment to that of its neighbors, and n
w
mass balance equations of the form
q
D
(i
w
) =
n
w

i
w
=1
n
L
(i
w
)

i
L
=1
q
ID
(i
w
, i
L
, i
S
), (12)
for a total of 2 N
S
equations.
The wellbore hydraulics relate the pressure of each segment to that of its neighbor, so
it is possible to reduce the number of unknown drawdowns to a single
wD
(i
w
, i
L
, 1)
per lateral by inserting Eq. 13 into the potential drawdown equations. If the starting
segments at the heel of each lateral are close together, we can assume
wD
(i
w
, 1, 1)

wD
(i
w
, i
L
, 1) and the number of potential drawdown unknowns is then reduced to n
w
.
Wellbore Hydraulics. Wellbore hydraulics were introduced into the general semi-
analytical approach in the original development of the method (Ouyang and Aziz,
1998; Ouyang, 1998). A considerable amount of work has been directed toward
establishing correlations for frictional pressure losses in horizontal or deviated wellbores;
much of this literature is reviewed in Ouyang and Aziz (1998). Here we will simply
present the frictional pressure loss correlations used in this work and outline the
basic implementation of wellbore hydraulics into the overall procedure. The wellbore
hydraulics equation relates the owing potential (pressure) of segment i to that of its
upstream neighbor, segment i + 1 (segment numbering increases as we move from the
9
heel to the toe of the lateral). Flow within the lateral is from the toe towards the heel, so

i+1
>
i
. In the case of innitely conductive laterals (no pressure losses within the
wellbore), all segments have the same potential; i.e.,
i
= constant. The basic equation
describing pressure losses in the wellbore can be expressed as follows:

i+1
=
i
+ p
f
+ p
a
, (13)
where p
f
represents frictional pressure losses and p
a
represents losses due to accel-
erational effects. The quantity p
a
is essentially zero when there is no inux into the
wellbore, but it can be signicant with inow.
The frictional pressure loss is expressed in dimensionless form via the friction factor f.
Many correlations exist for the friction factor for closed pipe (no inux); these require
some modication for the general case with inux. For the case of turbulent ow with
inux (producing well), Ouyang et al. (1998) presented the following correlation,
f = f
0
(1 0.0153Re
0.3978
w
), (14)
where f
0
is the friction factor in the absence of wall inux and Re
w
is the wall Reynolds
number (Reynolds number based on inux). The friction factor f
0
depends only on the
axial Reynolds number and the pipe roughness and is available fromstandard correlations.
Note that the frictional pressure losses during turbulent ow in wellbores with inux are
less than those in closed pipes. However, because of the accelerational pressure losses,
the total pressure drop is generally higher with inux. The wellbore hydraulics equation
introduces nonlinearity into the overall system of equations. For the cases considered
by Ouyang et al. (1998), however, the coupling between the hydraulics equation and the
reservoir ow equations was found to be rather weak. As discussed below, Newtons
method can be used when this approach does not converge.
Downhole Wellbore Control In addition to pressure losses due to frictional, accelera-
tional and hydrostatic effects, Valvatne et al. (2001) extended the model to include pres-
sure losses due to downhole inow control devices. The most common control device is a
choke. Because downhole chokes are relatively new inventions, specic correlations for
predicting pressure loss through these devices are generally proprietary and are not avail-
able in the literature. For this reason a simplied version of the Sachdeva et al. (1996)
model applicable to single-phase liquid ow was implemented by Valvatne et al. (2001).
Other more appropriate models could easily be used within the general semi-analytical
framework.
When downhole devices across which there can be very large pressure drops are included
in the formulation, the coupling between the reservoir and the downhole hydraulics be-
comes very tight; i.e., the successive substitution approach used by Ouyang and Aziz
(1998) can no longer be used. For such cases Valvatne et al. (2001) solved the full cou-
pled nonlinear system using Newtons method; i.e.,
J = R, (15)
where J is the Jacobian matrix, R the residual vector and is the update to the solution
vector. Elements of the Jacobian matrix are dened via J
ij
= R
i
/x
j
, where x represents
the vector of unknowns. The problem consists of potential residual equations R
p
, equal to
10
the total number of inow segments n
S
, and mass balance residual equations R
m
, equal
to the number of wells n
w
. The unknowns are segment inows q
I
and wellbore potentials

w
.
The potential equation R
p
(i
w
, i
S
) for a well i
w
and segment i
S
has the complexity that
wellbore potential is no longer constant along the well, due to friction and acceleration.
When multiple reservoirs are present in the problem, it is more convenient to use dimen-
sional residual equations. The dimensional and expanded forms of the equations are then
given by:
R
p
(i
w
, i
S
) =
n
w

j
w
=1
n
S
(j
w
)

j
S
=1
_
Bq
I
(j
w
, j
S
)
k

x
e

i
_

D
(j
w
, j
S
, i
w
, i
S
)

i
_
_

i

w
(i
w
)
k
S
=i
S

k
S
=1
(p
f
(i
w
, k
S
) + p
a
(i
w
, k
S
))
_
_
= 0,
R
m
(i
w
) =
n
S
(i
w
)

i
S
=1
q
I
(i
w
, i
S
) q (i
w
) = 0. (16)
Here,
w
(i
w
) is the wellbore potential for well i
w
, measured at the beginning of the rst
segment. It is convenient to keep
D
(j
w
, j
S
, i
w
, i
S
) in dimensionless form as this is com-
puted directly from the dimensionless Greens function. Valvatne et al. (2001) update the
Jacobian at each iteration. This is computationally inexpensive, however, as the great ma-
jority of the execution time is spent evaluating the potentials
D
(j
w
, j
S
, i
w
, i
S
), given by
the Greens functions. These are required only once for every timestep and do not need
to be updated when the Jacobian is recomputed. The inclusion of downhole devices has
only a slight effect on the execution time.
Incorporation of Skin in Solution The mechanical skin effect is generally used to ac-
count for altered permeability in the near wellbore region for an otherwise homogeneous
reservoir. This concept is commonly extended to account for non-Darcy ow effects,
partial completions (Thomas et al., 1996) and slanted wells. Wolfsteiner et al. (2000a)
introduced a skin that approximately accounts for the effects of near well heterogeneity
on wellbore ow.
Hawkins (1956) expression for skin s is:
s =
_
k
k
a
1
_
ln
r
a
r
w
, (17)
where r
a
is the radial distance over which permeability is altered fromk to k
a
(see Fig. 3).
We will generalize this skin concept to approximately model near well reservoir hetero-
geneity.
We now describe how skin is introduced into the semi-analytical formulation of the pre-
vious section. Wolfsteiner et al. (2000a) considered a lateral with four segments (Fig. 4)
where the fourth segment S
4
has a skin s
4
associated with it. The M
i
are the equivalent
pressure points located at a distance r
w
from the segment axis (in the isotropic case).
11
r
w
k
a
r
a r
k
Figure 3: Flow in series in a radial system (Wolfsteiner et al., 2000a).
M
1
!
M
4
!
M
3
!
M
2
!
S
1
4
3
2
skin
s
4
S
S
S
segment
pressure
point
Figure 4: Superposition of segment drawdowns with skin (Wolfsteiner
et al., 2000a).
A potential drawdown calculated at such a point is taken as the well drawdown at that
segment. As an example, the potential drawdown
wD
in S
1
given itself and all other
segments is, by superposition:

wD
(S
1
) =
4

i=1
q
ID
(S
i
)
D
(M
1
, S
i
), (18)
where
D
(M, S) is the Greens function from the IPSS solution and q
ID
(S) is the dimen-
sionless in/out ow rate of segment S. However, the drawdown in S
4
is changed due to
its skin,

wD
(S
4
) =
4

i=1
q
ID
(S
i
)
D
(M
4
, S
i
) +q
ID
(S
4
)s
4
. (19)
Note that Eq. 18 implicitly accounts for the skin s
4
because all of the segment drawdowns
are coupled through wellbore hydraulics. In the case of an innitely conductive well, the
pressure along the lateral is constant. In the Examples section below, we show a clear
validation of this skin implementation (cf. Fig. 5 below).
Using this procedure, numerical problems can appear if skin is very negative in some
segments of the well. This occurs, for example, in uvial reservoirs that display a high
degree of permeability variation over small distances. In these cases, the local skin can
become highly negative and the potential contribution from neighboring segments can
12
become dominant. This in turn creates instability in the solution. Valvatne et al. (2001)
found that a more stable approach is to use an effective wellbore radius r
we
, rather than
the actual radius r
w
, where r
we
= r
w
e
s
. The effective wellbore radius for each segment
will now vary depending on the magnitude of the local skin. In most cases, this method
ensures that the matrix is diagonally dominant. For the most difcult cases, a rescaling of
skin is necessary to ensure numerical stability. This rescaling reduces the magnitude of
the most negative s(i
w
, i
S
) to the value required for the stability of the numerical scheme.
Heterogeneity Representation with s and k* Wolfsteiner et al. (2000a,b) demonstrated
how reservoir heterogeneity can be included in an approximate sense in the semi-
analytical framework in terms of a background effective permeability k

and a local skin


along the wellbore. Additional skin effects, due to near-well damage or stimulation, could
also be included by summing up the contributions to give a total skin.
The skin can be computed using an expression of the form of Eq. 17. We need to estimate
the size and shape of the altered zone, the permeability k
a
in this altered zone, and
the scalar background permeability which will be denoted as k

s
. The scalar k

s
, to be
used in the skin equation in place of k, will be derived from the diagonal tensor k

,
as discussed below. The shape of the altered zone is taken to be an elliptical cylinder
with size and orientation determined from the statistical properties of the permeability
eld (Wolfsteiner et al., 2000a). The altered permeability k
a
in Eq. 17 is computed as a
weighted power average of the permeabilities in the search region. Following previous
studies in two (Durlofsky, 2000) and three dimensions (Desbarats, 1994), we represent
this average as follows:
k

a
=
1

a
_
a
k

(x)
r
n
dx, (20)
where designates the averaging exponent (power), r indicates radial distance, n is the
distance weighting, and the integral is evaluated over the search region a. The normal-
ization
a
is computed via
a
=
_
r
n
dx. We evaluate Eq. 20 through use of a Monte
Carlo type of integration (Press et al., 1993). The parameters and n, and even the size
and shape of the averaging region, could be tuned through some type of parameter t.
However, in our examples we use = 0 (corresponding to a logarithmic weighting) and
n = 2 (Wolfsteiner et al., 2000a).
The global (or background) permeability k

is simply computed as the effective perme-


ability for linear ow(Durlofsky, 2000). This can be computed most accurately by solving
the single phase incompressible pressure equation numerically on the geostatistical ne
grid in each coordinate direction. This single phase calculation is relatively inexpensive
and has to be done only once for each permeability realization. As an alternative, an ap-
proximative approach employing the statistical information of the permeability eld can
be used (Ababou, 1990). In three dimensions the diagonal entry of the tensor in coordinate
direction d = x, y, z is:
k

dd
= k
g
exp
_

2
lnk
2
_
1
2
3
l
h
l
d
__
. (21)
This formula was developed for log-normally distributed permeability elds which have a
geometric mean k
g
and correlation lengths l
d
(l
h
is the harmonic average of the three cor-
relation lengths). Note that this equation uses
2
lnk
, the variance of the log-permeability.
13
Table 1: Reservoir, uid and well properties (Wolfsteiner et al., 2000a).
drainage area 4500 4500 ft
2
thickness 175 ft
vert. bounds sealing (no ow)
horiz. bounds sealing (no ow)

porosity 0.25
compress. c
t
3.0 10
5
psi
1
at p
ini
density 60 lb
m
/ft
3
viscosity 1 cP
form. fac. B 1.05 RB/STB
initial potential
ini
3000 psi
total rate Q 10,000 STB/d (primary

)
min. bhp p
w
min
2800 psi (secondary

)
well radius r
w
2.4 in
report time 40 and 100 days

these specications differ for the xed pressure example


In the examples below we specify the log-normal permeability elds in terms of the mean
m and the coefcient of variation (standard deviation divided by mean)
v
rather than in
terms of
lnk
.
As indicated above, although our effective permeability (computed in either manner) is a
diagonal tensor, the skin formula is in terms of a scalar k. We therefore set k = k

s
=
3
_
k

xx
k

yy
k

zz
in Eq. 17 (i.e., k = k

s
is the geometric average of the k

permeability com-
ponents).
Our s-k

approach provides a varying skin distribution along the laterals of the wells.
Using this skin, the semi-analytical model computes a ow and pressure response (or
productivity) for each well segment. Comparisons between these results and those from
standard nite difference numerical simulations will be discussed in the next section.
Although the overall semi-analytical approach is only appropriate for single phase ow, it
is possible to use the technique to qualitatively represent some effects due to multiphase
ow (Yeten et al., 2000).
Examples The examples presented in this section demonstrate the capabilities of the
proposed method. Most of these results are compared with those from a nite difference
numerical simulator (Schlumberger, 1999). For the rst two cases, this reference solution
was computed using the detailed geostatistical model (45 45 35 cells) and the default
well model (Peaceman, 1983).
The time requirements for the semi-analytical solution method are essentially independent
of the level of detail in the geostatistical permeability description (though the numerical
calculation of k

will scale with the size of the ne grid model). Therefore, the efciency
advantage of the method over a nite difference technique will increase as the model size
14
Table 2: Statistics of the permeability elds (Wolfsteiner et al., 2000a).
realization permA permB / B2 permC
mean m 100 md 103 / 101 md 199 md
coefcient of
variation
v
1 1.8 / 1.6 1.4
corr. structure
{l
x
= l
y
, l
z
}
0.3, 0.3 0.3, 0.1 0.5, 0.05
background
perm. k

s
79 md 67 / 67 md 136 md
Table 3: Congurations of the examples presented.
0
-2
0
1
3
wellA wellA30 wellB
x
z
y
wellC wellD
increases. It is also important to note that the extra overhead of the s-k

method compared
to the base semi-analytical method is insignicant. In the calculations below, we consider
a slightly compressible single phase uid. In most cases, we use only straight well laterals,
which are aligned with the grid in the nite difference simulator. We do not include
wellbore hydraulics except when otherwise stated. Gravitational effects are included in
all of the calculations. Tables 1 to 3 provide the basic reservoir description, the statistics
of four different log-normally distributed permeability realizations (referred to as permX)
and schematics of the various well geometries (wellY ) respectively. The examples involve
different well geometries coupled with the different permeability realizations. In Table 2,
l
x
, l
y
, l
z
refer to dimensionless correlation lengths. Spherical variogram models with no
sill were used in all cases. Note that in Table 3 only the black well sections are perforated;
the gray lines indicate the location of the vertical main hole (the heel of lateral).
Anisotropic Reference Case. Our rst example serves to demonstrate that skin is im-
plemented correctly in the semi-analytical solution. The reservoir is homogeneous but
anisotropic with permeability prescribed as k
xx
= 200, k
yy
= 100, k
zz
= 50 md. The
15
0
100
200
300
0 1000 2000 3000 4000
coordinate along the well, ft
i
n
f
l
o
w
,

S
T
B
/
d
s-k*
fd
q= 1.3 STB/d
p= 0.7 psi
p
wf
= 2806.8 psi (fd)
Figure 5: Production prole for wellA in a homogeneous, anisotropic
reservoir with skin prescribed (Wolfsteiner et al., 2000a).
well conguration for this example is wellA of Table 3: the well is fully penetrating and
produces at 10,000 STB/d. Five different skin values (s = 0, 2, 0, 1, 3, see Table 3)
are specied along the well. In the semi-analytical solution, the well is modeled with 45
segments. The semi-analytical (solid curve) and nite difference (dashed curve) solutions
for inow as a function of position along the well are presented in Fig. 5. The curves
represent the total inow into each segment (solid curve) or grid block (dashed curve).
The agreement between the two curves is clearly very close, indicating that skin is prop-
erly implemented into the semi-analytical solution technique. As is evident from Fig. 5,
the inow is reduced in sections of positive skin, as would be expected. End effects at the
boundaries of the skin zones are also apparent; e.g., segments of low skin draw additional
uid from adjacent higher skin regions.
Because wellbore hydraulics are not included in these calculations, pressure is constant
from the heel ( = 0) to the toe ( = 4500 ft) of the well in both calculations. The
difference in the wellbore pressures for the two cases is only 0.7 psi, indicated as p in
the gure. In this and subsequent production plots, a measure of the difference between
the inow proles for the semi-analytical and the nite difference results is also provided.
This quantity, designated q, is simply the L
1
norm of the differences between the two
solutions; i.e., q =
1
n
S

n
S
i
|q
fd
q
sk
|
i
. We nowconsider several heterogeneous cases
and apply the procedures described above to determine the skin prole along the well and
k

.
Fixed Pressure Cases. For the following cases, we consider a 2100 ft partially penetrating
horizontal well (wellB of Table 3) operating at a xed pressure of p
w
=2800 psi. The well
is modeled using 21 segments in the s-k

calculations. Pressure at the bottom boundary


(z=175 ft) is xed at 3000 psi to model the effects of a strong aquifer. In the results
presented below, we consider two different realizations of the same geostatistical model,
permB and permB2. These two realizations do not have identical statistics although they
are very close (cf. Table 2).
16
0
500
1000
1500
2000
0 700 1400 2100
coordinate along the well, ft
i
n
f
l
o
w
,

S
T
B
/
d
s-k*
fd
k* only
Q
fd
= 19,900 STB/d
Q
s-k*
=17,300 STB/d
Q
k*
=18,700 STB/d
Figure 6: Production prole for xed pressure example (permB, wellB)
(Wolfsteiner et al., 2000a).
In Fig. 6 we show the well production prole using permB. Because p
w
is specied, p
is identically zero. The total inow rates are, however, different between the various
models. For the nite difference results, Q
fd
= 19, 900 STB/d, while the s-k

method
gives Q
sk
= 17, 300 STB/d, an error of about 13%. On a per segment basis, the error
in inow rate is about the same. An additional curve is also shown in the gure (denoted
as k

only); this corresponds to the semi-analytical result using k

but setting s = 0
everywhere. Such a result will be obtained with standard semi-analytical methods that
are unable to account for heterogeneity. This result is clearly unable to track the nite
difference result, though it does provide an accurate estimate of the total ow (Q
k
=
18, 700 STB/d, an error of only 6% relative to the nite difference result).
The next case involves permeability eld permB2 (Fig. 7). Here, the agreement of the
s-k

results and the nite difference results is quite close. The total inow rates are
Q
fd
= 10, 300 and Q
sk
= 10, 200 STB/d, respectively, an error of about 1%. On a
per segment basis, this error is also only 1%. The semi-analytical result using k

with
s = 0 is again shown. In this case the results for total inow using this method are poor;
Q
k
= 18, 500 STB/d, an error of 80% relative to the nite difference result. The re-
sults for Q
k
are quite similar for both permB and permB2 because k

is close for the


two cases. However, because the permeabilities along the well path are quite different
between the two realizations, the well productivities differ considerably. The use of k

alone is unable to capture this effect. This is consistent with earlier results for two di-
mensional systems, where the well productivity was seen to vary by orders of magnitude
between realizations of the same statistics, even though k

was the same between realiza-


tions (Durlofsky, 2000).
Dual-Lateral Case. We now consider the multilateral conguration wellC in Table 3. The
two horizontal sections are perforated and have a total production constraint of 10,000
STB/d. Each lateral is modeled using 18 segments. The production proles for the upper
lateral are shown in Fig. 8 and those for the lower lateral in Fig. 9. The inow pro-
17
le computed using s-k

is in close agreement with the nite difference results for both


laterals. As indicated in the gures, the differences in pressure between the nite differ-
ence and s-k

calculations are 4.6 psi for both laterals. The nite difference result for
the total production from the upper lateral is Q
fd
= 2870 STB/d, while the s-k

result
gives Q
sk
= 2570 STB/d, representing a difference of 10%. For the lower lateral,
Q
fd
= 7130 STB/d and Q
sk
= 7430 STB/d; an error of 4%. Differences in the contri-
butions of the two laterals to total well inow are likely due to gravitational effects and
permeability heterogeneity.
The close agreement between the nite difference and s-k

results in this case suggests


that both approaches are reasonably accurate. However, in some cases, such as when
the heels of the two laterals are relatively close together, the nite difference results may
begin to lose accuracy. This is because the nite difference results rely on the simpli-
ed Peaceman (1983) model to compute the well index. The assumptions of that model
(two dimensional radial ow due to an isolated well) are clearly violated when there is
strong communication between the two laterals. In such cases, the s-k

approach might
well provide more accurate results than the nite difference method. In fact, the semi-
analytical model could be used to compute an appropriate well index for use with the
nite difference simulator, as we will demonstrate later in this paper.
0
500
1000
1500
0 700 1400 2100
coordinate along the well, ft
i
n
f
l
o
w
,

S
T
B
/
d
s-k*
fd
k* only
Q
fd
= 10,300 STB/d
Q
s-k*
=10,200 STB/d
Q
k*
=18,500 STB/d
Figure 7: Same as Fig. 6 but with a different permeability realization,
permB2 (Wolfsteiner et al., 2000a).
18
0
100
200
300
0 600 1200 1800
coordinate along the well, ft
i
n
f
l
o
w
,

S
T
B
/
d
s-k*
fd
q= 16.9 STB/d
p= 4.6 psi
p
wf
= 2805.7 psi (fd)
Q
fd
= 2870 STB/d
Q
s-k*
=2570 STB/d
Figure 8: Production prole for the upper lateral of multilateral (permC,
wellC) (Wolfsteiner et al., 2000a).
0
200
400
600
800
0 600 1200 1800
coordinate along the well, ft
i
n
f
l
o
w
,

S
T
B
/
d
s-k*
fd
q= 31.8 STB/d
p= 4.6 psi
p
wf
= 2830.7 psi (fd)
Q
fd
= 7130 STB/d
Q
s-k*
=7430 STB/d
Figure 9: Production prole for the lower lateral of multilateral (permC,
wellC) (Wolfsteiner et al., 2000a).
19
Fluvial Channel Reservoir. The following two examples were generated using the proce-
dure of Valvatne et al. (2001). A heterogeneous uvial-based reservoir description from
Mao and Journel (1999) (Fig. 10) is considered. The permeability distribution was gen-
erated using geostatistical techniques, with channels introduced through use of an object-
based simulation technique (Deutsch, 1997) The overall mean permeability is 176 mD,
though there are two distinct permeability populations. The mean permeability of chan-
nels is about 200 mD, but with a long upper tail. The mudstone-dominated background
has an average permeability of about 6 mD, creating a very sharp contrast between the
background and channel permeabilities. The components of the effective permeability
tensor k

, obtained using a ow-based upscaling technique (Deutsch and Journel, 1998)


are k
xx
= 136, k
yy
= 135 and k
zz
= 12 mD.
This example involves a dual-lateral well, with each lateral penetrating a different channel
in the reservoir. Note that only the lateral branches of the well are perforated. The lower
lateral (lateral 1) is about 1900 ft in length and is perforated along its entire length. The
upper lateral (lateral 2) is of length 700 ft and is perforated only in the channel. The well
completion is modeled with separate segments representing tubing, annulus and chokes,
as depicted in Fig. 11. For the case under study, two chokes are also in this model, located
at the edges of the channels as indicated in the gure via circles. The well is specied to
produce at a total production rate of 10,000 STB/d.
Our computed inow proles at a time of 100 days, with and without choking, are shown
in Fig. 12. In the gure, we plot production rate per unit well length. When the well is not
choked, lateral 1 produces an average of 2.46 STB/d/ft while lateral 2 produces an average
of 7.54 STB/d/ft. Chokes are then set in an attempt to achieve a more nearly uniform (in
terms of production rate per unit length) inow prole. The inow prole in this case is
much atter than in the case with no choking. Both laterals now produce an average of
about 3.84 STB/d/ft. Thus, the objective of more uniform production is achieved through
the use of downhole chokes. Bottomhole pressure as a function of time for the two cases
is shown in Fig. 13. It is apparent that there is a signicant difference in bottomhole
pressure, with the choked well displaying pressures about 570 psi less than those of the
unchoked well. Choking therefore has an adverse impact on the well productivity index.
However, the more uniform inow prole may more than compensate for this effect.
Multiple Reservoirs. Formation properties such as permeability or pressure may vary con-
siderably if the eld is comprised of multiple reservoirs or fault blocks. Using downhole
chokes it is possible to achieve balanced commingled production since the different in-
ow zones do not need to be produced at the same bottomhole pressure. This has in fact
been the most common reason for installing downhole chokes (Lie and Wallace, 2000).
To demonstrate the application of the semi-analytical approach to such cases, Valvatne
et al. (2001) consider a reservoir containing three non-communicating fault blocks
(Fig. 14), with each fault having a vertical throw of 100 ft. The boundary conditions in
the y-direction (front and back of the model) are constant pressure; in the other directions
the boundaries are no-ow. This results in the reservoir achieving steady state conditions,
as might be approximated if a strong aquifer were present.
There is a 500 psi pressure differential across each fault, with the rightmost block having
the highest initial potential of 3,500 psi. The well is horizontal and is completed over
20
1000 MD
100 MD
10 MD
1 MD
Lateral 1
Lateral 2
Well
chokes
Figure 10: Fluvial reservoir permeability eld and a dual-lateral well
(Valvatne et al., 2001).
Figure 11: Segment structure when modeling chokes (Valvatne et al.,
2001).
1,000 ft in each block. The permeability along the well is again highly variable (Fig. 15).
Conventional completions (i.e., completions without chokes) would not be feasible in this
case due to the large pressure differences between fault blocks.
Results for the inow proles and bottomhole pressures (at sand face), with and without
choking, are shown in Fig. 16 and Fig. 17. Without choking the low pressure in the left
block results in cross ow from the other blocks (apparent from the negative inow along
the rst 1,000 ft of the well in Fig. 16). Using downhole chokes, it is possible to produce
each reservoir with an optimal bottomhole pressure, as shown in Figs. 16 and 17, such
that the inow is more uniform and is distributed according to the perforated well length
(Table 4), or any other criterion.
In Fig. 18 the optimized inow prole computed using the semi-analytical model is com-
pared with that obtained using a reservoir simulator (Schlumberger, 1999). The agreement
between the two results is quite close. In some cases, such as when the well is not ori-
ented along a coordinate direction or when the well is a complex multilateral, the results
21
Figure 1: Permeability Distribution in fluvial reservoir
Figure 2: Optimized inflow profile with downhole chokes
Figure 3: Chokes influence on BHP
Lateral 1
Lateral 2
Well
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500
cumulative length, ft
i
n
f
l
o
w
,

s
t
b
/
d
a
y
/
f
t
Choking
No Choking
Lateral 1 Lateral 2
0
500
1000
1500
2000
2500
3000
0 200 400 600 800 1000
time, days
B
H
P
,

p
s
i
a
Choking
No Choking
Figure 12: Inow proles with and without choking for dual-lateral
well (Valvatne et al., 2001).
Figure 1: Permeability Distribution in fluvial reservoir
Figure 2: Optimized inflow profile with downhole chokes
Figure 3: Chokes influence on BHP
Lateral 1
Lateral 2
Well
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500
cumulative length, ft
i
n
f
l
o
w
,

s
t
b
/
d
a
y
/
f
t
Choking
No Choking
Lateral 1 Lateral 2
0
500
1000
1500
2000
2500
3000
0 200 400 600 800 1000
time, days
B
H
P
,

p
s
i
a
Choking
No Choking
Figure 13: Bottomhole pressure with and without choking for dual-
lateral well (Valvatne et al., 2001).
from the semi-analytical approach might even be more accurate than those from the nite
difference simulation. This is because the nite difference results can lose accuracy when
the well is not oriented along the grid or when the various branches of the multilateral in-
teract strongly. The semi-analytical calculations, by contrast, do not suffer any additional
inaccuracies in this case as they are not based on an underlying grid.
22
Figure 14: Reservoir geometry with fault blocks (Valvatne et al., 2001).
Figure 15: Altered permeability k
a
in the near-well area (Valvatne et al.,
2001).
Figure 16: Inow prole in faulted reservoir (Valvatne et al., 2001).
23
Figure 17: Bottomhole pressure (at sand face) required to optimize in-
ow prole in faulted reservoir (Valvatne et al., 2001).
Table 4: Optimized inow using downhole chokes (Valvatne et al.,
2001).
zone
inow without
choking (stb/d)
inow with
choking (stb/d)
1 -3,657 9,836
2 5,915 9,722
3 27,742 10,442
Figure 18: Comparison with reference simulator when using downhole
chokes in faulted reservoir (Valvatne et al., 2001).
24
Well Indices for NCWs
The concept of well indices carries directly over to more complex well congurations. As
for conventional well models, some formof reference solution (analytical, semi-analytical
or numerical solution of the pressure equation) is required for NCWs. A comparison to
the nite difference model yields the well index. Mochizuki (1995) extended Peacemans
original model to account for deviated well segments in a Cartesian grid block. Klausen
and Aavatsmark (2000) solved the single phase pressure equation for an innite well in
an innite reservoir analytically and obtained well connection factors by comparison to
a nite difference model. G okas
.
and Ertekin (1999a,b) modeled slanted and undulating
wells using local renement of coarse blocks that contained deviated segments.
Modeling NCWs on grids that closely follow the trajectory shifts the problem complexity
towards gridding and discretization issues. The benet is that simpler (i.e., conventional)
well models or no well models (in the case of explicit wellbore resolution) can be used.
Deb and Reddy (1995) and Reddy et al. (1997) took a nite element approach using
highly exible grids to explicitly model well trajectories. The explicit modeling tech-
nique must be used with care as discussed by Wan et al. (2000). Along the same lines
is the work of Baumann et al. (1999) who used a generalized nite element approach to
solve the pressure equation for the modeling of multiple wells in heterogeneous reser-
voirs. Similar approaches using control volume methods, which are more common in
reservoir simulation, have also been presented. Gunasekera and Cox (1997) employed
k-orthogonal grids, with the permeability dened on tetrahedra, for the solution of the
pressure equation. Heinemann et al. (2000) resolved NCWs accurately using three di-
mensional Voronoi grids embedded in two and a half dimensional grids of the same type.
For a more complete review refer to Wolfsteiner (2002).
We will use the semi-analytical technique discussed earlier to compute well indices for
NCWs. While the s-k

methodology is useful as a fast stand alone model, the overall


applicability of the method is limited because it is restricted to single phase ow. The
semi-analytical solution methodology does, however, offer a new way to obtain the refer-
ence well pressure for the determination of accurate WI. These accurate WI can then be
used within general reservoir simulators for the modeling of multiphase, multi-component
systems.
As discussed earlier, the reservoir simulator cannot directly model the details of ow
around the well within a gridblock. For this we need a well model to relate the well
ow rate to the difference in the wellblock and well pressures. Eq. 8 is the default well
model in most simulators. This simple model is based on a number of assumptions. The
model works well for isolated wells in uniform Cartesian grids where the ow around the
well is essentially radial (one dimensional). In many situations, particulary for NCWs
in irregular grids, Eq. 8 may not be appropriate. For these situations we require a more
general approach. The basic idea is evident from Eq. 6:
WI
i
=
m
w
i

(p
i
p
w
i
)
, (22)
or we can dene a per block productivity index as
PI
i
= WI
i

. (23)
25
p
w
j
reference
solution
map onto
target grid
preprocessing
simulation with

WI*
i
=
2 segments j
intersect
48 grid blocks i
prescribed
yields
q
w
j
p
w
i
q
w
i
q
w
i
p
i
q
w
i
p
w
i
local problem
subscripts:
p
i
Figure 19: Flowchart for obtaining coarse scale well indices.
Often Eq. 22 is written in terms of volumetric rather than mass ow rate as:
WI
i
=
q
w
i

p
i
p
w
i
. (24)
We will use this form in the following discussion. Note that WI has units of permeability
times length (cf. Eq. 8), or equivalently, length cubed.
Clearly, if for a well or a well segment, the ow rate, well pressure and gridblock pressure
are known, WI
i
can be readily computed. Here we discuss the use of semi-analytical
techniques to obtain accurate WI for arbitrary wells and grids. The gridblock pressure
can be obtained from single-phase numerical calculations by specifying the well (or seg-
ment) ow rate. Details of this procedure are discussed next. One of our objectives is to
approximately account for near well heterogeneities in the calculation of WI.
Calculation of WI Using s-k

This section shows how the s-k

semi-analytical solution
described earlier can be used as a reference solution for computing well indices for an
arbitrary coarse grid. The concept of well indices is general and can also be applied
to unstructured coarse grids. Note that Eq. 22 makes no assumptions about the block
geometry.
Our method is different from existing approaches as the entire well trajectory (or the tra-
jectory of multiple wells that inuence each other) is modeled in a local well problem and
26
the resulting inuxes and pressure distributions are then mapped onto the wellblocks of
the nite difference simulation. Also of considerable importance is the fact that our model
considers reservoir heterogeneity. We are interested in obtaining appropriate well indices
for each wellblock i for an arbitrary target grid and any complex well trajectory. We rst
describe aspects of the general method applicable to both homogeneous and heterogenous
reservoirs and then discuss some details of the heterogeneous problem.
Let us consider a local well driven ow problem as depicted on the top of Fig. 19. The
well domain selected is approximately the drainage volume for a single or multiple wells.
The boundary conditions can be constant potential or no ow, as appropriate. Here we
specify the well ow rate and no ow boundaries for all boundaries except the bottom,
where constant pressure is specied to model a strong aquifer. Because of the constant
pressure boundary, the well will reach steady-state. The local problem is solved using the
semi-analytical s-k

method described above. The solution provides the well pressure p


w
j
and inow q
w
j
for each segment j. This is considered to be the reference solution and
will be used to compute the appropriate well indices for the simulation grid. The third
parameter in Eq. 22, p
w
i
, is discussed next.
The same local domain and boundary conditions as used in the semi-analytical solution
are applied to obtain the numerical solution for the target grid. For a heterogeneous prob-
lem, the permeability information must rst be upscaled to the target grid. Rather than
specifying the total wellow rate, here we prescribe sources q
w
i
for each block intersected
by the well. As a segment j is generally not contained by exactly one block i (as indicated
in Fig. 19), we need to apply a mapping procedure to compute the intersections of the well
with the grid. Our current implementation of this algorithm is general (Arvo, 1991) in that
any polyhedral unstructured grid can be considered as long as the grid cells are convex.
With this information, the rates q
w
i
are interpolated based on the fractional exposure of the
well segment to a given block i. Pressures p
w
i
, needed for the nal calculation of the WI

i
, are the reference p
w
j
s corrected by the hydrostatic head between the segment and block
midpoints. The prescribed block rates q
w
i
act as simple source terms in the simulator; i.e.,
the well model of the target simulator is not used in this procedure. From this step, the
pressure of each well block p
i
is determined.
The correct well index for block i follows now from combining the semi-analytical (refer-
ence) well pressure and rate information, interpolated onto the well blocks, and the block
pressures obtained in the simulation step:
WI

i
=
q
w
i

p
i
p
w
i
. (25)
The * on WI

i
is used to distinguish it from the default WI or WI
d
computed by Eq. 8.
The overall procedure is similar for a heterogenous system. In this case, the s-k

reference
solution is only approximate, though it does use the geostatistical ne grid permeability
information. This adds an upscaling component to the method, as the effects of near well
ne scale heterogeneity are included in the coarse grid WI

i
. Due to the approximate
nature of the s-k

approach, and slight inconsistencies between the semi-analytical and


nite difference solution methods, it is possible that the procedure will result in a block
pressure p
i
that is actually higher than the wellbore pressure p
w
i
for an injection well.
Application of Eq. 25 for the calculation of WI

would, in this case, provide a negative


well index. When this occurs, we introduce an additional iteration loop into our procedure
27
Figure 20: Pinchout multi-block grid with NCW (three segments).
0
2
4
6
8
1 8 15 22 29 36
segment number
i
n
f
l
o
w

p
e
r

l
e
n
g
t
h
,

S
T
B
/
d
/
f
t
2600
2700
2800
2900
3000
3100
s
e
g
m
e
n
t

p
r
e
s
s
u
r
e
,

p
s
i
influx
segment
pressure
Figure 21: Semi-analytical reference solution for the pinchout example.
to force the calculation of positive WI

.
Examples We rst demonstrate the proposed method on a general three dimensional
grid containing a complex well trajectory. The example illustrates the generality of the
method. In subsequent cases we use Cartesian grids to allow for comparisons between
our results and those obtained using conventional models (i.e., Peaceman well index WI
d
from Eq. 8). Homogeneous cases will be used to assess the geometrical effects; studies
with heterogenous reservoirs will demonstrate the ability of this approach to match some
aspects of the ne grid results. For the Cartesian cases presented, we compute the WI

distribution for a reservoir with a constant pressure boundary at the bottom. This allows us
to use long time semi-analytical solutions together with fast steady state nite difference
calculations. Tables 1 to 3 give properties for the homogenous and heterogenous runs as
well as the well trajectories used for the Cartesian grid examples.
Pinchout Multi-Block Grid. We will use an unstructured grid constructed to model a pin-
chout in our rst example. The NCW trajectory is modeled with only three segments as
28
Figure 22: Intersection of NCW with grid yields well blocks.
Figure 23: Pressure eld from nite difference simulation.
shown in Fig. 20.
The multiblock grid used here is locally structured and globally unstructured (Jenny et al.,
2001). The reservoir is homogenous and isotropic. The rst step in our procedure (see
Fig. 19) is to obtain the semi-analytical solution, shown in Fig. 21.
As a next step we perform an intersection of the NCW with the grid as shown in Fig. 22.
This enables us to map the reference ow rates onto the target grid, which are subse-
quently used as source terms in the nite difference calculations. The pressure eld for
this solution is shown in Fig. 23. The resulting well index distribution WI

is shown
in Fig. 24. We now apply these WI

i
s in our simulator with a typical well constraint,
such as a target rate. Fig. 25 compares inow proles and well segement pressures along
the well trajectory computed using the nite difference (triangles and quads) and semi-
analytical solution techniques (solid line). As expected the nite difference solution with
WI

traces the reference result very accurately. This demonstrates the applicability of the
method for complex grids and general well trajectories.
29
0.0
0.1
0.1
0.2
0.2
0.3
0.3
0.4
1 8 15 22 29 36
segment number
W
I
*

p
e
r

l
e
n
g
t
h
,

S
T
B
.
c
p
/
d
/
p
s
i
/
f
t
Figure 24: WI

distribution.
0
2
4
6
8
1 8 15 22 29 36
segment number
i
n
f
l
o
w

p
e
r

l
e
n
g
t
h
,

S
T
B
/
d
/
f
t
2600
2700
2800
2900
3000
3100
s
e
g
m
e
n
t

p
r
e
s
s
u
r
e
,

p
s
i
influx
segment
pressure
Figure 25: Inux and pressure match for pinchout example (solid: s-k

reference, bullets
application of WI

).
Skewed Horizontal Well. The following examples involve Cartesian grids of dimension
45 45 33 cells. The well in the rst example follows conguration wellA30 (see
Table 3). The well is rotated 30 degrees in the horizontal plane relative to the orientation
of the grid lines. The reservoir is homogenous with k
xx
= 200, k
yy
= 100, k
zz
= 50
md and has a constant pressure aquifer (p
aqu
=3000 psi). We compute WI

for this
case as described above. In Fig. 26 the well pressure is shown as a function of time.
Finite difference results using both the default well index WI and WI

are compared
to the semi-analytical reference solution. Note that the pressure axis is scaled such that
its maximum value identies the initial wellbore pressure (here 2963.5 psia). Hence, the
interval between any of the lines and the top of the graph is equal to the drawdown at a
given time.
Most notable in Fig. 26 is the fact that simulation results using a default Peaceman index
differ from the exact reference solution, while those computed using WI

agree closely.
It is interesting to note that, although the WI

i
s were calculated from the steady state
problem, the curve matches the reference in the transient region as well. The WI

distri-
bution is compared to the default well index WI in Fig. 27. The default well index for
30
2930
2935
2940
2945
2950
2955
2960
0.01 0.1 1 10 100
time, days
w
e
l
l
b
o
r
e

p
r
e
s
s
u
r
e
,

p
s
i
s-k* ref
fd WId
fd WI*
p
i
=2963.5
Figure 26: Skewed well drawdown.
0.09
0.10
0.11
0.12
0.13
0.14
0.15
1 17 33 49
segment number
w
e
l
l
i
n
d
e
x

p
e
r

l
e
n
g
t
h
,

s
t
d
/
d
/
p
s
i
/
f
t
WI*
WId
Figure 27: WI
d
and WI

for the skewed well case.


this geometry and anisotropy has a value of 12.32 (in appropriate units) and is constant
for all blocks. The WI

prole is clearly different.


Multi-lateral Well. We repeat the WI

calculation procedure for a more complex well tra-


jectory (wellD in Table 3). The well is now vertically dipping and has a horizontal lateral
extending from it. The reservoir remains unchanged (constant pressure aquifer). We plot
the well pressure as a function of time in Fig. 28. The ordinate is again scaled such that
the topmost value identies the intial static pressure level. In this case an even higher error
for the default WI is observed. We now test if changing the reservoir boundary conditions
has an inuence on the quality of nite difference results computed using WI

(recall that
WI

is computed from a steady state solution). Fig. 29 shows the well drawdown for the
same well conguration in a closed reservoir (i.e., no aquifer). The exact solution is still
closely reproduced. Note that while the absolute deviation of the two simulator solutions
is nearly the same (about 6 psi), the relative error in terms of drawdown is smaller for the
closed reservoir case, due to the changed ow scenario.
31
2940
2945
2950
2955
2960
0.01 0.1 1 10 100
time, days
w
e
l
l
b
o
r
e

p
r
e
s
s
u
r
e
,

p
s
i
s-k* ref
fd WId
fd WI*
p
i
=2963.5
Figure 28: Multi-lateral with constant pressure aquifer.
2675
2725
2775
2825
2875
2925
0 20 40 60 80 100
time, days
w
e
l
l
b
o
r
e

p
r
e
s
s
u
r
e
,

p
s
i
s-k* ref
fd WId
fd WI*
p
i
=2939.4
Figure 29: Multi-lateral example in closed reservoir.
Heterogeneous Reservoir. In the previous example we assessed the accuracy and robust-
ness of our procedure for different boundary conditions. We now consider reservoir het-
erogeneity. Recall that in this case our representation of permeability variation via s-k

is only approximate. We therefore choose a case for which we can regard the ne grid
nite difference result as the basis for comparison. The coarse Cartesian target grid is
of dimensions 15 15 11; i.e. an upscaling factor of three in each direction. The
permeability eld is dened on the underlying ne grid and was upscaled to the coarse
grid (using a ow based upscaling procedure (Deutsch and Journel, 1998)). In order to
minimize the effects of geometry we consider the case of a fully penetrating horizontal
well that is centered on both the ne and coarse grid. For the ne grid case the default
well indices should be sufciently accurate, allowing us to regard this as the reference
solution.
Well pressure as a function of time is plotted in Fig. 30. The comparison between the ne
grid and coarse grid simulation, both generated using the default well index, shows that
the coarse grid result does not degrade much with upscaling. The accuracy of the ne
grid result was veried by a run on a downscaled grid of size 115 115 99 cells, which
was obtained by replacing each geostatistical ne grid cell by 3 3 3 downscaled cells
32
2925
2930
2935
2940
2945
2950
2955
2960
0.01 0.1 1 10 100
time, days
w
e
l
l
b
o
r
e

p
r
e
s
s
u
r
e
,

p
s
i
s-k* ref
fine WId
coarse WId
coarse WI*
p
i
=2963.5
Figure 30: Well drawdowns for heterogeneous reservoir.
0
200
400
600
800
1000
1200
1400
1 5 9 13
segment number
i
n
f
l
u
x
,

s
t
d
/
d
s-k* ref
fine WId
coarse WId
coarse WI*
Figure 31: Inux proles for heterogeneous case.
of equal permability. The plot also reveals that the s-k

reference result (using ne grid


information) is about as accurate as the coarse grid result in terms of wellbore pressure
predictions. For both cases, the deviation from the ne grid result is about 2 psi. This
result can be attributed to the fact that the coarse grid solution is surprisingly good. The
WI

i
s were computed for the coarse grid and they reproduce the semi-analytical solution,
as expected.
For heterogeneous reservoirs, the inow prole along the well trajectory can be even
more important than the wellbore pressure. In Fig. 31 we display the various solutions
for inow as a function of position along the wellbore. Here, the coarse grid result with
default well indices is in signicant error relative to the ne grid result, while the coarse
grid result using WI

tracks the ne grid solution closely. This illustrates the ability of


our procedure to provide accurate inow proles, even in cases when wellbore pressures
show some errors.
33
Conclusions
In this paper we have demonstrated the ability of newly developed semi-analytical tech-
niques to model a variety of problems associated with NCWs. In particular we have
shown that such models can be used for quick estimates of the well productivity with
single phase ow in homogeneous and heterogeneous reservoirs. The NCWs discussed
in this paper include
horizontal wells,
skewed wells not aligned with reservoir boundaries,
advanced wells with control devices, and
multilateral wells.
Since the WI calculation requires a reference solution that gives the exact well pressure for
the corresponding single phase problem, we have used the new semi-analytical technique
for this purpose. A preprocessing procedure is described that gives the exact well index
for arbitrary NCW congurations on any kind of grid.
The semi-analytical technique can also be used to obtain approximate solutions for hetero-
geneous systems, by accounting for near well variations in permeability. We have shown
how such solutions can be used to account for near well heterogeneities in the calculation
of WI.
Acknowledgments
The research reported here was supported by the U.S. Department of Energy
under contract number DE-AC26-99BC15213, and by a consortium of compa-
nies under the SUPRI-HW Industrial Afliates Program at Stanford University
(http://ekosk.stanford.edu/horiz.html). Results from the research of several previous
students are included in this paper. We wish to particularly acknowledge the contributions
of Dr. Liang-Biao Ouyang and Mr. Per H. Valvatne. We also thank Mr. Jerome Serve,
who generated the results for the dual lateral well in a uvial reservoir.
34
Nomenclature
A area open to ow
B formation volume factor
c
t
total compressibility
f friction factor with inow
f
o
friction factor without inow
g gravitational acceleration
g
c
gravitational conversion constant
h segment length
H vertical level
J jacobian matrix
k permeability (scalar)
k
r,p
relative pemeability for phase p
k permeability tensor
k

effective permeability tensor


k
s
geometric mean of k

components
l dimensionless correlation length
m mean (expected value)
or mass ow rate
M (equivalent) pressure point
or total mass
n number
N
S
total number of segments
p pressure
p
w
owing wellbore pressure
PI productivity index
q volumetric ow rate
Q total inow (into lateral)
r radius
R residual equation
Re
w
inow Reynolds Number
R residual vector
s skin factor
S segment or saturation
t time
T transmissibility
V block volume
WI well index
y
c,p
mass fraction of component c
in phase p
, conversion factors
porosity
potential or Greens function
normalization factor
t timestep
x gridblock size
viscosity
density
standard deviation

v
coefcient of variation (/m)
power averaging exponent
coordinate along the lateral
Subscripts
0 initial
a altered zone (search region)
or accelerational
aqu aquifer
c component
d coordinate direction
D dimensionless
e effective or total extent
eq equivalent
f frictional
fd nite difference result
h harmonic average
i block index
I inow
k

result using k

only
l index for block connected to block i
L lateral
m mass balance
p phase or potential
S segment
s-k

result using s-k

method
x, y, z coordinate direction
w well
Superscripts
effective property
d default
w well
35
References
Ababou, R. (1990). Identication of Effective Conductivity Tensor in Randomly Hetero-
geneous and Stratied Aquifers, presented at the Canadian-American Conference on
Hydrogeology, Calgary, Alberta, September 1820.
Almehaideb, R., Aziz, K., and Pedrosa, O. A. (1989). A Reservoir/Wellbore Model
for Multiphase Injection and Pressure Transient Analysis, paper SPE 17941 presented
at the SPE Middle East Oil Technical Conference and Exhibition, Manama, Bahrain,
March 1114.
Arvo, J., editor (1991). Graphic Gems II. Academic Press, Inc.
Aziz, K. and Settari, A. (1979). Petroleum Reservoir Simulation. Applied Science Pub-
lishers Ltd., London.
Aziz, K., Arbabi, S., and Deutsch, C. V. (1999). Why is it so Difcult to Predict the
Performance of Horizontal Wells? JCPT October, pp 3745.
Babu, D. and Odeh, A. (1989). Productivity of a Horizontal Well. SPERE November, pp
41721.
Basquet, R., Alabert, F. G., Caltagirone, J. P., and Batsale, J. C. (1998). ASemi-Analytical
Approach for Productivity Evaluation of Wells with Complex Geometry in Multilay-
ered Reservoirs, paper SPE 49232 presented at the SPE Annual Technical Conference
and Exhibition, New Orleans, LA, September 2730.
Baumann, C. E., Price, H. S., Reddy, M. P., and Thuren, J. B. (1999). Full Field Pressure
Simulations Using a Very Accurate, Yet Inexpensive Well Model, paper SPE 56620
presented at the SPE Annual Technical Conference and Exhibition, Houston, TX, Oc-
tober 36.
Bendakhlia, H. and Aziz, K. (1989). Inow performance Relationships for Horizontal
Wells in Solution Gas Drive Reservoirs, paper SPE 19823 presented at the SPE Annual
Technical Conference and Exhibition, San Antonio, TX, October 811.
Besson, J. (1990). Performance of Slanted and Horizontal Wells on an Anisotropic
Medium, paper SPE 20965 presented at the Europec, The Hague, Netherlands, October
2224.
Brigham, W. (1990). Discussion of Productivity of a Horizontal Well. SPERE May, pp
25455.
Byer, T., Edwards, M., and Aziz, K. (1999). A Preconditioned Adaptive Implicit Method
for Reservoir Simulation and Surface facilities, paper SPE 51895 presented at the SPE
Symposium on Reservoir Simulation, Houston, TX, February 1417.
Carslaw, H. S. and Jaeger, J. C. (1959). Conduction of Heat in Solids. Oxford University
Press, United Kingdom.
Clonts, M. D. and Ramey, Jr., H. J. (1986). Pressure Transient Analysis for Wells with
Horizontal Drainholes, paper SPE 15116 presented at the SPE California Regional
Meeting, Oakland, CA, April 24.
36
Coats, K. H., Thomas, L. K., and Pierson, R. G. (1995). Compositional and Black Oil
Reservoir Simulation, paper SPE 29111 presented at the SPE Symposium on Reservoir
Simulation, San Antonio, TX, February 1215.
Daviau, F., Mouronval, G., Bourdarot, G., and Curutchet, P. (1988). Pressure Analysis for
Horizontal Wells. SPEFE December, pp 71624.
de Souza, A., Arbabi, S., and Aziz, K. (1998). Practical Procedures to Predict Cresting
Behavior of Horizontal Wells. SPEJ December, pp 38292.
Deb, M. K. and Reddy, M. P. (1995). A New Generation Solution Adaptive Reservoir
Simulator, paper SPE 30720 presented at the SPE Annual Technical Conference and
Exhibition, Dallas, TX, October 2225.
Desbarats, A. (1994). Spatial Averaging of Hydraulic Conductivity Under Radial Flow
Conditions. Math. Geol., pp 121.
Deutsch, C. (1997). Developing Hierarchical Object-Based Stochastic Modeling of Flu-
vial Reservoirs. Technical report, SCRF.
Deutsch, C. and Journel, A. G. (1998). GSLIB: Geostatistical Software Library and
Users Guide. Oxford University Press, Reading, MA, 2nd edition.
Dikken, B. J. (1990). Pressure Drop in Horizontal Wells and Its Effects on Production
Performance. JPT November, pp 142633.
Ding, Y. (1995). Scaling-up in the Vicinity of Wells in Heterogeneous Field, paper SPE
29137 presented at the SPE Symposium on Reservoir Simulation, San Antonio, TX,
February 1215.
Durlofsky, L. J. (2000). An Approximate Model for Well Productivity in Heterogeneous
Porous Media. Math. Geol., pp 42138.
Durlofsky, L. J., Milliken, W. J., and Bernath, A. (2000). Scale Up in the Near-Well
Region. SPEJ March, pp 11017.
Economides, M. J., Brand, C. W., and Frick, T. P. (1996). Well Congurations in
Anisotropic Reservoirs. SPEFE December, pp 25762.
G okas
.
, B. and Ertekin, T. (1999a). Development of a Local Grid-Renement Technique
for Accurate Representation of Cavity-Completed Wells in Reservoir Simulators. SPEJ
September, pp 18795.
G okas
.
, B. and Ertekin, T. (1999b). Implementation of a Local Grid Renement Tech-
nique in Modeling Slanted, Undulating Horizontal and Multi-Lateral Wells, paper SPE
56624 presented at the SPE Annual Technical Conference and Exhibition, Houston,
TX, October 36.
Goode, P. A. and Thambynayagam, R. K. M. (1987). Pressure Drawdown and Buildup
Anlysis of Horizontal Wells in Anisotropic Media. SPEFE December, pp 68397.
Gringarten, A. C. and Ramey, Jr., H. J. (1973). The Use of Source and Greens Functions
in Solving Unsteady-Flow Problems in Reservoirs. SPEJ October, pp 28596.
37
Gunasekera, D. and Cox, J. (1997). The Generation and Application of K-Orthogonal
Grid Systems, paper SPE 37998 presented at the SPE Symposium on Reservoir Simu-
lation, Dallas, TX, June 811.
Hawkins, M. F. (1956). A Note on the Skin Effect. Trans. AIME, pp 35657.
Heinemann, G., Abdelmawla, A., and Brockhauser, S. (2000). Modelling of Fluid Flow
Around and Within Highly Deviated Horizontal Wells. In Proceedings of the Seventh
European Conference on the Mathematics of Oil Recovery.
Holmes, J. (2001). Modeling Advanced Wells in Reservoir Simulation. JPT. To be
published.
Jenny, P., Wolfsteiner, C., Lee, S. H., and Durlofsky, L. J. (2001). Modeling Flow in
Geometrically Complex Reservoirs Using Hexahedral Multi-Block Grids, paper SPE
66357 presented at the SPE Reservoir Simulation Symposium, Houston, TX, February
1114.
Joshi, S. (1991). Horizontal Well Technology. PennWell Books, Tulsa, OK.
Joshi, S. D. (1988). Augmentation of Well Productivity With Slant and Horizontal Wells.
JPT June, pp 72939.
Klausen, R. A. and Aavatsmark, I. (2000). Connection Transmissibility Factors in Reser-
voir Simulation for Slanted Wells in 3D Grids. In Proceedings of the Seventh European
Conference on the Mathematics of Oil Recovery.
Kuchuk, F. and Brigham, W. (1979). Transient Flow in Elliptical Systems. SPEJ Decem-
ber, pp 4018.
Kuchuk, F. J., Goode, P. A., Brice, B. W., Sherrared, D. W., and Thambynayagam, R.
K. M. (1988). Pressure Transient Analysis and Inow Performance for Horizontal
Wells, paper SPE 18300 presented at the SPE Annual Technical Conference and Exhi-
bition, Houston, Texas, October 25.
Landman, M. J. (1994). Analytical Modeling of Selectivity Perforated Horizontal Wells.
J. Pet. Sci. and Eng., pp 17988.
Lee, S. H. and Milliken, W. J. (1993). The Productivity Index of an Inclined Well in Finite-
Difference Reservoir Simulation, paper SPE 25247 presented at the SPE Symposium
on Reservoir Simulation, New Orleans, LA, February 28March 3.
Lie, O. and Wallace, W. (2000). Intelligent Recompletion Eliminates the Need for Ad-
ditional Well, paper SPE 59210 presented at the IADC/SPE Drilling Conference, New
Orleans, LA, February 2325.
Maizeret, P. D. (1996). Well Indices for Non-Conventional Wells. Masters thesis, Stanford
University, CA.
Mao, S. and Journel, A. (1999). Generation of a Reference Petrophysical / Seismic Data
Set: The Stanford IV Reservoir. Technical report, SCRF.
38
Mascarenhas, O. and Durlofsky, L. J. (2000). Coarse Scale Simulation of Horizontal
Wells in Heterogeneous Reservoirs. J. Pet. Sci. and Eng. March, pp 13547.
Mochizuki, S. (1995). Well Productivity for Arbitrarily Inclined Well, paper SPE 29133
presented at the SPE Symposium on Reservoir Simulation, San Antonio, TX, February
1213.
Muggeridge, A. H., Cuypers, M., Bacquet, C., and Barker, J. W. (1999). Scale-up of Well
Performance for Reservoir Flow Simulation, presented at the European Symposium on
Improved Oil Recovery, Brighton, United Kingdom, August 1820.
Muskat, M. (1937). Flow of Homogeneous Fluids Through Porous Media. McGraw Hill.
Nolan, J. S. (1990). Treatment of Wells in Reservoir Simulation, presented at the Third
International Forum on Reservoir Simulation, Baden, Austria, July 23-27.
Novy, R. A. (1995). Pressure Drops in Horizontal Wells: When Can They Be Ignored?
SPERE February, pp 2935.
Ouyang, L.-B. (1998). Single Phase and Multiphase Fluid Flow in Horizontal Wells.
Ph.D. thesis, Stanford University, CA.
Ouyang, L.-B. and Aziz, K. (1998). A Simplied Approach to Couple Wellbore Flow and
Reservoir Inow for Arbitrary Well Conguration, paper SPE 48936 presented at the
SPE Annual Technical Conference and Exhibition, New Orleans, LA, Sept. 2730.
Ouyang, L.-B., Arbabi, S., and Aziz, K. (1998). A Single-Phase Wellbore-Flow Model
for Horizontal, Vertical and Slanted Wells. SPEJ June, pp 12433.
Ozkan, E., Raghavan, R., and Joshi, S. (1989). Horizontal Well Pressure Analysis. SPEFE
December, pp 56775.
Ozkan, E., Sarica, C., and Haci, M. (1999). Inuence of Pressure Drop Along the Well-
bore on Horizontal-Well Productivity. SPEJ September, pp 288301.
Palagi, C. and Aziz, K. (1994). Modeling of Vertical and Horizontal Wells With Voronoi
Grid. SPERE February, pp 1521.
Peaceman, D. W. (1978). Interpretation of Wellblock Pressures in Numerical Reservoir
Simulation. SPEJ June, pp 18394.
Peaceman, D. W. (1983). Interpretation of Well-Block Pressure in Numerical Reservoir
Simulation with Nonsquare Grid Blocks and Anisotropic Permeability. SPEJ June, pp
53143.
Peaceman, D. W. (1990). Interpretation of Wellblock Pressures in Numerical Reservoir
Simulation Part 3 Off Center and Multiple Wells Within a Wellblock. SPERE May,
pp 22732.
Pedrosa, O. A. and Aziz, K. (1986). Use of Hybrid Grid in Reservoir Simulation. SPERE
November, pp 61121.
39
Penmatcha, V. R. and Aziz, K. (1998). A Comprehensive Reservoir/Wellbore Model for
Horizontal Wells, paper SPE 39521 presented at the India Oil and Gas Conference and
Exhibition, New Dehli, India, February 1719.
Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P. (1993). Numerical
Recipes in C, The Art of Scientic Computing. Cambridge University Press, 2nd edition.
Reddy, M. P., Deb, M. K., Bass, J. M., and Ning, H. (1997). Numerical Simulation of
Non-Conventional Wells Using Adaptive Finite Element Analysis. Comp. Meth. in
Appl. Mech. and Eng. December, pp 10924.
Sachdeva, R., Schmidt, Z., Brill, J., and Blais, R. (1996). Two-Phase Flow Through
Chokes, paper SPE 15657 presented at the SPE Annual Technical Conference and Ex-
hibition, New Orleans, LA, October 58.
Schlumberger, G. (1999). Eclipse 100 Reference Manual. Technical report.
Soeriawinata, T., Kasap, E., and Kelkar, M. (1997). Permeability Upscaling for Near-
Wellbore Heterogeneities. SPEFE December, pp 25562.
Thomas, L. K., Todd, B. J., Evans, C. E., and Pierson, R. G. (1996). Horizontal Well IPR
Calculations, paper SPE 36753 presented at the SPE Annual Technical Conference and
Exhibition, Denver, CO, October 69.
Valvatne, P. H., Durlofsky, L. J., and Aziz, K. (2001). Semi-Analytical Modeling of the
Performance of Intelligent Well Completions, paper SPE 66368 presented at the SPE
Reservoir Simulation Symposium, Houston, TX, February 1114.
Vogel, J. (1968). Inow performance Relationships for Solution Gas-Drive Wells. JPT
January, pp 8392.
Wan, J., Penmatcha, V., Arbabi, S., and Aziz, K. (2000). Effect of Grid Systems on
Predicting Horizontal-Well Productivity. SPEJ September, pp 30914.
Wolfsteiner, C. (2002). Modeling and Upscaling of Non-Conventional Wells in Heteroge-
neous Reservoirs. Ph.D. thesis, (to be completed), Stanford University, CA.
Wolfsteiner, C., Durlofsky, L. J., and Aziz, K. (2000a). An Approximate Model for the
Productivity of Non-Conventional Wells in Heterogeneous Reservoirs. SPEJ June, pp
21826.
Wolfsteiner, C., Durlofsky, L. J., and Aziz, K. (2000b). Efcient Estimation of the Ef-
fects of Wellbore Hydraulics and Reservoir Heterogeneity on the Productivity of Non-
Conventional Wells, paper SPE 59399 presented at the SPE Asia Pacic Conference,
Yokohama, Japan, April 2526.
Yeten, B., Wolfsteiner, C., Durlofsky, L. J., and Aziz, K. (2000). Approximate Finite
Difference Modeling of the Performance of Horizontal Wells in Heterogeneous Reser-
voirs, paper SPE 62555 presented at the Western Regional Meeting in Long Beach,
Long Beach, CA, June 1923.
40

Anda mungkin juga menyukai