Initialization
Introduction
‘The primary objective of initialization is to establish
the correct amount and distribution of the original reser-
woir fluids. The majority of simulations are initialized
based on static gravity capillary pressure equilibrium.
This works well in the majority of cases. However, there is
‘ho requirement the fluids in the earth are static and, in
truth, this is rarely completely correct.
A number of different methods are used to enter
fluid saturations.
Grid-block center capillary pressure initialization
It the grid blocks are smaller than the capillary pres
sure transition zone, then the saturations of grid blocks
can be accurately estimated from the capillary pressure
curve at the midpoint of the grid block. This also requires
the depth of the relevant contacts be specified.
This was the method used by early reservoir simular
tots, The distribution of fluids in the grid blocks was
assumed to be uniform
Pseudo capillary pressure
Later, it became possible to use a specialized tech.
nique to simplify reservoir problems that involved capil
lary pressure transition zones smaller than the grid block
height. Known as vertical equilibrium (VE), this method
requited the saturations in the grid block be averaged.
Since the transition zone is assumed to be of negligible
thickness, the saturation of the block can be calculated
using a linear relationship based on the distance from the
specified gasoil contact (GOC) or wateroil contact
(OC). This is shown in Figure 9-1
Recall from chapter 8 that VE generated straightline
relative permeabilities. Capillary pressure was ignored in
the displacement calculations. As shown earlier, capil-
lary pressure does affect frontal displacement and intro
ducing this fudging stategy to the relative permeability
will affect displacement processes in the simulator.
>2i7Practical Reservoir Simulation
ety
a +
on ss
| Negigite
Poo+ i Tamsin
I Bone
| Thickness,
-o7h —_ vo
Fig 9-1 eto Copia Presse Cise-¥E
Finite transition zone thickness
A problem soon becamne apparent regarding relatively
thick blocks existing with transition zones consisting of a
fraction of the grid block, Such a situation is shown in
Figure 9-2.
Early efforts involved generating a pseudo capillary
pressure graphically. Such a curve has the same appear:
ance as the results shown in Figure 9-3 for the previous
data
This will correctly calculate the original fluids in
place. Apparently i is difticut co program an automatic
integration of the P, curve. This system had two prob:
lems. The capillary pressure curve is dependent on grid
block thicknesses, This makes calculation of pseudo
capillary pressures difficult where gridlock thickness
es vary; the solution is that one must use uniform grid:
block sizes. In addition, the pseudo capillary pressure
affects displacement calculations
Often, getting the correct amount of fluid in place
was more important than displacement, such as pci
Mary production performance. Modern simulators now
use a different approach.
Grid-block subdivision
By arbitrarily subdividing grid blocks during the is
tialization process, as shown in Figure 4, it is possible
to pick multiple points on the P, curve and average them
to derive an averaged grid-block saturation.
For most programs, this is user transparent, However,
there ate programs allowing users to specify tow many
subdivisions are (0 be made. It's also necessary ta specit:
ically request depth averaging for this type of calculation
to be done. This is much easier to implement and should
give satisfactory saturation accuracy
like
Fuyther thought is needed regarding
gridding With a fluid distribution ier the
grid diock. does the computed pseudo
capillary pressure accurately represent the
displacement process? The author has not
done any tests on this yet, but it may be a
serivus consideration jn constrecting a
eservoir arid
Type of capillary pressure data
At present. capillary pressure data for
initialization is nosmally abeained from
drainage capillary pressure tests. Tis is
fone area where simulation technology may change. As
discussed earlier. the real reservoir process is one of
countercurrent imbibition, Eventually. with more devel-
opment, initialization may be based on more realistic
laboratory data
Tilted water contacts
Tilted water contacts have been recognized for a
long time. Figures 9-5 and 9-6 show some diagrams for
reservoirs where tilted water contacts are well
established.! There are many reservoirs with tilted water,
contacts, which are caused by the subsurface move-
ment of formation fluids, These movements are studied
by hydrogeoiogists
Hydrogeology
‘The study of hydrogeology is far from new Most
provincial governments and the federal government are
busy modeling surface groundwater flow on several
watersheds in the province at any given time. Most of
these studies do not extend far past the Pleistocene sta
low formations and are intended to manage farm, munic-
ipal, and industrial groundwater use. A significant
amount of original research on this topic was done by the
Dutch to suppor theie system of dykes
‘The application to oil and gas was recognized many
years ago by M. King Hubbert! He demonstrated the
movement of water in the subsurface would alter che
location of oil an gas pools and showed how to calcu
late this effect. He died a few years ago and was widely
recognized for his achievements. However, the practical
application of this theory has been stow.
Due to the foresight of the Canadian government and
industry all DST data is in the public domain atter one
year. This source of data is critical to performing basin
Seale hydrogeology. Brian Hitchon of the NationalP. Data
0S. 10
494Pa
54 = 51.89%
=24.5 kPa
5. = 37.8%
P.=38.0kPa
30.8%
P.=49.0kPa
Fig 9-2 Grid Biock with Finite
Thickness
P Zone
—
Se
Fig. 9-3 Pseudo Copillary Pressure Curve Caiculation—
Finite P, Zone
Fg 9-4 Generation of Average Grid Block Saturation
Using Venical Grid RefinementFig. 9-6 Example of Tilted Water Contact MontanaResearch Council Canada (NRC) has done pioneering
work in this area One company remarketed the DST
data with interpretations in a form shat had been quality
controf checked, but they are no longer actively selling
this data. More recently, Hydro Petroleum Canada has
prepared complete subsurface potential maps for almost
the entire basin, and it includes water chemistry as well.
The trapping of most oi! and gas pools in the Province of
Alberta can be expiained based on fluid movements and
number of trapping mechanisms.
Many exploration wells during the years have been
termed geological successes but economic failures.
Restated, the geology extrapolated was accurate, but n0
oil or gas accumulation was found. The industry has not
historically been able to predict the structures, reels, et
containing oil or the ones containing oil and gas.
Ieis much easier to trap gas, and this is why 60% of
eserves in Canada on a barrels of oil equivalent (BOE)
basis are gas
‘Manual input
Water, ol, and gas saturations caa be input manually
in a simulator using one of many different read options.
This would include grid-block by grid-block specifica
tions as well as block or layer reads
This is normally done either on simple data sets,
where itis faster to hardwire a saturation,
‘or when dynamic conditions exist
Inthe latter ease, itis possible to input
saturations based directly from log analy.
sis, At present, simulators do not allow
for dynamic initialization, However, a
simulator can be run to achieve this
objective and the output arrays convert:
ed to input arrays. Developing the
dynamic initialization may tequire exper-
mentation, and the reservoir simulator
must be run for a sufficient time to emu-
late steady state conditions.
Reservoir engineering implica
tions of dynamic trapping
Tilted contacts and dynamic condi
tions ae frequently not recognized. Quite
bizarre interpretations are often requited
in onder to force (incorrectly) a static
interpretation. Some examples are out
lined following
‘Actual Conditions,
DynamicEquiibrim
Phantom Faults
rterreted to Match
Precancelved Satie
Equlltiam assumption
Initialization
Example 1. ifthe contact is tited, the obvious adjustment
to the geological interpretation is compartmentalization
‘with separate water contacts. Very oten, these end up as
the mystery pisichouts and phantom faults, ie, can't be
seen on seismic, such as those depicted in Figure $7.
Clearly, when a simulation is done on a reservoir that
‘s actually continuous, i (S going 10 be difficult 0 history
march reservoir performance. In one case, we had a client
‘who was going to stop injection since they felt communi
cation did not exist, despite the fact pressure data proved.
conclusively that communication was occurring
Example 2 Very often, most of the oil or gas will be dis-
placed off the top of the structure and reservoir perform
ance will reflect the original distribution of fluids. This is.
shown in Figure 9-8,
Water influx will occur from updip and downdip, and
the best wells will actually be offset from the structural
high. This example is based on a reservoir in
Saskatchewan and is easily demonstrated based on well
cumulative productions. da static gravity equilibrium is
used here, it will be difficult (impossible?) to match the
cumulative production, since the oil will not be located in
the correct location.
Fig.9-7 Effect of Misinterpreting a Tited Water Contact as a Series of Barriers
>aPractical Reservoir Simulation
sere wetsocated
offsets hg
Fig 9-8 étfect on Well Placement When a Tilted Water Comact
Has Been Misinterpreted
I mater
seen derma
saentoemarrate,
ae
00s Water cut
Conventional watertlood approaches
have stiowa placing an injector in the cen:
ter of high-permeability in an isolated
pool will give the highest flood efficiency
Based on the assumption used. this is true.
However, if one places injection in the
center of a hydrodynamic pool, then the
oil is ejected out of the trap, This is depict
ed in Figure 9-11
This has happened in a few pools.
Watertiooding with inconectly placed injec-
tors in @ hydrodynamically tapped pool
will reduce recovery
Simulation failures
‘These are examples of so-called sime
lation failures. Of course. the simulator is
an inanimate object and isnot responsible
for anything, The failures that have
occurted did so because die underlying
physics were not understood. Sadly, the
poor understanding often dates back to
the 1950s. However, in Canada, the wealth
‘of public data has made available detailed
mapping of the sedimentary basin, and
the full implications of hydrogeology can
be applied.
Fig. 9-9 Explanation of Wels shat Originaly Produce 100% Wover
Followed by Increasing Oil Cuts
Example 3. [n some reservotrs,he water cut will go down,
when the well is put on production at high rates. The situ-
ation is shown in Figure 9-8.
The gradients induced by reservoir production are
much stronger than the potential gradients associated
with trapping, The performance described previously is
common in the Sawtooth in the southern part of
Alberta, Often, these wells will mysteriously produce
many times the OOIP for their driting spacing unit. A
‘gravity static equilibrium is unlikely (0 calculate the cor
rect cumulative oil, since the updip pinchous location is
unknown and the sirong aquifer precludes accurate mate.
rial balance calculations.
Example 4. From an areal perspective,cil trapped hydro-
dynamically in high-permeability formations will natural:
ly shrink to a central point. The influx occurs from all,
directions. Water displacement will progress as shown in
Figure 9-10.
222 <
Modeling dynamic equilibria
Geologists are quite interested in
hydrogeology, and the author has prepared a talk on
reservoir engineering implications for the Canadian
Society of Petroleum Geologists (CSPG). The positions of
cilwater and gascil contacts were generated using a
reservoir simulator. Most hydrogeology calculations ut
lize steady state flow assumptions and could solve this.
kind of problem faster than a reservoir simulator.
However, a model specifically for initialization has yet to
be built.
‘An anticlinaliike structure with a production well
located on the structure is shown in Figure 9-12. At pres-
ent, the fluids show a distribution of uids that would be
obtained via static gravity equilibrium. Capillary presare
is included in this model. Note the gradual change in the
wwateroil transition,
‘The situation has been made a little more realistic
by including layering in the problem. The permeability.
distribution could change to something similar to
Figure 13,POOL WISPLACED
PL BOUT
WE Re Query
feos eae
PRIMARY
Fig. 9-11 Elect of Misplacing an injector in Dynamically Tapped Fool
initialization
To obtain dynamic equilibrium, the simu:
lator was deceived by using an injection and
production wel] combination in the aquifer,
The author has used a high horizontal poten.
tial gradient to make the point more clearly
The resultant potential distribution is shown in
Figure 9-14
The potential field is not identical to the
pressure distribution, which is shown in
Figure 9-15,
‘The locations of the fluids are shown in
Figure 9-16.
The direction of water flow can also be
reversed. The effect of this is shown in
Figure 917.
No Flow ~» Static Equilibrium,
1g 9-12 Reservoir Simulation Tests for Dynamic Equilbrium—Staicinitialzationsarvoir Simulation
EEE Medium |
7983 mam High
sos.
aor.
ai
asa.
az
Fig 9-13 Reservoir Simulation Tests for Dynarine Equitbrium—Permeabiity Distnbution
798
P datum =
oF potential
038)
078.
ai
aes
207.
z
Fig 9-14 Reservoir Sirulation Tests for Dynamic Eauilibrium—Flow Potential (Datum Pressure}Pressure, initial
Fig. 9-15 Reservoir Simulation Tests for Dynarnie Equilionium—Raw Pressure
7950:
_- oil
7999
036.
ain
3.68
207
Fig, 9-16 Reservoir Simulation Tests for Dynamic Eqactica
son.
07.
07:
Fig 9-17 Simulation Tests for Dynamic Equilibrium—Reversed Flow Location of Fluids
2.000 3000 k
ate SC, bbl/day
3
Injection (WF)
reduces recovery slightly
1.000
_- Natural Drive
Waterflood~~
1.000 12.900 100
Time day
Fig. 9-18 Reservoir Simulation Tests for Dynamic Equiltvium—Watertiood Sensitivity
Deemer eemmmeemmaanntae, steesaremeame ssThe location of fluids is determined by the potential
ivld and, when steady state is achieved, is independent
of tre absolute permeability. Layers will alfect location of
rhe hydrocarbons. You can increase permeability to
speed up the development of the saturation distribution,
The output from such a run can be used to manually
specify the saturation array in the simulator,
Water injection
The elfects of water disposal were modeled using a
restart fle. Cases were run for both initial conditions in
which water flow starts from both directions. In both
cases, recovery from the production well went down with
waterflooding instead of increasing, The production fore-
cases aze shown in Figure 9-18,
The effect of the high-permeability layer is shown in
Figure 9-19. Note the drop in oil productivity and the
change inthe GOR profile. The waterflood starts at 19,000
days, which was the time required to develop the steady
state flow conditions for initialization,
2.000
Uniform Reservoir
Oil Rate
‘olf Rate SC, bbliday
Layered Reservoir
Rates:
10,000
1200
Time, day
Fig. 9-19 Reservoir Simulation Tests for Dynamic
Initiglicaeion
Hydrogeology
Recently this problem has been resolved by Henning
Lies of Hydro Petroleum Canada (HPC). At the time of
this writing, most of his material remains unpublised,
which is unfortunate. HPC has created regional hydroge-
ological maps of most of the Western Canadian
Sedimentary Basin, They have examined what causes the
trapping of probably 80% of the poois in Alberta, British
Columbia, Saskatchewan, and Manitoba.
Another Key area where HPC has done practical
research is in modeling the fluid movement in the sub
surface. At present. only One of these projects has been
published: “Hydrogeological Trapping Model of the Mitk
River Gas Field" by H. Lies"
No production runs
During the early days of simulation, it was common
0 nun the simulation with no well production. The
objective of this test was to see if gravity equilibrium
had, in fact, been obtained. The output log would indi-
cate no material balance changes and no saturation
changes in any grid blocks. With depth averaging, this
may not always be the case and will definitely not be
3,000
Layered Reservoir
ORs
2000 B
=
a i
Uniform Reservoir 3
ei ro 8
Reversed Waterflow
GOR
14000
Equilibriam—fect of Layering on Waterflgod
> 27Practical Reservoir Simulation
true if a dynamic system is madeted. It is still useful. It
will indicate how much effect the depth-averaging cal
culations have on the reservoir as fluid distributions,
equalize. The reservoir simulation calculation is still
based on grid-block average properties that mathemat>
cally occur exactly at the grid-block center
‘Summary
Initialization appearsto be a straightforward matter on
the surface or initially (two bad puns). Gravity icitializa-
4ion is relatively simple to implement with modern fear
tures, An engineer must be aware of what features are
available, There is no law, however, stating that static grav
ity equilibrium mest exist In fact, the opposite is true. If
fone has trouble getting a history match, this may be a
good place to reexafnine basic assumptions. Dynamic
conditions are common in the Sawtooth of Southern
Alberta, all of Saskatchewan, Indonesia, and in the
Middle East. Qatar has some strong examples ofthis.
228 <
References
Dahlberg. €.C. Applied Hscrdynannies in Petroleum
Exploration 2 ed, SpringNerkog. 195.
Hubbert. LK, The Theory of Growl ater Mion ane
Plated Papers, Hafner Publishing Company 1969.
Hitchon, 8 etal, "Culling Criteria for Standant
Formation Water Analyses” Applied Geochemisiry
ol.9, 1998,
Lies, H.Hydrogeotogical Tapping Model of the Milk
River Gas Feld Journal of Canadian Fetolewra
Technology, (December): 25-3), 1895,