Professional Documents
Culture Documents
E3304
‘@
Reservoir Simulation Models–
An Engineering Overview
H. M. Staggs, SPE-AIME,Atlantic RichfieldCo.
E. F. Herbeck, SPE-AIME,Atlantic RichfieldCo.
Introduction
The multicell reservoir simulation models have
Basic Model Considerations
reached that stage of development at which they are Defining a Model
being passed from the hands of the scientists and Stated simply, a multicell model simulates fluid flow
mathematicians who originated them into the hands in an oil or gas reservoir. Models cannot describe
of the reservoir engineer for his everyday use. Whereas flow exactly as it occurs, but they do give valid ap-
numerous papers have been published describing the proximations. The mathematics of these models
theoretical aspects and mathematical approaches de- requires that the reservoir be treated as if it were
veloped for these models, this paper is intended as a composed of many individual segments. These seg-
look at multicell models from the engineer’s rather ments are usually called cells, but are also referred
than the scientist’s point of view. to as grids, nodes, mesh points, or a network. Models
Engineers are quickly realizing that these models are made to represent reservoir fluid flow from cell
are the best tool developed in recent years for under- to cell in one, two, or three dimensions and accord-
standing oil and gas reservoirs and predicting their ingly are termed one-dimensional, two-dimensional,
~~,.”,,,.-,,-.
m..f,w.ln .Tlf.r=
Mn~~l;mg
1.. ”..1.1.. a reservoir hy dividing it or three-dimensional models. These are illustrated in
into cells provides a flexibility to the engineer that Figs. 1 through 3.
he never had before. High-speed computers permit Each cell is assigned its specific reservoir properties
multiple runs of a reservoir model to test different of size, porosity, permeability, elevation, pressure,
methods of field operations or to check the sensitivity and fluid saturations. In addition to cell properties,
of reservoir behavior to unknown rock or fluid well data must also be provided. These include loca-
properties. .:--
UUII,
-..-.4..,.+:.;+.,
pl uuu&L1 Vlcy
;nAmv rlaci .c.A
AIIUS,A , “w..’ w=
nrnrl~~&f
~. “...
Qr j~je~-
The multicell models have many valid applications, tion rates, and limiting conditions such as economic
but they can also be misused because the model is limit, maximum water cut and GOR, and minimum
only providing answers that the input data are forcing bottom-hole flowing pressure. General fluid and rock
it to provide. Therefore, a careful analysis and selec- data must also be provided for the entire field or sec-
tion of input data is imperative. The engineer can tion of field being studied. These usually include PVT
usually obtain representative data on reservoir rock data for the oil, gas, and water; rock compressiiiility;
and fluid properties. The greater difficulty lies in and relative permeability for each flowing phase.
properly selecting such items as relative permeability, Models using PVT data for oil, gas, and water are
vertical permeability, and cell size. frequently referred to as “black-oil models.” Special
For the engineer with limited experience in using models, these case histories can be
helpful. They show that difficulties can arise with cross-section models and with improper
relative permeability data and that it’s easy to get carried away with your own jaked-in
input data. But they also show that, properly used, models can be a godsend.
lA?$l
JOURNAL OF PETROLEUM TECHNOLOGY
“gasmodels” have been developed for the study of 4. Calculate the flow between cells during the time
gas reservoirs. In some models the hydrocarbons are step and the new saturation for each cell.
divided into components, in which case the term 5. Set a new time step and repeat the process until
“compositional model” is used. The reservoir engineer the model has calculated performance for the desired
will rarely use a compositional model because it is total time. High-speed computers are required for
more complex than is generally necessary. solving problems involved in a multicell model be-
A system of mathematical equations is used to cal- cause of the extremely large number of calculations
culate the flow between cells and the fluid saturation made in a typical study.
and volumes in each cell. The equations used are
derived from: (1) the continuity equation or mass The Need for Engineering Judgment
balance, (2) Darcy’s law of flow through porous Certain judgment decisions must be made by the engi-
media, and (3) equation of state. neer before and during modei runs. Tine engineer
Finite-difference methods are used to solve the must match the solution that is required with the most
model equations. The methods do not give exact an- economical model that will do the job. The more
swers and are recognized as having some inherent complex the model, the longer the time required to
error.l’ z The reservoir engineer sbouici be aware of Ccmpiete t}le SLUUy. T+,.
.-4.. -I..
ILGmIS
“ .3. -L
3W11
““ .x-L.* fif A“mian -
am nuliluw. “1 “L.w.l
these potential errors, even though for the most part sions and cells strongly influence running time. It is
he will not be too concerned with them because the quite often necessary to make preliminary model runs
mathematical approaches in modeling are still gen- to determine what dimensions and cell count will
erally more accurate than the engineer’s ability to ultimately be required.
refine the. input data, In multicell modeling, representative basic data are
indispensable. If little basic data are available, the
How Models Work engineer must decide whether more should be gath-
A simplfied calculation procedure used by a model ered — whether, for example, wells should be drilled
would be as follows: to take fresh cores for analysis or to obtain such basic
1. Start with the cells’ having certain initial satura- data as a PVT analysis, The engineer working on a
tions and conditions given. field study must be careful to avoid the overkill that
2. Select a time step over which the model is to results from using expensive, time-consuming simu-
calculate. (Initial time steps are usually short periods lator methods on va~e, hastily prepared basic data.
of about a day, but with successive steps the time Engineering judgment is also required in selecting
periods can gradually be extended until they cover the proper relative permeability data because it repre-
some months.) sents the flow mechanism that occurs in the reservoir.
3. Calculate or use an assigned production or in- This often requires the formulation of pseudo relative
jection volume for each well for the time period. permeability data that represents not only displace-
6!!!!!3
HORIZONTAL VERTICAL TILTED RADIAL
Fig. l-One-dimensional models.
Zzz!2e’”
/
um recovery by determining locations of ‘wells in ,/’
reservoirs with varying rock properties. The two-
dimensional areal models have also been valuable in
determining aquifer influx for reservoir configurations
that have no analytical solution. / /
Another use f& two-dimensional radial models is
T I 1 1 I’–c-’it’
m bottotti-iiOie pi~SSLif3 03H.D) a.lw,j..
-hQis . RV
-J .mafch-
ing a well’s BHP response with a model, the BHP *I
~/ /
pe-tiormance can be interpreted more accurately. I
I I I I
Threedimensional models are the present ultimate
in the engineer’s reservoir simulation library. This Fig. 3—Three-dimensional model.
m
are that (1) a larger number of cells will be used to
t- ,- 90 cover the area; and (2) smaller cells may dictate
1
1 shorter time steps because large saturation and pres-
: I sure changes in a time step will cause some models
+- k-fi~m- .,”.+ hb omcl CAW- llnr~alictic an~w~-~o -A.
‘4
‘A
I
i LV UGWW LUID .a”Aw LU.U &. w u... -..u..- -..
I
■
I crease in the number of calculations that must be
A. --m md -J made and in excessive requirements of computer time.
Eli Our experience has been that 1,000 to 1,500 cells are
5X5
:~~ a reasonable number to use, but that a system of
2,000 to 2,500 is approaching the practicai iimit.
-- ACTUAL 5-SPOT ELEMENT OF 5 ACRES This limit will, of course, be relaxed as higher-speed
PRODUCING WELL computers are developed.
: INJECTION WELL
To simulate ~e sweepout of a reservoir undergoing
Fig. 4—Grid systems used to study influence of
grid size on model calculations.
an injection program, one should use a grid system
that has at least two cells between offsetting produc-
tion and injection wells.
Relative Permeability
The key to good results from a multicell model study
is good data. Perhaps the most critical are the relative
permeability relationships. It should be realized that
in many cases the choice of a certain relative permea-
bility relationship will essentially fix the answers the
model determines. Generally, the concept associated
with relative permeabili~ curves is that they repre-
sent fluid movement and displacement in a uniform
3 -o I 2 3 4 5 6 7
section of reservoir rock. However, in model work,
TIME-YEARS
relative permeability must represent this displacement
Fig. 5—Comparison of calculated producing rates for in a reservoir volume of cell size. It therefore has to
a five-spot element using varying grid systems. account for fingering, stratification, and inbomo-
Reservoir A, one of the first fields we studied with a based to some degree on laboratory work. The water-
multicell model, brought to our attention the im- oil relationship is plotted against fraction of total
portance of relative permeability data. pore volume, whereas the gas-oil relationship is
Reservoir A produces from a 7,500-ft lime. The plotted against fraction of hydrocarbon pore volume.
crude is of good quality — 35 ‘API gravity. Original Model runs were made, first to match history.
reservoir pressure was 3,500 psi, and the bubble- These were followed by projection runs to simuiate
point pressure was about 50 percent of original. The continued alternate gas-water injection and then to
pressure had been permitted to drop to near the simulate water injection only. Under primary per-
bubble point before a secondary recovery project was formance the history matched; but we had consider-
begun. The secondary program was one of alternating .1.1.
aolt?
J.mm
u~m~ul[y
1. . . .— .:--
Lrylllg
.-
LU
—..4-I.
lJJIJIW
k: -.,....,
lJISLUIY
4+-..
&LC1
:..:n,.+:n*
lll~cuJull
gas and water injection. One of the purposes of the started, and the relative permeability curves had to
multicell model study was to determine if the gas- be revised. The two projection runs were made using
water injection program should continue or if an in- the set of curves that then gave the closest match.
Results revealed that slightly higher recovery could
be expected for alternate gas-water injection than for
water alone.
tii
In analyzing the results of this study several things
came to light. In this reservoir study three different
/ - // + recovery mechanisms were involved: primary recov-
.’<”
I I I I I I I
ery by Oii expansion, secondary recovery by alterna-
ting gas-water injection, and seconda~ recovery by
water injection only. During the primary recovery
period only one phase — oil — was flowing over most
of the model area during most of the time period. A
history match of this period was easy to obtain.
Actually it would have been difficult not to obtain a
match with any reasonable relative permeability curve
as long as the k,o at the comate water saturation was
approximately correct. The fact that the same relative
permeability curves did not hold for the gas-water
/ :-: --.: - -.4..L ..L.-...l.-l..-+ L..,o ha=m+,-,
UIJCU UOJJIIMWII MKJ LUU lW L ua VG UW,U LUO ,I”av?met,d
WIUAFUWWM.
II The match of reservoir performance that was obtained
for one producing mechanism did not validate the
relative permeability curves for another producing
I?K!c!mlisrn.
An examination of the output data from the pro-
jection runs showed that under alternate gas-water
:-:--.:-..
lIIJGU
● L.- an- ..,-.
LIU1l,
“A;”- ;“+~ al,.+”,-,” ~n the nil
UIG &J3 W c13 &J1ll& 111 LU aUlu LIULL I*I u,- WM,
a
wATER 5ATURb.TION
% TOTAL PORE VOLUME
GkS SATURATION
% HYDROCARBONPORE VOLUME
could expect not only gas swelling effects, but also the
effect that the presence of gas would have on the
residual oil saturation to water injection. The presence
of free gas could possibly reduce the residual oil satu-
rations.4 The extent to which this would take place
Fig. 8-Relative permeability curves developed for
model study of Reservoir A. would have to be determined in the laboratory by
Case History 6
1 1 1 11 [111
1 1 II I
1
HISTORY PREDICTED
H, SJOIW I PRED:TED
I
5000 16
L
14
A @ = SMALL AQUIFER
4000 @ = STRONG INFLUX
@) = LARGE AQUIFER
b @ = MODEL PREDICTION @) = WATERFLOOD
3000
\ .-
-g d -“m
~A \ Yb
2000
~~ \
P
$ ti’ -
1000
)
2
I
0 4 8 12 16 20 24
I
0
RECOVERY- % OF ORIGINAL-OIL- IN- PLACE 12345676 9101112
varying degrees of water influx for Reservoir D, Fig. 12—Predicted oil producing rates for Reservoir D.
m“nm. ,”r, , <071
1435
in field development. Reservoir engineers must dem-
Overview onstrate to management that these new tools are
Experience has shown that multicell reservoir simu- worth their enormous costs.
lators are a vital tool for the reservoir engineer. This brings to mind another side of modeling,
Models supply and solve a set of equations descri- which has been described as field-study overkill, or
bingfluid flow, but it is up to the engineer to enter the application of too much engineering time and
into the model the data that satisfactorily describe the computer time to field problems that do not justify
system being studied. No statement so well describes the expense. The engineer should apply the simplest
reservom model computer runs as .~- LUG
<+-rail .U- orn term available solution to his problem. For example, for a
“Garbage in — Garbage out.” The reservoir engineer simple waterflood caicuiation, a tllree-dhnensiomd
must examine all the output and apply his judgment model is seldom necessary. There are many good
to the results. If a model run gives answers that are techniques besides multicell models for waterflood
inconsistent with experience in similar cases, there is forecasting that are faster and give excellent results.
a good chance something is wrong. It is not a satis- The engineer should ask himself:
factory response to management to say “I don’t 1. What is the problem I need to solve?
understand it, but this is what the model calculated.” 2. Will the model answer the question or will it
I@gineers -k “ld spend the maiority of their time
~lloul only tell me what I am forcing it to tell me by the
analyzing results of runs, making ;ure the calculations input data?
done by the computer are consistent with flow theory. 3. Even if the multicell model can give me the
The sophisticated simulator can never replace a answer, is there a simpler way of getting it?
sound geological study of the reservoir and rock
propefi;es or-a laboratory study of the relative per- References
meability characteristics of the rock. I$4ul$icel!models !. Cmts K. H.: “Use and Misuse of Reservoir Simulation
enable the engineer to test the effect of variance in Models,” J. Pet. Tech. (Nov., i%9j i39i-i39t3.
the input data and inform management of the risks 2. van Poollen, H. K., Bixel~ H. C. and Jargon, J. R.: “Res-
ervoir Modeling — 3: Finite Differences,” Oil and Gas
and ranges of producing conditions to be encountered J. (Sept. 15, 1969) 120-121.
3. Caudle, B. H. and Witte, M. D.: “Production Potential
OriRinal manuscript received in Societv of Petroleum Encineere
office-March 24, 197”1. Revised manuscript received Aug. 19; 1971.
Changes During Sweepout in a Five-Spot System,” J. Pet.
Paper (SPE 3304) waa presented at SPE Permian Basin 011 Re- Tech. (Dec., 1959 ) 63-65.
covery Conference, held in Midland, Tex,, May 6-7, 1971. @) COPY 4. Schneider, F. N. and Owens, W. W,: “Sandstone and
right 1971 American Institute of Mining, Metallurgical, and Carbonate Three-Phase Relative Permeability Charac-
Petroleum Enginears, Inc.
teristics,” Sot. Pet. Eng. J. (March, 1970) 75-84. JPT