Integration and Gridding
Introduction
Putting the simulation together was reserved deliber- Integration
ately for this later point in Part |, There are a number of At this stage, you should have a good idea of where
steps at this stage. the study is heading, Normally, the most critical aspects,
+ Until all ofthe input has been evaluated, the are the geological and production performance review
model cannot be finalized, Te key point to AlLof the data should have been screened, If some of the
emphasize here isto keep an open mind at this data is weak or unavailable, part of the study may be to
stage. The study scope can change significantly _ determine the influence of the assumptions. This is a piv
depending on the data available ‘otal point, which the author calls the integration phase.
Frequently, material balance or other ancillary The result of this phase will be a complete data set.
studies are conducted as well. These are dis- Gridding is normally determined at this point, Each ofthe
cussed briefly Sometimes they provide useful preceding four points will be discussed.
insight, and, at other times.the results can be mis-
leading or ambiguous.
Itis also important to review the original objec
tives in terms of the data reviewed and the latest
understanding of the reservoir
‘The final step in this chapter concerns creating
a grid.Practical Reservoir Simulation
Data
Missing data
Io a number of cases, individuals felt studies should
fot be conducted because some element of data sas
missing In paricular, this seems to apply to relative per-
meabilty data, Although this is undoubtediy important
data, the wath is that limited daa is not much better than
no data, since there can be alot of variation
In one study for retrograde condensate systems, there
was no relative permeability data available at all. Using
sensitivities, the author demonstrated the relative per
formance between fraced and unfraced wells was similar
regardless of the relative permeability data assumed. This
is a serious consideration since, even for Jow levels of
condensate dropout, a frac can significantly reduce pro-
ductivity impairment caused by retrograde liquid con:
densation. The simulation could not predict the absolute
fevel of production (test data did), but it could indicate
the direction and trends. Restated, an engineering solu
tion and a scientific solution are not the same. It may be
sufficient to put an upper and lower bound on an answer
rather than provide a precise solution,
Scientific results versus pragmatic results
[rote a paper on the study described previousiy. The
paper was severely criticized due to the “Naky” nature of
the data, which relied on general gasoil relative perme-
ability correlations developed by Arps As a practicing
resenoir engineer, | was horrified by this response. The
client had spent considerable money on exploration and
needed to make an expensive nomination for a sour gas
plant on short notice. A nasanswer was untenable. They
needed a reasoned, educated guess. To put the simula-
tion in perspective, the land and wells were drilled based
‘on substantially higher levels of risk
[An engineer's job will sometimes be to give the boss-
es or clients the best answer possible with the data avail-
able, At such a point, they need a risk assessment with
the answer i, 2 qualified response
Ancillary Studies
Material balance calculations
This calculation only works weil in reservoirs with
good communication fe. in those reservoirs having high
Petmeabilties. This is not just a mater of calculating an
average pressure, although this is a significant issue.
There are also other considerations, Material balance
studies normally have considerable uncertainty when a
45 cap and/or a water leg is involved ?
In this area, one often finds contradictory advice in
the literature. The author has worked on a number of
reservoirs where the material balance study provided
inaccurate results and actually provided results very dit
fecent from simulation results. The material balance does
not take info account pressure variations existing in
moderate. (0 low-permeability reservoirs. Having said
this, ARE recently did a study. and an individual who
‘worked on the project matched the OOIP and gas cap vol-
ume using a material balance program. The reservoir in|
question had braided channels, and the matetial batance
‘was of no help in determining these locations, Based on
previous experience, the author had been extremely
doubitul that accurate results would be obtained.
Material balance studies have obscured ot misdirect:
‘ed solutions as often as they have provided useful insight.
In general, the author no longer does material balance
studies as part of presimulation work, unless the reservoir
has very high permeability and is of limited size
However, this is ot an absolute rule. Reservoir engineers’
views on this matter will vary according to the proportion
of success they have observed.Integration and Gridd
Objectives
Importance of objec
Depending on its design, a numerical simulation will
take into account various factors to varving degrees
“radeots are usually requited. Hence, most studies are
normally aimed to address specific issues.
At fist glance, determining the objective might seem
fa trite comment, but, in many cases, problems are
defined in terms of the solution, Far example, ifthe pro-
duction performance from a reservoir is proving to be
enigmatic, one will be requested to do an areal simula
tion. However, the underlying problem is that reservoir
mechanisms are not understood, In this case, the real
objective isto propase possible reservoir configurations,
fluid properties, or geological models to expiain the
observed behavior. smaller model can be used to
determine physically plausible proposals
Common situations
Here are a number of common situations:
The most common situation concerns a reservir
on primary production, which an operating com-
pany has now decided to waterflood. They are
looking for a waterflood reserves and production
forecast based on the results of a simulation
Another situation involves a recently drilled well
in a new field, which has a gas-il, gas-water.or
wateroil contact in close proximity to the com
pleted interval. simulation has been contemplat-
ed to support a recovery factor as well as to pre
dict GOR or watercut trends and, possibly.to eval:
uate mitigation strategies.
[An operating company has @ mature field they
have decided to model to optimize production.
Small increases in overall recovery factors may
represent major increases in remaining reserves,
Often, iis hard to distinguish incremental
reserves fromm production acceleration,
A large developed gas field exhibits retrograde
liquid condensation) behavior A study has beer
commissioned to determine natural gas liquids
(NGL) and gas recoven
A large miscible flood has been installed.
The operating company has decided to
pertorm a simulation. Since limited production
history exists. a simulation is often the bes
method of engineering analysis available to
estimate recove
Each of these situations involves somewhat different
issues chat must be resolved. In truth, one cannot get all
of the details of the real reservoir into a simulator It is
necessary to get the important things right so the mecha
nistic behaviors—the physies—are captured correctly
Critical issues
The following is a list of critical issues likely to arise
‘with the situations listed previously.
‘or the waterfiood implemented alter period of
primary production. one of the most critical
aspects of mode! design is characterizing layering
or heterogeneity This caa be done either with
detailed layering built into a model or via pseudo
relative permeability curves.
For a well drilled into a pool with a gas-il,gas-
water or wateroil contact, the critical modeling
issue will likely be accounting properly for aear
wellbore effects. This usually volves a detailed
Coning study. defining layers, covering the correct
range Of production rates, and including the
elects of hysteresis,
For a mature feld optimization, study the most
critical aspect for the most likely way to build a
balanced model. This could require preliminary
coning and cross-sectional studies that must be
integrated {ater into an areal model. Integration is
the major issue
Fora retrograde gas condensate reservoir,a C01
positional simulator is likely requited. Proper PVT
characterization will comprise a major portion of
the study.
Fora miscible ood, a number of critical issues
4a possible The same simulator may not address
all of the issues effectively f miscibility isthe crit-
cal issue, a compositional similatoris in order. It
maximizing sweep efficiency is the critical cor
cemn,then a pseudo miscible simutator is probar
ly the most effective solution,Simulation
Other factors, suck as well spacing, laboratory relative
permeability. PLT properties, injectivity. and timing of
wells are still important
Multiphase studies
Historical: a single model could aot address all
‘ssues. Therefore it has been common to assemble many
diferent models deat withspecticfisues. A largescale
ation may be split up as tollows
Coning Study.A detailed radial grid study would
be performed to evaluate coning For instance.
reservoir appears to have No gas Coning despite a
7.000 kPa oF 1.000 psi pressure drop Experience
ievelscargand not everyone may find the solution
obvious 8 relatively cheap coning study can deter
nine the issue and cause subsequent elements of
the study to be changed. Ita gas cap realy does
exit itis possible co use ths information to evalu:
ate well pseudo relative permeabilty curves
CrossSectional Study The objective of a crosssee
ional study is similar tothe coning stu that is
:0 sieplty the problem to gain general under
standing Often these studies are used to generate
pseudo relative permeability curves. as shown in
the problem set in chapter 20 of this bork. The.
use ofthese curves can simplify a multilayer
problein ino a single-layer model. which may be
used to develop an areal modet
steal Study Tis is probably the most common
sype of stud allows complex watertlood pat
terns. which seem occur as an accident of his
tonto be evaltiated, However withthe lager
cconipaters available todayitis getting easier to
make multilayer. tulscale -0 studies
Each of the stages will build on the earlier stages 10
develop a full Reid model
Gridding
Realistic grids
Ideally, a very fine grid could always be used.
However, as the dimensions of a grid increase, the
amount of calculation time increases rapidly:
Determining the amount of work varies depending on
the mathematical implementation used in the simulator
In general terms, the amount of computer time increas-
es to the fourth power of the grid dimension (1) for a
common direct matrix solver implementation and a 2D
square grid, ie. cells in each dimension or total of 2
xem -calls in the grid. With layers, the amount of time
increases by the cube of the number of layers (P), A
large grid can, therefore, exceed the capacities of a
supercomputer for a large reservoir; hence, constructing
2 model usually involves some compromises,
General guidelines
Some engineers use the number of acres in a grid
block as a guideline, but this is not recommended,
Others use two grid blocks between wells, but this is nei
ther adequate nor recommended
A rreservoir simulation is often equated with a materi
al balance; this is true, but only in a limited sense. In the
‘numerical model, the blocks are linked via a mass trans-
fer equation (Darcy's Law).
By virtue of this interblock transfer, a reservoir simula
tion is designed to calculate the transient behavior aceass
the grid. Therefore, itis entitely different from a material
balance calculation, which presumes tanklike conditions
of equilibrium at each timestep.
Flow visualization
The key difference between the reservoir simulator
and a material balance equation is interblock tans.
The best guideline is a flow net. Muskat frequently used
flow nets in his work, Flow of Homogeneous Fluids
through Porous Media." They are relatively easy to draw by
hand. Two examples are shown in Figure 10-I and 10-2."Headpond
Earth-Fill Dam
(permeable]
s
gine’
ot!
Isopotential
Lines
"18 vectogrmscon be gown by hansth an ocuoey ofaronmaey:
os
Fig. 10-1 Flow Net for Eant-it Oam
Headpond
Py
(Concrete,
no permeability)
T T
Flow Lines
Na Thee gras con be crn by Rand th macro af aroumnaty 10% “iced matareofeamen onion
Fig. 10-2 Flow Net for Earth Fill Dam with Grout Curtain
PREY py i schon ntimulation
Irregular grids
The use of flow nets suggests the use of iregular
grids. Where the flow net is fine, more grid blocks ae
required. In fact, i is possible to use the grid from a
flow net to make a curvilinear grid system. This can pro-
urate results. For example, a grid for the leak
age under a dam is shown in Figure 10-3. From this
information, a finer grid could be constructed where
low concent
ions exist
Grid sensitivities
One cannot analytically calculate the amount of ertor
from grid discretization, However, it is possible to deter
mine how the error varies. In practice, a grid sensitivity is
030m
com | Datum
Flow Concentrati
Fig. 10-3 Flow Concentration for Earth Si! Dam with Grout Curtain
Grid Sen:
required By
rate to within an approx
and 10-5 show a gasoit and wateroil coning grid sens.
uwity done by ARE during a reservoir simulation.
Sadly, grid-sensitvity checks are rarely run in field
studies. The example in Figures 10-5 and 10-6 shows the
areal simulation is only accurate to within 210%. The
nathematicians are horiied with this level of accuracy
however. the brutal
hanging the grid dimensions and making
ns, itis possible to show the model is accu:
gures 104
ate percentage
ruth is most reservoir simulation
put is not known with this degree of accuracy. With the
capability of computers continuing to increase, this may
be less problematic in the future.
Insufficient grid blocks
Reducing the number of grid blocks
can be taken too far, The following exam
ple occurred in a study proposal submit
ted to the author. The objective was 10
determine horizontal well potential in a
tight sandstone reservoir. Production in
vertical wells was uneconomic due to low
rates and GOR penalties. The study objec:
tive was to determine if a horizontal well
‘would inerease productivity sufficiently to
enable economic production and if GORs
‘would be low enough to avoid penalties
Asdiscussed previously a flow net isa use-
{ultool for visualizing flow inthe reserva
‘The proposed grid and the expected flow
pattern are shown in Figure 10-6.
2000)
Ratia SC, mim
g
GeO Seve Tine
G35 Ol Rabo SC vs.
{G35 Oi Rabo SC vs. Tine
{G35 Ol Ratio SCs. Ti
&
Since the reservoir is tight, less than 1
mD, localized gas saturations would be
expected to arise, A fine grid made with
small grid blocks and thin layers would be
required around the well. An element of
symmetry could be utilized through the
center of the well to reduce computation,
The proposal tor the situation
described previously recommended three
layers with the wellbore ceotered in
blocks with 100 ft. horizontal x and y
dimensions. Although these large grid
blocks could not correctly represent the
0 000
Time, day
Fig. 10-4 Grid Sensitivity GOR from Field Study
mt <
physics in the reservoir, the model would
still produce results, The proposal was not
considered favorably by the author
B00
aciniGrid Sensitivity Analysis
Wares ve Time 5
Water Cut sCve Time, 14-25
mm Water Cut v5 Teme, 4-2
a Water Cut Cove Time, 14-25
oa
2,
5 .
‘0 sooo
Time day
Fig. 10-5 Grid Sensitivity Water Cut rom Field Study
View Down Centerline of Well
Solution Gas Breakout
ig. 10-6 Flow Concentrations around a Horizontal Wel!
Actually. this problem is quite common in the author’
opinion. Simulation output should have a warning at the
top of the output file
Warning:
Simulation may give nonphysical results
if too coarse a grid has been used
in reatity, there are no warnings at all, although
Eclipse documentation is reported to have some warn-
ings. The fact is the simulator runs much faster and gives
fewer problems if too coarse a grid is used, Buyer beware!
‘A good example of these differences can be found in
the SPE comparative solution on horizontal wells, One of
the simulators used a high-permeability grid block to rep-
Most Severe Pressure Gradients
Detail Grid Required
resent the horizontal well
Figure 10-7
shows the predicted GORS for a horizontal
well The author does not recommend
the approach in the example given.
| Two basic approaches to gridding
Based on experience with reservoir
engineers new to simulation, gridding is
fot intuitively obvious. There are two
basic approaches for seting up @ 3. grid
The first is based on setting up a spatial
grid end then assigning rock properties
based on geometrical position. The sec
cond isto assign grid blocks based on geo!
ogy and then alter the dimensions of the
Tova grid block. These two methods are shown
n Figures 10-6 and 10-9.
In the author’ experience, its rare to
assign grids based on fluid type, i, oil,
gas, and water. Nevertheless, uids can be
‘a consideration
Capillary pressure transition
Normally capillary pressure curves
are relied upon to give the fluid satura-
tions in a grid block, However, if flow
visualization is appiied to wells in a tran-
sition zone, it may have an important
effect on reservoir performance. For
instance, would the two situations shown
in Figure 10-10 be equivalent?
Grid orientation effect
Grid orientation can be summarized
quickly with Figure 10-11. Grid orientation becomes sig
nificant at high adverse mobility ratios. This commonly
‘occurs whenever there is gas injection, Since most hydro-
carbon miscible floods (HCMF) use enriched gas, grid
orientation is important in EOR design, {tis also a major
factor in thermal simulation; however, these models woe
cally incorporate a nine spot finite difference molecule to
control this problem, This effect will not show up as a
material balance error or anywhere else for that matte.
‘The best method of handling ths type of problem isto
develop a streamline model and then make detailed
‘ross sections. For miscible floods, detailed heterogene-
ity can be incorporated using geostatistical realizations.
This technique has been used by Chevron in the Mitsue
Gilwood Unit No. 1Practical Reservoir Simulation
GOR sctistb
0
oe a ao
Time, days
Fig, 10-7 SPE Comparative Solution—Varying Grid Refinement
Fix Grid Geometrically, Fit Geology to Grid
ig. 10-8 Basis for Grid Constmction—Fixed Grid in Space
Fix Grid to Geological Layers, Fix Structure to Geology
ig. (0-9 Basis for Grid Construction Fit Blocks to Geology
Reality check
‘The perception of where good
neering is required can be somewhat dis
torted, For example, consider the case of
& massive high-permeabilty carbonate
reservoir above the bubblepoint, which
has strong bottom water influx, and a pro
uction history water fre, since the wells
are completed away from the water inthux
No gas coning is involved because the
reservoir is undersaturated, Large grid
blocks with lab relative permeability
curves will work very well. The pressure
differences between grid blocks will not
be that large due to the high permeability.
Water influx is nearly vertical because of
the high permeability Furchermore, geo-
logical layers stabilize vertical displace
ments rather than accentuating break
through. Costs are not a real concern
Frequently, the biggest problem is getting
the tubing pressure drop curves matched,
Simulation. in some situations, is actually
auite easy.
if the same simulation engineer who
has been working on the reservoir
described previously later works on a large
carbonate reservoir with low permeability
and with water and gas contacts, he (or
she) will find life gets very hard. The gas
will cone. GORs will increase rapidly due
to the low permeabiliy The water will
cone in downdip wells. There will be sig:
nificant differences in pressures between
atid blocks due 10 the steep pressure grax
dents in the reservoir A waterflood will be
implemented in a pattern to provide for
voidage replacement, Large changes on
an individual well basis are required for
well pseudos on gas cap wells and
downdip water producers. The rock rela
tive permeability curves will be changed to
a pseudo relative permeability. This is
required, since layering control the break
through of the displacing fluids for hori
zontal floods. It is the reservoir from hellFig. 10-10 Basis for Grid Constactuoe—Diserbution of Flu in Reservoir
Not only that, buz everyone is constantly concerned about
costs and when the project will be finished. The engineer
may ask, “Why am ! working here?" «
Simulation and reservoir engineering requirements
are often more severe in low-quality reservoirs (read
mature basins) than in massive high-gualty (high:
permeability) reservoirs. This is often
opposed to the economic margins, This
is frequently true of the more classical
reservoir engineering techniques such as
material balance
Limitations of a simulator
A simulator cannot direct an engineer
to the right solution. It will answer yes or
1no, but is incapable of a creative process,
On the positive side, the creativity is really
the fun part, So perhaps it is bew the com
puter can’t do this. Management is unlikely to replace us
computationally challenged biological units wi
puter for a long time.
Parallel Grid Parallel Grid
5.=051 /
/ ~_]
4 / :
Y= 048 ep / bw = 0.048 cp
Diagonal Grd
Diagonal Grid
Fig. 10-1 Grid Orientation BtectsPractical Resarvoir Simulation
Summary
In order to provide effective answers, review all of
the data, production performance, and geotogy
determine ail of the issues that must be resolved. This
will require some flexibility Its easy to fall into traps it
tthe problem is deseribed in terms of a potential solu-
tion. Lack of data need not be a deterrent. Either a
series of sensitivities may show a range of solutions, or
it may conclusively show the missing data does not
affect overall results significantly.
Experience has shown there are a number of well:
known critical issues for certain situations, Be aware of
these problems and address them in developing the
approach. Breaking the study into different components
allows one to evaluate various aspects in detail. Certain
situations allow specific creative solutions. Radial sector
studies often work well on salt dome structures; a
Cartesian grid can be used for fractured wells and nor:
symmetric reservoir discontinuities
Developing a grid requires a number of tradeotts. The
capacity of your computer will have a major impact on
how fine a grid you can use. Be careful of some of the
rules of thumb. [tis impossible to base grid requirements
‘on blocks of x acres or two grid blocks between wells,
Grids depend on many factors. Flow nets are probably
the best tools for indicating grid requirements. The only
‘way to tell i it is“right"is to perform a grid sensitivity
Setting up a grid can be difficult, particularly tor
novice users, Again, new engineers can cheat, i., look
up grids used in existing papers and textbooks, Grid ori-
entation is important in a number of advanced applica-
tions. Often, miscible flooding is best handled with
streamline models.
Finally, fundamental grid errors are user transparent
‘The simulator can and will calculate physically meaning
less results with incredible precision, Moreover, it will ran
faster and give you fewer problems as it merrily leads you
to perdition. Buyer beware!
208 «
References
Arps.J. “Analysis of Decline Curves” Ransactions ofthe
UME Nol, 160, 1945,
Carlson.M R.Tips.Tricks and Traps of Material
Balance Calculations’ Journal of Canadian
Petroleum Technology, 1997,
Muskat,M. Flow of Homogeneous Fluids Through
Porous Media, MeGraw Hil 1937,
ter Wyckolf and Reed, Physics, date and.
issue unknown.
Nghiem, et al.“Seventh SPE Comparative Solution
Project: Modelling of Horizontal Wells in Reservoir
‘Simulation’ SPE 21221, 11" SPE Symposium
fon Reservoie
Behrens,R.A.etal,"Implementation of a Streamline
Method for Flow Simulation of Large Fields” Journal
of Canadian Peroiewr Technology, 1996,