..
A. S. Odeh, SPE-AIME,Mobil Research& DevelopmentCorp.
Introduction
Reservoir simulation is based on well known reservoir
engineering equations and techniques — the same
equations and techniques the reservoir engineer has
been using for years.
In generai, sirmukition .e.w.. .- Lie representation
.-f m-c tO
of some process by either a theoretical or a physical
model. Here, we limit ourselves to the simulation of
petroleum reservoirs. Our concern is the development
and use of models that describe the reservoir perform-
ance under various operating conditions.
Reservoir simulation itself is not really new. Engi-
neers have long used mathematical models in per-
forming reservoir engineering calculations. Before the
development of modem digital computers, however,
the models were relatively simple. For example, when
calculating the oil in place volumetrically, the engi-
neer simulated the reservoir by a simple model in
which average values for the porosity, saturation, and
thickness were used.
Although simulation in the petroleum industry is
not new, the new aspects are that more detailed reser-
voir features, and thus more accurate simulations,
have become practical because of the capability af-
forded by the computers now available. The more de-
tailed description, however, requires complex mathe-
matical expressions that are difficult to understand,
and this difiicuhy has caused some engineers to shy
away from using simulators, and others to misuse
them.
We in the petroleum industry are in the reservoir
simulation revolution. As time goes on, simulators will
be used more and more, so a basic understanding of
. ..
reservolr modelmg is esseiitia,. “ ‘ The engineer! espe-
cially, must become competent in setting up simula-
tion problems, in deciding on appropriate input data,
and in evaluating the results.
. 7-Z3t-
Basic Analysis
If a reservoir is fairly homogeneous, average values
of the reservoir properties, such as porosity, are ade-
quate to describe it. The average pressure, time, and
production behavior of such a reservoir under a solu-
. .
tion gas drive, for ---+
~Acwy.-,
nre
--
MNTI-MUycalculated
by the familiar methods’ of Tamer, Muskat, or Tracy.
All of these methods use the material balance equa-
tion normally referred to as the MBE. A simple ex-
pression for the oil MBE is the following
(cumulative net withdrawal in STB) = (original
oil in place in STB) — (oil remaining in place
in STB)
The cumulative net withdrawal is the difference be-
tween the oil that leaves the reservoir and the oil that
enters it. In this basic analysis, there is no oil entering
the reservoir since the boundaries are considered im-
permeable to flow. Thus, the MBE reduces to its
simplest form. Such a reservoir model is called the
tank model (Fig. 1). It is zero dimensional because
rock, fluid properties, and pressure values do not vary
from point to point. Instead, they are calculated as
average values for the whole reservoir. This tank
model is the basic building block of reservoir simu-
lators.
Now let us consider a reservoir represented by a
sandbar. Let the two halves of the sandbar vary in
lithology. The sandbar as a whole cannot be repre-
sented by average properties, but each half can. Thus,
the sandbar consists of two tank units, or cells, as they
are normally called. The MBE describes the fluid
behavior in each ceU as in the previous tank model.
.,M. . . . ~] te~ of the MBE is
However, the net wl,,,d,~.v-. .+...-
more complicated because there can be migration of
fluid from one cell to another, depending on the aver-
1383
age pressure values of the two cells. This fluid transfer
between the two cells is calculated by Darcy’s law.
The MBE, together with Darcy’s law, describes the
behavior of each cell. This model is not a zero-dimen-
sional reservoir simulator since reservoir parameters
may vary between the two cells. Instead it is a one-
dimensional model, because it consists of more than
one cell in one direction and of only one cell in the
other two directions (Fig. 2).
This analysis can be extended to reservoirs where
properties as well as pressure values vary in two
dimensions, and to others where the variation occurs
in three dimensions. The simulators representing these
reservoirs are called, respectively, two-dimensional
and three-dimensional simulators, as illustrated in
Figs. 3 and 4. Thus, a two-dimensional reservoir simu-
lator consists of more than one cell in two dimensions
and of one cell in the thhxi dimension. And a three-
dimensional simulator consists of more than one cell
in all of the three dimensions.
Regardless of the number of dimensions used, the
MBE is the basic equation describing the fluid be-
havior within a cell; and Darcy’s law describes the
interaction between the cells. In one-, two-, and three-
dimensional models each cell, except the boundary
cell, interacts respectively with 2, 4, and 6 cells. Since
a simulator can consist of hundreds of cells, keeping
account of the MBE for each cell is a formidable
bookkeeping operation ideally suited to digital com-
~utation. But we emphasize once again that the prin-
ciples and equations used in reservoir simulation are
not new. They only appear so because of the complex-
ity of the bookkeeping.
Types of Reservoir Skmihiiim
There are several types of reservoir simulators. Choice
of the proper simulator to represent a particular res-
ervoir requires an understanding of the reservoir and
a careful examination of the data available. A model
that fits Reservoir A may not be appropriate for Res-
ervoir B, in spite of apparent similarities between
Reservoirs A and B. A reservoir model is useful only
when it fits the field case.
One basis for classif@g models, as discussed ear-
lier, is the number of dimensions. The two-dimen-
sional model is the most commordy used. There are
/
Fig. l—Tsmk model.
several two-dimensional geometries, the most popu-
lar of which is the horizontal (x-y) geometry; but the
vertical (x-z) and the radial (r-z) geometries are also
used quite often.
Simulators can be classified also according to the
type of reservoir or process they are intended to simu-
late. There are, for example, gas, black oil, gas con-
densate, end miscible displacement reservoir simulat-
ors. Moreover, there are one-, two- and three-phase
reservoir models. Furthermore, any of these simula-
tors may or may not account for gravitational or
capillary forces. It is not enough to choose the proper
simulator with respect to dirnensionality; the simu-
lator must represent the type of hydrocarbon and the
fluid phases present.
Simulation Steps
Preparation of Data
After the type of model to use in a study has been
selected, the next step is to divide the reservoir into a
number of cells, as illustrated in Figs. 2 through 4.
This is accomplished by laying out a grid system for
the reservoir. In a two-dimensional study, the grid is
established by drawing lines on a map of the reser-
voir. All grid lines must extend across the reservoir.
Each cell is identified by its x, y, z coordinates. Then
the flow conditions around the perimeter of the res-
ervoir are established. Normally the reservoir bound-
ary is considered sealed, but influx or efflux at an
assigned pressure or rate may also be specified.
m. .
. .. -following
1rie next s’fip is to ass:gn the - for each
celk rock properties, geometry, initial fluid distribu-
tion, and fluid properties. The rock properties consist
of specific permeability, porosity, relative permeabili-
Cj ~d ~~m~e&~.e$~Apillary pressure. The Cell geome-
try includes the depth, thickness and locations of
wells. Usually the wells are assumed to be located at
the centers of ‘&e cells in which *Aeyfal!a The initial
fluid distribution consists of the oil, water and gas
saturations at the beginning of simulation. Also, the
average pressure of the cell at that time is assigned or
calculated from known data. Fluid properties are
specitied by the usual PVT data. In addition, for each
well it is necessary to provide a production schedule
and a productivity index or a skin value (i.e., damage
or ,improvement).
Flow
5TY?l--
Fig. 24ne-dimensional simulator.
Flow
Fig. 3-Two-dimensional simulator.
JOURNAL OF PETROLEUM TECHNOLOGY
 The engineer should scrutinize carefully these basic
data for consistency and accuracy. For example, if
pressure buildup data are available on a well, the
permeability-thickness product of the cell where the
well is located and the flow rate assigned to the well,
should be compatible with the buildup data. The time
spent in examining the basic data is well spent, for it
can lead to fewer simulation runs. Moreover one must
always remember that the answer is only as good as
the input data.
History Matchfng and Performance Predktion
The rn-fi ~urpose of reservoir simulation is to prediCt
the rate of hydrocarbon recovery for dtierent meth-
.*
ods of field operation. d adeq-ua .-.-LG CI,.J
~eiu
A+. ex!st~
-a~
reasonably accurate performance predictions can be
made. If data are incomplete or suspect, simulators
may be used only to compare semi-quantitatively the
results of dtierent ways of operating the reservoir. In
either case, the accuracy of the simulator can be imp-
roved by history matching.
The first step in a history match is to calculate res-
ervoir performance using the best data available. The
results are compared with the field recorded histories
of the wells. If the agreement is not satisfactory, such
data as permeability, relative permeabiUty, and po-
ro@ are va~~ed from one computer run to another
until a match is achieved. The s&ndator is then used
to predict performance for alternative plans of oper-
ating the reservoir.
The behavior of the reservoir is influenced by many
factors — permeability, porosity, thickness, satura-
tion distributions, relative permeability, etc. — that
are never known precisely aU over the reservoir. What
the engineer arrives at is only a combination of these
variables, which results in a match. ‘T’is colmbiiation
is not unique, so it may not represent precisely the
condition of the reservoir. When the simulator, after
a match, is used to predict, it is not certain that the
physical picture of the reservoir described in the simu-
lator will give predictions sufficiently close to the ac-
tual reservoir performance. In generaI, the longer the
matched history period, the more reliable the pre-
dicted performance wiU be. It behooves the engineer
to monitor periodically the predicted vs the actual
performance and to update his physioal picture of
Flow
Fig. 4-Three-dimensional simulator.
the reservoir.
Mathematical Considerations
Derivation of Equations
For the engineer to adequately understand reservoir
simulation, he should be acquainted with the equa-
tions used. These are basicaUy material balances about
cells for each phase, and Darcy’s law, which describes
the interactions between cells. For illustration, we
derive here the fundamental equations for a black oil
system. and exp!ain thei: physical significance.
For the sake of simphclty, consider a cdl in a one-
dimensional reservoir simulator, as shown in Fig. 5.
Tine sane mia!jjsis is applicable to a cell in two- and
th.me-dimensional models. (An expression for the oil
material balance of the cell was given eariier.)
(Oil volume entering the ceU during a time incre-
ment At, in STB) minus (oil volume leaving the ceU
during the same time increment, in STB) equals (the
change in oil volume in the ceU, in STB).
Volume of oil entering the cell during At, in STB,
equals QinAt.
Volume of oil leaving the cell during At, in STB,
equals (At + dA~.
Change in volume of oil in the cell during At, in
STB, equals
where Qi. is the average flOWmte of ofl into the cell
during At in STB/unit time, Q..t is the avenge flow
rate of oil leaving the cell to its neighbors during At
in STB/unit time, and QOis the oil production rate
from the cell, if it contains a well, in STB/unit time;
AxAyh@.
represents the volume of oil in the cell at
l?.
any time, n+ 1 refers to the end of the time step, and
n to the beginning.
Substitution in the oil MBE, after dividing through
by At, gives
Qin – Qcmt – go = %w’)w%)rnl
.,. . . . . . . . . . (1)
However, by Drircy’s&.w, assuming the flaw to be
Qout
Qin
Fig. 5-Cell in a one-dimensional simulator.
. 1385
from left to right as shown in Fig. 5, average value for @o—i.e., at the (n+ %2 )-time level—
while the forward difference method uses @. at the
beginning of the time step — i.e., at the n-time level.
The implicit method is the most stable of the three.
and The time at which hkJBop~ is evaluated was left un-
specified. Most authors use the n-time level, but some
use the (n+ 1)-time level. This will be discussed later.
Similar derivations can be made for the water and
gas. The water and gas equations in vector notation
where Ayh is the cross-sectional area of the cell, Ax
form are, respectively,
is the length of the cell, 00 is the flow potential in the
oil phase, i refers to the cell of interest, 1—1 refers to
the left-hand neighbor, and i+ 1 refers to the right-
hand neighbor. The flow potential @oequals pressure
and
plus capillary pressure plus gravitational potential,
and its Id$eat the (n+ 1)-time level is explained later.
Substituting Eq. 2 in Eq. 1 and dividing through by v ‘ (Afv@g) + v “ O&AJv%)
AyAx gives
1 hko *“+’ + V ● (R,,oA,oV@t.) – *Y
——
[(
Ax p.. B.
Oi-1 –*F

Ax )
.— hko
/.LO
Ba

where in Eq. 6 the gas dissolved in the oil and water

is accounted for.
Eqs. 4b, 5, and 6 are the MB eqiaaticms for three-
phase immiscible flow in a black oil system, and were
Eq. 3 is rearranged to give derived by Muskat.3 Written in difference form, these
are virtually the only equations used in the most com-
mon type of simulation, that of a black oil reservoir.

@y
—
@ —— Method of Solution

( Ax
%
)1 AxAy
qo
Eq. 4 and its comparable forms for the water and gas
give the relationships, for each cell, among pressure;
S*[(*)”+l-(*)”] , ~ (4) and oil, water, and gas saturations; and time. If there
.- are m cells, then we have m equations for each phase,
giving a tdai of 3m equations. The solution of these
where A = * , and the subscripts i+% and i —1% equations is the major chore of reservoir simulation.
~B
Two methods of solution are generally used; these
indicate that the quantity is evaluated as an average are the implicit-implicit” 3 and the implicit-explicit.”
for the (i+ 1, i) and (i, i – 1) cells, respectively. DiEer- They are similar in one respect. Given a value for the
ent investigators use different averaging techniques. saturations and pressure at each cell at the beginning
The upstream value for A is the most commonly used. of a time step, new saturations and pressure values
Eq. 4 is the oil mass balance equation in one di- are found at the end of the time step. These values in
mension, in difference form, which is used in the simu- turn represent the starting point for the next time step.
lation calculations. In two and three dimensions, y- This stepwise process is continued until the desired
and z-&ection terms identical with the x-direction amount of time elapses.
term are added. The implicit-implicit”’ method solves Eqs. 4 and
Eq. 4 may be written in differential form as the difference forms of Eqs. 5 and 6 directly. The
solution usually involves an iterative procedure. Cap-
a+. qo _ a +JW’
& Ao~
() –—–—
AxAy 2t ()
—
B.
. . (4a) illary pressure occasionally tames instability prob-
lems.* The implicit-implicit method overcomes the
problem by expressing saturation as a function of cap-
and in vector notation as
illary pressure. To start the calculations, values of
saturations are assumed, and the pressures in the oil,
.~,ater, ~d gas phasesarecalculated. These calculated

These three forms of the MBE are used inter-
changeably in the literature. Because of its compact-
ness, Eq. 4b is the most commonly used.
In deriving Eq. 4 we used the value of % at the
(n+ 1)-time level, i.e., at the end of the time step.
This diflerencing technique is called the implicit2 or
backward difference method and is the most com-
monly used. The Crank-Nicholson methodz uses an
pressures result in new capillary pressures, which are
used to calculate saturations. These are compared
with the assumed values, and if necessary the calcula-
tions are repeated.
Using the fact that the oil, water, and gas satura-
tions add up to one, we can manipulate the three MB
“For definition, refer to the section on Computetionel Consid-
eration.

JOURNAL OF PETROLEUM TECHNOLOGY

. .
equations in such a way as to result in a pressure
equation. The pressure equation in symbolic form
and vector notation is
Zp
v “ k,up –
q’=c’-z ‘ “ “ “ “
(7)
.&,KeKe).~ is the total effective mobility of the three
phases, q’ is the total production, and c’ is the total
effective compressibili~. (Capillary and gravity forces
have been neglected.) In the implicit-explicit’ method,
at any given time, the pressure equation @q. 7) is
solved first, giving the pressure distribution at each
cell. Then the saturations are determined from the
solution of the three MB equations.
To illustrate the method of solving Eq. 7 we write
it in difference form in one dimension. We also assume
that k’ = c’ = 1, and that q* = O. The implicit dif-
ference formulation’ is
where i, i — 1, and i+ 1 refer to the cell of interest and
its two neighbors, and n refers to the time level. Eq.
8 gives P ~“, the pressure tO be dete~ed> as a finc-
tion of two unknowns PY1 and P U. TM We cannot
solve for p ~1 with this equation alone. For this reason
we call this an implicit equation in pressure. However,
similar equations can be written for all cells, resulting
in m equations with m unknowns.
Several methods have been devised to solve the m
pressure equations. The simplest are tbe relaxation
techniques.’ The pressures at i– 1 and i+ 1 are as-
sumed, and the pressure p~l is calculated. This tnal-
and-error process is repeated at each point in turn
until a sufficiently accurate solution to the m equa-
tions is found. Given this pressure solution, we then
solve explicitly for the saturations, using the three
MB equations.
The coeflkients ~’ and CTcontain effeCtiVeperme-
~~i~te$, @-e~i~e~3 and fo~ation volume factors, SO
they are functions of saturation and pressure. Until
now we have ignored this fact. However, if we want
to account for i.hs“ ~eYwL,uw..-
‘ --”dI=n”Yy which is sometimes
necessary due to instability, then an iterative method
is used. The method is summarized by the following
steps, which are symbolically correct. In actuality the
calculations are more involved. 7
1. Begin with known pressure and saturation dis-
tribution at time n and, using the pressure equation,
and the values for ATand CTcalculated from the sat-
uration distribution and the pressure values at time
n, solve for pressures at time n+ 1.
2. Solve for the saturation diSifibiitiOtl at time n + 1
using the three MB equations.
3. Using the new saturations and the pressures cal-
culated in Step 1, recalculate & and cT vaiues.
4. Repeat Steps 1 through 3 until the convergence
criterion is achieved. In repeating Step 1, the values
Of& and c’ Of Step 3 are used.
5. Proceed with the next time step.
Methods that cycle between pressure and satura-
tion equations-are called fully implicit or iterative,’
whereas those that do not, and in which essentially
only Steps 1 and 2 and then 5 are executed, are called
mixed. 7 Mixed methods are extensively used because
they require less computer time.
One criterion for dete rmining the compatibility of
the pressure and saturation values is the material bal-
ance error. One form of the material balance is the
summation of the stock tank oil at the beginnimg and
at tiie end of the time step. The ditlerence between
the values should be equal to the totai production dtii-
ing the time step. The incremental error is calculated
by the following equation.
incremental MBE error =
where V is the volume of the cell and the summation
is taken over the m ceils.
Some authors use cumulative MBE error, which is
given by the following equation.
cumulative MBE error =
initial oil in place
-! [v’@ln:_l.
cumulative total production
A low value for MBE error is a necessary but not
a sufficient criterion for a correct solution. In essence,
low error indicates that the total oil in the reservoir at
time n+ 1 is correct, but it does not guarantee that
the oil is distributed properly.
Computational Considerations
Computing Time
For a given computer, the time required for a par-
ticular reservoir simulation depends primarily upon
(1) the number of cells, and (2) the number of time
steps.
The computing time required for a time step is
proportional to the number of mesh points. Doubling
the number of mesh points approximately cioutiles the
computer time per time step.
The ~ti,m&r Of time stepsrequired to simulate an
~~assigned number of years dep-nds on the allowed
length of the time step At. The maximum value At
may take is a function of the volume and shape of the
cell. In a two-dimensional horizontal model, for ex-
ample, the cell volume is AXAy times thickness, and
the shape is given by the ratio of Ax/Ay, where Ax
and Ay are the horizontal dimensions of the cell. The
allowed time step decreases as AX and Ay decrease,
IAX
and as 1’ incieases. For example, the ‘d-
Fy –
Iowed At in a simulation study in which AX = Ay =
300 ft will be about four times the Al if AX = Ay =
150 ft.
Instability
Numerical techniques do not yield exact solutions.
There is an error associated with the answers. This
error sometimes grows very rapidly, causing the solu-
1387

tion to “blow up”; in other words, the solutions be-
come ph ysicaiiy ‘-mrea!istic. The most common cause
of this instability is excessively large changes in sat-
urations and pressures during the time step. Usually
this may be remedied by reducing the size of the
time step.
Nnmerfctd Dispersion
This is an inherent property of digital simulation. It
is due to the representation of the reservoir by cells in
which properties are averaged. When a saturation
front enters the ceil, it is S~ir3Xl out over the cell to
arrive at average satmdicm values. Numerical disper-
sion can be minimized by decreasing the dimensions
of the cells. However, this leads to increased com-
puter time.
Validity of Solution
Once a simulation run has been made, the question
ansex “How good is the solution?” Small MBE error
indkates that the total fluid volumes are correct, but
does not guarantee that the fluid distribution is valid.
If the resulting fluid distribution is questionable, a
systematic analysis is needed. Variables that influence
the saturation distribution are the time step size At,
and the cell dimensions ~ and Ay. For a correct
mathematical analysis, the sensitivity of the results to
ti, Ay, and At should be examined. A change in Ax,
Ay values may require a major revision of the data,
which is not practical. A common practice is to study
Norriseal 3-way
2-position
motor valves.
. .
only the effect of the time-step size. This is done by
rerunning the simulation with reduced time steps and
compa~ng ~~e ~e~u!t~.The time stepisreduced until
further reduction does not change the results signifi-
cantly, thus indicating that the best solution has been
obtained for the chosen sizes of the cells.
Acknowledgment
I should liie to thank G. L. Smitt, J. W. Watts and
J. E. Walraven for helpful comments, and Mobil Re-
search & Development Corp. for permission to pub-
lish this paper.
References
1.Craft, B. C. and Hawkins, M. F., Jr.: Applied Petroleum
Reservoir Engineering, Prentice-HallInc., EnglewoodCliffS,
N. J. (1959).
2. Smith, G. D.: Numerical Solution of Partial Differential
Equations, Oxford U. Press, Inc., New York (1965).
3. Muskat, M.: Physical Principles of Oil Production, Mc-
Graw-Hill Book Co., Inc., New York (1949 ).
4. ~hss, Jim, Jr., Peacemen, D. W. and Rachford, H. H.,
“ “A Method for Calculating Multi-Dimensional lm-
~ncible Displacement”, Trans., AIME ( 1959) 216, 297-
5. ~oa~, K. H., Nielsen, R. L., Terhune, M. H. and Weber,
“ “Simulation of Three-Dimensional, Two-Phase
Fiow ;n Oil and Gas Reservoirs”, Sot. Pet. Eng. J. (Dec.,
1967) 377-388.
6. Fagin, R. G. and Stewart, C. H., Jr.: “A New .Approach
to the Two-Dimensional Multiphase Reservor Sumdator”,
Sot. Pet. Eng. J. (June, 1966) 175-182.
7. Blair, P. M. and Weinaug, C. F.: “Solution of. Two-Phase
Flow Problems Using Implicit Difference Equations”, paper
SPE 2185 presented at SPE 43rd Annual Fall Meeting,
Houston, Tex., Sept.29-Ott. 2, 1968. JPT
for controlled,
high pressure automation!
Series 3023 with
working pressures up to 3600 psig;
temperatures from -20° to 180° F.
a
Available in 1“, 2“ and 3“
sizes, the Norriseai
Series 3023 will he.ndle
metering, blending or diverting
for “tough”’ high
differential end static
pressure applications. Cast
~-
steel body with hammer- —
nut top closure .. . stainless “~ ,
steel plugs and seats ...
selection of hard
or soft cagegu ided trims . .. wide
range of seals.
Write for Bulletin 3-5 7-69. I
P. O.Box 9528, Houston, Taxss 77011 -
(713) 928-2701
JOURNAL OF PETROLEUM TECHNOLOGY