ROUNDTABLE: RESERVOIR SIMULATION
Exploring the Role of Reserv
Manfred Wittmann
(moderator)
Schlumberger Overseas S.A
Jakarta, Indonesia
Reservoir simulation isthe fool engineers use to model the
possible flow behaviors of the reservoir over fime—from
pessimistic, 1 optimistic, to probable. Here are the views of
six simulation experts on the scope of reservoir simulation,
its proper inputs, quality control, and trends forthe Future.
‘Abdul Karim AL-Rabah
Kuwait Oi Company
Almac, Kuwait
Pramod P. Bansal
Mobil Exploration ana Producing
Services In.
Dallas, Texas, USA
‘THE FUNCTION AND SCOPE OF RESERVOIR SIMULATION
MODERATOR
Manfred Wittmann
Al, since you area builder of reservoir simulation models, lot's
start wth you. What does a reservoir simulation do!
AL BREITENBACH
The function of reservoir simulation isto solve the fundamental
flow equations insufficient detail so that you can predict the
consecences of changes in operating practice. Successful reser
voir simulation, therefore, requires solving those equations as
precisely as possible, This means two things. First it means
developing a numerical simulator that will give accurate solu
tions for the data input. Second, it means being able to describe
the reservoir with suficient accuracy so that the numerical solu:
tions give an accurate basis on which to make predictions.
MODERATOR
‘You used “Suficient” twice. What is suficient detail and sufi
cient accuracy?
NANSEN SALERI
Sufficiency is reservoirspecific, dependi
nosily on reservoir
size and complexity. You may have close to sufficient data in
small, simple reservoirs, and may never approach it in large or
‘complex ones, no matter how hard you ty, Because data about
reservoirs almost always lack a degree of detail and accuracy,
reservoir simulation cannot be considered a deterministic tool.
is more probabilistic, along the lines of weather forecasting mod:
cls, if we accept reservoie simulation as such, then I think itcan
be put to optimal use
Oilfield Reviewoir Simulation
EA. Breitenbach
Scientific Softwae-ntecomp, tn:
Denver, Colorado, USA
Larry D. Hallenbeck
Philips Petroleum Cornoany
Tananger, Norway
MODERATOR
‘What is optimal use—why undertake a reservoir simulation?
PRAMOD BANSAL
‘We typically use simulation studies in two settings: In develop-
ment ofa new field, a simulation can help us decide how many
‘ells we need and where they should be. In an established field
that has unexpectedly poor performance, a simulation can tell us,
why and help us find ways to improve performance. In ether
setting, itis important atthe outset to define the purpose ofthe
simulation and do the necessary homework before undertaking,
the simulation. This homework involves use of simple reservoir
engineering tools, such as material balance calculations and
boundary conditions and aquifer size.
In the case of poor performance of an established field, material
balance equations may not tll you why you have poor perfor:
‘mance; you may need more description and simulation
ABDUL KARIM AL-RABAH
| question the value of simulation studies atthe development
stage. Without the benefit of production history, think results of
simulations during the development stage are doubsful
SALERI
(On the contrary, reservoir simulation may be at its best when
data are sparse. Itis an effective tool for comparing different
probabilities influencing flow—pressure-volume-temperature
(PVD variations or variations in the continuity of shales or faults,
‘rin the connectivity of compartments ofthe reservoir. Simula
lion studies are the most effective way to develop a sensible
‘engineering solution to the different possible scenarios.
Volume 2 Number 2
Nansen G. Saleri
(Chevron Exploration and
Production Services Company
Houston, Texas, USA
D. Nathan Meehan
Union Pacific Resources Company.
Fort Wont, Texas, USA
AL-RABAH
| agree that when data are sparse you need to run some basic
simulation for planning. But | don’t think you should develop a
field based solely on simulation studies if production history is
lacking, Ifyou have some wells, with atleast a minimum of pro:
duction history, then you can go back and modify your mode!
But without that, I don't feel that | can decide how to develop a
field of 100 wells
SALERI
Unfortunately we dont always have time to gather all that infor
mation. In some exploration projects we have only 4 to 12
Weeks to make the best technical evaluation we can, on which
large investments will be based. In this setting, reservoir simula-
tion plays a key role
BREITENBACH
| think perhaps Nansen (Saler and Abdul Karim [Al-Rabah] are
talking about two different things. When you're considering,
‘major investment decisions like location of offshore platforms
‘and pipelines, you may not have a great deal of information, but
‘you still have to decicle where and how many wells to drill. Even,
if you might not be able to produce fora year or several years,
reservoir simulation isa useul sensitivity tool. It gives you devel-
‘opment strategies for a range of possible reservoir conditions —
the presence or absence of faults, whether they are sealing oF
‘er contains to his atc, thank alo to Luis Ayetaran, Schlumberger Technical
Seren Dut UAE Tom Gul Siti Son tacone, fe, Deve Ol
‘alo USA’ and kia Sh Seino, Katt Ol Conny Aras ew
Roundubl candace i San Amoi, Teas, USA, Oxi 1, 1989.
Repeat Formation Tee (FT fa atk of Schaberge
Cray tea wademarcof Cay Reseach eROUNDTABLE: RESERVOIR SIMULATION
Exploring the Role of Reserv oir Simulation
‘Manfred Wittmann,
(moderator)
Schlumberger Overseas S.A
Jakarta, Indonesia
Reservoir simulation is the tool engineers use to model the
possible flow behaviors of the reservoir over fime—from
istic, to optimistic, to probable. Here are the views of
wlation experts on the scope of reservoir simulation,
‘proper inputs, quality control, and trends forthe future.
Abdul Karim AL-Rabah Pramod P. Bansal
Kuwait Of Company Mobil Exploration and Producing
Abad, Kunvait Services Inc
Dallas, Texas, USA
‘THE FUNCTION AND SCOPE OF RESERVOIR SIMULATION
MODERATOR
‘Manfred Wittmann,
AN, since you are a builder of reservoir simulation models, let's
start with you. What does a reservoir simulation do?
AL BREITENBACH
‘The function of reservoir simulation isto solve the fundamental
flow equations in suificient detail so that you can predict the
consequences of changes in operating practice. Successful reser
voir simulation, therefore, requires solving those equations as
precisely as possible, This means two things. First it means
developing a numerical simulator that will give accurate solu
tions for the data input. Second, it means being able to describe
the reservoir with sufficient accuracy so that the numerical sou
tions give an accurate basis on which to make predictions
MODERATOR
‘You used “slficient” twice. What is sufficient detail and su
cient accuracy?
NANSEN SALERI
Sufficiency is reservoirspecific, depending mostly on reservoir
size and complexity, You may have close to sufficient data in
small, simple reservoir, and may never approach itn large or
‘complex ones, no matter how hard you ty. Because data about
reservoirs almost always lack a degree of detail and accuracy,
reservoir simulation cannot be considered a deterministic tool.
is more probabilistic, along the lines of weather forecasting mod
cls we accept reservoir simulation as such, then I think itcan
be put to optimal use,
Oilfield Review
EA. Breitenbach Larry D. Hallenbeck
Scientific Software-intercomp, Inc. Philips Petroleum Company
Denver, Colorado, USA Tananger, Norway
MODERATOR
What is optimal use—why undertake a reservoir simulation?
PRAMOD BANSAL
‘We typically use simulation studies in two settings: In develop
ment ofa new field, a simulation can help us decide how many
‘wells we need and where they should be. In an established field
that has unexpectedly poor performance, a simulation can tel us
‘why and help us find ways to improve performance. in either
setting, itis important atthe outset to define the purpose ofthe
simulation and do the necessary homework before undertaking
the simulation. This homework involves use of simple reservoir
engineering tools, such as material balance calculations and
boundary conditions and aquifer siz.
In the case of poor performance ofan established field, material
balance equations may not tell you why you have poor perfor:
‘mance; you may need more description and simulation
ABDUL KARIM AL-RABAH
| question the value of simulation studies atthe development
stage. Without the benefit of production history, | think results of
simulations during the development stage are doubsful
SALERI
(On the contrary, reservoir simulation may be ati best when
data are sparse. Its an effective tol for comparing different
probabilities influencing flow—pressure-volume-temperature
{PVD variations or variations in the continuity of shales o fault,
or in the connectivity of compartments ofthe reservoir. Simula-
lion studies are the most effective way to develop a sensible
engineering solution to the different possible scenarios
Volume 2 Number 2
D. Nathan Meehan Nansen G. Saleti
Union Pacific Resources Company Chevron Exploration and
Fort Worth, Texas, USA Prod ton Seices Company
Houston, Texas,
AL-RABAH
| agree that when data are sparse you need to un some basic
simulation for planning. But I don't think you should develop a
field based solely on simulation studies if production history is
lacking, If you have some wells, wth atleast a minimum of pro-
dluction history, then you can go back and modify your model
But without that, I don’t feel that | can decide how to develop a
field of 100 wells
SALERI
Unfortunately we don’t always have time to gather all that infor
mation. In some exploration projects we have only 4 to 12
‘weeks to make the best technical evaluation we can, on which
large investments willbe based. In this setting, reservoir simula:
tion plays a key role
BREITENBACH
I think perhaps Nansen [Salei} and Abdul Karim [Al-Rabab) are
talking about two citferent things. When you're considering,
_major investment decisions, like location of offshore platforms
and pipelines, you may not have a great deal of information, but
you still have to decide where and how many wells to drill. Even
if you might not be able to produce for a year or several years,
reservoir simulation isa useful sensitivity tool. It gives you devel-
‘opment strategies for a range of possible reservoir condlons—
the presence or absence of faults, whether they are sealing or
or conbutons to his aril tok aloo Luis Ayestan, Schnee Teil
Senvics, Dus, UAE Ta oul, Seni Sowarhacom, ey Deve Cob
‘ao, USA; an kha Singh shea, Kawa Ca Company, Aes, Kaw
Rourcable conducted in San Antonio, Tes, USA, Octobe 1, 198,
19Donte
andansone
Limesione
se) ese
- Bee.
ronsealing, whether there are aquifers and if they are large or
small above). In many places, simulation has been done with a
paucity of data and the results were taken, unfortunately, as
gospel. In at least one instance, this turned into disaster because
hundreds of millions of dollars were spent on pipeline facilities
and the reservoir in realty was inno way close tothe simulation
description. In this case, people missed the objective ofthe
study. The simulation could have been used correct if, as
[Nansen sad, they took it asa sensitivity study, and eecognized,
for example, that they did not know the stratigraphy and had not
‘examined all the possible solutions for boundary conditions of
the reservoir,
MODERATOR
If we accept, then, that reservoir simulations fulfil these two
functions, when do such studies become economic? Is eco-
nomics related to field size?
SALERI
Economics is based more on what you're tying to find out than
(on reservoir size. And you don’t need complex reservoir models
to answer complex reservoir questions. Simple, tank-type
‘models can often answer many questions about the reservoir.
BREITENBACH
‘More generally, reservoir simulation becomes economic if you
think itis going to change your production practices.
BANSAL
But you can't know a prior that simulation is going to change
‘production practices.
TA ankaype mode conde he reso angle, ple tak conaining ld
‘ate hana srs of ano astm fnrconnected pares Kemgheszes te
‘ternal boundaties ofthese abd gts eral ese heleropenees.
20
BREITENBACH,
In many cases you do. When a company comes to us and
‘expresses an interest in doing reservoir simlation, we ask,
“What exacly do you want to get out ofthis? What is your
goal" What we want to ear is, “We want to increase produe-
tion" or "We're looking fr alternatives to enhanced ol recow-
ry," oF "We're trying to find zones for more recovery.” In these
inslances, there i an opportunity to take some action. But ithe
person says, “We'd like to understand the reservoir beter,” we
28k: "I this gong to have any eflect on production o develop
‘ment practice? And if they say *No, our production is con-
trolled and we won't be able to change ifr several yeas," then
"suggest that we do the simulation in several years
SALERL
‘We canals take a ifferent approach. When people say they re
‘operating in a dosnothing mode, tl them that a do-nothing
mode is actually doing something that could be detrimental to
the reservoir and their careers—maybe inthe reverse order. The
value of doing a study is that it forces people to consider options
tat they otherwise might have overlooked or ignored
NATHAN MEEHAN
How do you quantify the value of sucha study?
SALERI
In Chevron, every study has a dollar value associated with it. The
study is done ifthe given operating company thinks it’s a good
investment. The decision i ently upto the operating company.
LARRY HALLENBECK
‘Who determines priority? You cant do every study for every
‘operating company:
Oilfield ReviewSALERI
‘We have a price structure for eservoir simulation studies that
maintains a eeasonable balance between our resources fr doing
stutlies and the needs ofthe operating companies. We would
have a problem if we dropped the price to an artificially low
level, offering basically free engineering. Even in a large company
like Chevron, computer ancl manpower resources are limited
MEEHAN
| view the economics of resewvoir characterization asthe pat=
‘chase of uncertain information, the value of which should be
estimated. Enhanced oil recovery (EOR) projecs are so hopelessly
aud to quantify
dificult to optimize that their information value is
Bur sometimes value can be easily evaluated. ve recently worked
in tight gas heterogeneous systems, where economies are con-
tolled by facture length and well spacing, There were only a
couple of things you could vary inthe simulation, soit was easy
to determine whether the simulation was worthwhile,
‘BREITENBACH
think the decis
based mostly on cost-benefit analysis. As a manager you have to
ask: “If did a ester ob of simulating this reservoir, what would |
save? Manpower? Addltional wells? Am | making rote oil? Pr:
dlucing at a greater rate?” You need to identify the specific payof.
MEEHAN
So the use of reservoir simulation is to specifically address some.
decision you're likely to make inthe future
BREITENBACH
Exactly
about whether to do a simulation study is
Volume 2 Number 2
SIMULATOR INPUTS
MODERATOR
Lets take the case of reservar simulation for early characteriza
tion of a prospect. What are the inpuls you use?
SALERI
‘The most impegant input in early prospects isthe boundary con:
ditions of the reservoir. As AI [Breitenbach] said eatie, his
includes the extent of the reservoir, the size of the aquifer, the
prese 1 of high permeabilty streaks, the nature of
faults (if presen, sealing or nonsealing, etc. (See “Key Steps in
Reservoir Simulation,” pages 22-23) Is OK if we can't describe
al hese features, as lang as we come forward and say, for exam:
pile, “We don't know the sealing or nonsealing nature of the
reservoir” We must recognize the possibilty of either.
‘The other major point to keep in mind isthe scale of data,
Descriptions of fine-scale heterogeneities are important in some
systems, such as in EOR. But for most two-phase systems, like
simple waterlooding, we're finding out that mexa-and giga
scale heterogeneities are of overwhelming importance (see page
24}, By mega we mean data obtained on the scale of well
tests—150 to 600 meters [500 to 2,000 feet from the wellbore.
The giga scale is thousands of meters In the early phases of a
reservoir’ life, properties onthe giga- and mega-scales really
dictate what goes on. Macro scale data—logs and cores—and
microscale data thin sections -tend to become important later.
HALLENBECK
‘Most of my experience concerns this later stage, since my inter
‘est is how simulation studies can assist reservoir management
2 oF aber
The development of relative
permeability curves ina simu-
lation study should be based
‘on the objective of the study
cand on the amount of field
performance data availabe, I
believe that calibration of fetd
performance data, such as oil
‘and gas production and pres-
sure performance, with simu-
lator prediction is the ultimate
‘method for obtaining reliable
simulator forecasts.
‘i
i:
a
a8) General description of reserve rom:
“Seismic
*Study of other reservoirs inthe same formation and outcrops.
) Correlate major depositional unt.
*Use log and core data to identty low k or om layers that may
fact as batters. Divide reservok into layers,
(C} Determina directonal permoabliy and pay coninuty
‘assess directional permeabilty, facture of channel conto of Fuk
flow with pressure Vansient toss
‘Assess degree of stratfcation.
‘assess vertical communication between ayes.
‘Evaluate pay continuity betweon inocors and producers,
D) Determine lw characterises.
“Obiaincelatve permeabilty and caplary pressure relationships
from natve stato cores, I poceibo, oF rostord stato cores,
‘Assess effect of irjcted us on cor characteristics,
) Determine gard k distibuton in each layer
‘Uso cote data and statistical analysis to map areal itrbution of
ard kin each layer.
‘Determine whether ¢and kcan be correlated
‘Determine i ateral continuity ots
“Dive reservorino Bock,
Key Steps i
‘Divide reservoir nto sections. if appropriate.
*Soloct numberof dimonclone, layors and aroal Docks.
Coosa moda rd so that block boundaries coincide wit changes
in reservoir characteristics determined from geolale stud (or
‘example, 20 that boundaries coincide wih barriers Yo ow)
“Study effect of atc block size on reservoir sensitivity.
‘Use pseudo funetons to account fr layering within each model
layer and numerical ippersion,
‘Determine optimal gid sz.
“Verily and refne reservoir description by history matching the fla
pressure production performance,
“Consult geologst before making major changes in description,
‘Spek the most reasonable ofthe muple solutions.
“Evaluate sonsity of reservokr performance to changes In barr
sterath
‘Choose tho baso case that bost maiches etd performance.
+Evalatealtemative operating plans against the base case using
the model. This includes changes in injection rate, management of
(985 cap voidago (gs injections or blowdown), nfl cling,
Calizing locaton of injectors and preducers, and changes in
Production rate
‘Schedule dling of adcitonal wos
“Plan itu evaluation ofthe progrem,
‘Determine when and how injection and production wos should
be adjusted,
Repeat steps 4 trough 6
Oilfield Review_ Reservoir Simulation : (
(Choose the right.
‘model z
neuro conlatoney | Model constuction
|2nong geooge, : i
| simulation, lab, and) {
eld data -Avoi arbiaioss | i
= FR istry rata |
Not Vaiton fey ita
IN oc
| Predetons [ESS -eoanze
| reser properties,
1 Peeters
Intrprt,negrate and cocoate, rvaly and vertically :
RESERVOIR DESCRIPTION
| Rock Characterization || Fluld Cheractertzstion Geologie Model |
Residual of -Aceal and verical variation |J_|-Volumetic maps
Relative poemoabity Sratication
) capita pressure Barer maps
Oil and water tingereinting +Reservor continuity
8 qualiy maps.
ty key Identity main drive J] Determine volumes (OOP, |
rs ‘mecharisms ‘OaiP, OwIP)
Volume 2 Number 2 )2
|) General description of reservoir rom:
‘Selemic
‘Study of other reservoirs in he same formation and outcrops.
') Cortelata major depostional unit
“Use log and core data to ently low k o low @ layers that may
‘ct as barriers. Divide roservolr nt ayers.
(C} Determine dectonal permeability and pay conn.
‘Assess decional permeabity, facture or channel conto of fhid
flow vith prossure varsiont tests,
‘Assess degre of satiation.
“Assess verical communication botwoon layers.
“Evaluate pay continu between injectors and producers.
) Determine fow charactors.
“Optaln lave permeabiily ard caplary pressure relationships
{rom nalvestate cores, I poss, or restored stato cores.
“/Asg0ss affect of injactod uid on core characteristics,
) Dotermine ¢and k istibution in each layer.
“Use core data and statistical anelyss to map areal distribution ot
‘and kin each ler
*Datormine whether 4 and kcan bo corelated
*Datorring if ateral continuity exists
“Divide reservor into blocks.
“Divide reserva ino sections it appropiato.
“Select numbor of dimensions, layore an area block.
‘Choose mode gid go that block boundaries coincide with changes |
in reseroic characteristics determined from gocloge stu (or
‘2xamle, 6 thal Boundaries coincide with baits to fo).
Study effect of ord block size on reservar sensitiv,
Use neous functions to account fr layering within each model
layer and numerical cspersion
Determine opimal grid size
“Verily and roine reservoir description by history matching the leis
pressure production performance.
“Consult gelogists before making mejor changes in desertion.
‘Seek the most reasonable ofthe multiple solutions.
-EvaluateSanstvly of reservolr performance o changes in bain
strength
sCho0se the base casa that best matches felé performance
‘Evaluate aerative operating plans against the base caso using
the model. Th includes changes in injection rate, management of
(2s cap voidage (ga injocions or blondon), inl illng,
Dptimizing locaton of irjectors and producers and changes in
production rate
‘Schedule cling f addtional wall
“Plan future evaluation ofthe program.
‘Determine when and how ijpcton and production wells should
be agus,
Repeal stops 4 through 6
Oilfield Review
Key Steps in Reservoir Simulation
[Sot clear objecives |
and protize
Tater Sat [ Recenol
tion ING and Toronyi chafacterizaion
RM “Engineering Control in Reser-
‘vol Simulation " paper SPE 18305,
‘Presented at the 63rd SPE Annual oe
‘Technical Conference and Exhibi- ‘Chose te ht
ton, Houston, Texas, USA, October
26, 1988; and attr Meohan DN:
‘personal communication, 1989)
a
Ensure consistency =e! Mocel construction
among goologic, Eis |
simulation ab, and
Model Validation [a=
uy
Predictions — [age
¥
Documentation
eS
interpre. nterate and correlate, aeally and voricaly
ee
cies
Residual i
-Roiatve permesbilty
-Capitary pressure
soll and water fingerprinting
dently main deve
‘mechanisms
Ident key
reserve feats
Volume 2 Number 2
‘Avid arirainass
in history matching
*Apaly engineering
‘contol
Recognize
‘ncaraines In
resent properties,
‘geology, boundary
‘condbons, et.
Determine volumes (OO,
‘OGiP, OwIP)
2BMicroscopic
(meron)
Macroscope
(ches ee)
Pore throat
Elements of Data Scale
Megascono,
(00 000 ee)
Drainage Area of Well
Seales of date at
Which a reservotr is
Gigaecopie
(> 1,000 ee)
Reservoir
‘Our frst criteria is: Does the physics make sense? Is the flow or
recovery mechanism in the model plausible? A case in point is
the Fkotisk field in which we performed a pilot waterflood
study in a part of the reservoir that, according to our laboratory
data, had poor imbibition characteristics.) By matching perior-
_mance data of the offset producers, we began to see immediately
that input data for imbibition would have o be revised to
account for the favorable behavior that was observed, As more
performance data were obtained, a final interpretation utilizing
the simulator was accepted.
MODERATOR
‘Nansen, in your paper you describe a reservoir with a transition
zone of 325 feet, based on capillary pressures obtained from
Core analysis. Closely spaced drillsem tests, however, proved
that in reality the transition zone was shorter. | have had a similar
experience in which, according to capillary pressure measure
‘ments, an entire field would have been inthe transition zone
when it actually produced clean oil. How do you account for
this discrepancy?
SALERI
Its mainly due to differences in scale—macro versus mega- oF
Biga-scale. Also, discrepancy arises when you don't have repre:
sentative samples-—if you have a 10 millidarcy im) sample from,
a reservoir that averages 500 md.
2: hth a acted hak fld inthe Nomen scar ofthe Non Se, with 7
Dilionbaeset on pace
2. Spf Hee LD srl Thomas LX: thas Formation Pilot Watelons,” paper
SPC 10376, pews athe GS SPE Annual Telia Cenftence ard Exim
Heuston ens SA, Oca 5,198
allesseck LD, Sy sD] and Thomas LK: nplerenation of te Ek
Feld Wael” paper SPE 1983, peste athe th SP Al Techni
Confer si ation San Asn, Fes, USA, Oct 6-1, 1989.
4. Sales NG an! orn Rie “trgering Cont nese Silat” paper SP
Ta, preeed he SPE Ara Tf Cnlerece a ato,
Houston, Tors, USA, Octobe 25, 1988
Slack“ a cl ayesha ierchangeshle tems forthe uns sed 0
Grideihe reer The rsewoirefpcly da i go blocks at
‘lored scl nthe srs del Whlope nd perphysical poper
{hs are apne on x ely bs cal sro ceisler
‘which saluion sane rate ef cites equations The fne e
{the fe soliton othe flentlesne,Calls ae anlagou pace
{ona compat sermon the more py, he smoother te ce thal an be daw,
24
MEEHAN
Even worse than that if your sample is 10 md and the reservoir
averages 10 md but has a large standard deviation. A homoge-
neous 10-md core plug will have a relative permeability and
capillary pressure substantially diflecent from the reservoir. A lot
‘of properties that are critical for reservoir simulation are scale=
dependent
MODERATOR
What can you do to overcome this uncertainty in order to make
8 good model?
[BREITENBACH
First we must recognize that rock formations are not made up of
‘one, two or three rack types, but 20, 30 or 40 rock types that can
have widely diferent pore geometries and therefore clilferent
‘capillary pressures and relative pecmeabilies. You can define
these properties by careful well log analysis and subdivide the
rocks 50 that you can take a paucity of data and make a better
thanveducated guess, But, as Nathan (Meehan) said, you run into
problems if you ty to apply an average of the data tothe reser-
voir. When we examined the accuracy of predictions of simula-
tions we did 10 oF 20 years ago, we found that the averaging
problem was best solved by breaking out the rock types as finely
8 possible. Then we went back tothe well logs and foot by foot
divided the well into types A, B, , D, and so on. Then we
applied the appropriate rock properties for each type.
AL-RABAH
Do you construct the rock tables before or afte the history
‘match?
BREITENBACH
‘Before. We start with the kind of cossplotting common to log
interpretation. We look for patterns and clusters in how, say,
sonic and density logs compare, and how different capillary
Pressure curves compare. In most cases, we come up with 10-
plus rock types. Sometimes we have 40 oF 50 rock types. In the
Middle East, we've seen rock types in the hundreds. AS see it,
this isthe only way to solve the kind of questions such as where
the transition zone is in a carbonate, For example, it’ the only
way that you can say with certainty I have a transition zone at
60 percent water saturation that goes all the way to the top of the
steveture.”|Also, diserepaney
arises when you
don’t have repre-
sentative samples —
ifyou have a 10
milidarey (md)
sample froma
reservoir that aver-
ages 500 md.
Water font
Layer
Tanks
Modoiwih Goning
Layer
Tanktype Mode!
Layer
Station
cn
Lidentiication of the main encroachment mechantsms of the
‘reservoir. This isan important objective in simulation because
encroachment affects how the reservolr ls medled—as essentially
‘strated or tank type model. Thorough characterization of usd
ow would indicate which encroachment scheme is right; this in
turn simplifies model selection. Nansen Salori observed that
"seservolr models must not be expected fo impart this under-
standing, only to retina it.”
(Alter Sater! NG and Torony Re, 1988.)
Even worse than
‘hat, i your sample
is 10 md and the
reservoir averages
10 ma but has a
large standard
deviation.
HALLENBECK
The development of relative permeability curves in a simulation
study should be based on the objective ofthe study and on the
amount of field performance data available. | believe that cali-
bration o field performance data, such as oil and gas production
and pressure performance, with simulator prediction isthe ult:
mate method for obtaining reliable simulator forecasts. OF cours,
the relative permeability curves are not the only parameters that,
can be adjusted but are generally the most difficult to aban
‘experimentally. I there are no field performance data, and rela
tive permeability curves from the laboratory must be used, then
the study should include a range of relative permeability
values—pessimistic, optimistic and the most probable,
MODERATOR
‘Another problem is how you constrain the simulator. Normally, a
simulator has an average pressure per block,? which may or may
‘not be what the RFT tool measures. How do you deal with these
potential differences?
SALERI
These differences are generally significant only in stratified reser:
voirs the RFT tool indicates a stratified reservoir with six or
seven pressure regimes, but the simulator says you have vertical
equilibrium, then there's something wrong with the model (Je
We saw ths in West Africa, A reservoie study concluded that the
reservoir was in vertical equilibrium, Yet, several RFT measure:
mens indicated pressure differences of 500 to 700 pounds
between different zones. Just on the basis ofthe RFT measure
ments, one could have recommended peripherally flooding each
layer separatelySIMULATOR QUALITY CONTROL
| MODERATOR
‘What are the steps you take to ensure simulator quality?
SALERI
In every model, we try to identify the key items that are critical to
that reservoir—whether its PVT relationships or the geology or
the aquifer—and test them for physical reasonableness. We also
identity key areas of uncertainty to see that the model adresses
them properly.
‘More generally, I don't think we as an industry pay enough
attention o production logging below). Flowmeters and spinner
surveys and temperature logs give the engineer a valuable three-
dimensional view of both pressure and saturations. Reservoir
engineering is often relegated to interpretation of production data
taken atthe surface, which severely constrains the perspective of
the reservoir. Production logging allows you to monitor the
advance of fronts, profile the pressure vertically ancl compare it
with simulator values. Often, production logs are available but
no one makes the effort to compare them with the simulator pre-
diction, There are two reasons for this. Firs, a general methodol-
‘oy is often lacking. In Chevron, we've come a long way toward
developing and enforcing a methodology, so that critical data are
always examined. Second, comparing production log data and
simulator predictions requires extra labor and computer time. It
aft
0g 00 E100
&
26
-may also require a longer elfortat history matching than simply
#8 field-observed production with model-predicted pro-
MODERATOR
‘Ave there standards you can apply belore you say “I'm happy
with this simulation,” or do you rely on intuition?
[BREITENBACH
‘As Nansen said, there are no real standards. We've looked at
‘mechanizing standard, even arificial intelligence (Al) programs,
bbut with the diversity of problems in reservoir description, we
have been able to come up with litle more than loose standards.
Quality control is necessary today because so much is done by
Intuition, for example, regarding the depositional environment.
Inthe nest decade or so, we will have much finer-scale data that
will dramatically change how we work. We might be working
with 100 million grid coll, allowing us to use a more determin.
istic approach than now possible,
BANSAL
like to mention one practical detail conceming the comparison
‘of predicted and measured data, Iris important to match predict
‘ed pressure against both the shut-in and bottomhole flowing,
pressures. People sometimes ignore the bottomhole flowing pres-
sure, which is very important in determining the outflow perfor-
mance ofthe wel. Performance predicted by the model when
‘both types of pressures have been matched! will be more reliable.
Oilfield Review‘CONTRIBUTORS TO A SIMULATION STUDY
MODERATOR
| sometimes find inconsistencies in how different people within
the same organization approach reservoir characterization I this
Intentional or something to be avoided?
BANSAL
In Mobil, we emphasize multdisciplinary cooperation at every
level ofa simulation, whether large or small, What we mean by
interdisciplinary cooperation is making sure geologists and seis:
‘mic people and reservoir engineers sit down together and rank
the reservoir parameters by importance. The group | am in con:
sults with people from 14 division offices worldwide who are
doing simulation studies. When they solicit our help, the first
thing laskis whether they are working with the geologists. we
are visting them in their ofice, we make sure that the geologists
are sitting with the reservoir engineers. We won't do any consul
ing unless they are sitting together
SALERI
| agzee about the importance of collaboration, but you've got 10
10 beyond just having the reservoir engineer talk to the geo
physicist, the setsmic people, the production engineers, etc
Often those in formation evaluation or seismics aren’ familie
with wh
controls reservoir behavior. I's upto the reservoir
engineers to define the most important aspects ofthe data. And
unless you identify what is atthe top of your data heirarhy, you
face the most common pitfall in reservoir simulation today: too
much data that requires too much analysis and in the process of,
that analysis, alos vision ofthe objective of the study.
HALLENBECK
‘We've been somewhat successul in cross-departmental commu:
nications with geologists. When we embarked on the full-field
Ekofsk waterflood in 1986, we integrated the recommendations
cof the geologists and geophysics into our implementation phi.
losophy. Intaly, the geologists were somewhat eluctant to ded.
icate suicien time, since exploration normally takes priority,
We believe, however, that we have made progress toward more
effective communication
MEEHAN
"had something funny happen to me recently. | gave a talk about,
Union Pacific Resources o engineering students at the University
(of Texas and Texas A&M University. Itold the students at each
gathering that their work would requite clase collaboration with
‘geologists and geophyscists. | wasnt too surprised when a nun
ber of them hissed (ater which I said that at least some of them
can save their bid points and not worry about interviewing with
tus), That type of reaction is part ofthe long-standing cfference in
the kinds of attitudes and behaviors that are rewarded, But the
truth is, we engineers can use a lot of information that geolo-
Bists provide. Geologists do a great service in explaining the
nature ofthe racks, the depositional envionment, the petrology —
informatio
hat we don't use. Conversely, a number of things
ve really do want to know, geologists can't or don't routinely
provide. So our job is to educate them, We need to make sure
Volume 2 Number 2
that the earth scientists we work with know enough about reser:
voir simulation inputs, operations and outputs—not in tremen-
ddous detail—to know which oftheir tools will be most helpful
This cooperation is especially important in geostatistical mode!
ing in which you generate maps that honor variograms or some
‘other model of spatial heterogeneity. You need input from your
geologists to reduce the numberof possible scenario in your
reservoir description, which you then plu into your simulation
‘model. I is impossible to do a simulation for each scenario that,
the geostatisticians develop,
What we mean by
interdisciplinary
cooperation is making
sure geologists and
seismic people and
reservoir engineers sit
down together and rank
the reservoir parameters
byimportance,
27|
PREPARING FOR TOMORROW: RESERVOIR SIMULATION IN THE
NEXT DECADE
MODERATOR
‘What critical developments do we need today to advance reser
vpoie simulation into the next century?
BREITENBACH
| see the need for wo key developments: improvements in field
‘measurement technique to allow us to becorne more de
tic and less stochastic and the evolution of computer tools to
‘make these measurements availabe tothe occasional user. Let
‘me elaborate on the second point, then Ill come back to the
firs.
IFEOR, for example, is going to advance, then we have to
‘make the technology available to people who don't have access
toa first-class team of geologists, statigraphers, petrophysicist,
‘geophysicists and reservoir engineers. The technology needs to
evolve to the point where the production engineer responsible
for a five-well field, who has a background in reservoir engineer
ing, can do a good job—not just a passable job—solving EOR
problems with reservoir simulation. Now, to give the occasional
ser nearexpert results on a desktop system would require qual
ty controls that could be accomplished only through tremendous.
advances in computer systems and software programs—we'd
‘need something like a desktop Cray 1. Such systems are just now
becoming availabe. In the near future, software systems will also
become available to aid the occasional user. An example ofthis
is software for adaptive history matching, which uses statistical
processes and the user's existing knowledge to help him or her
‘overcome a lack of experience and do better history matching
(below and next page)
Nowy to return ta my first point: What do | mean by “becom
ing moce deterministic"? A deterministic system is one in which
1Plem viow of simulated pressure distibu-
ton in one layer of reservoir, produced
by a Sctontise Software Intorcomp, Inc.
(SD program on @ microcomputer. The
Color barat the bottom gives simulated
reservoir pressure in pounds per square
Inch: grid cell numbers appear on the ver-
tical ands. The swirling ‘potential flow
lunes" ropresont the pressure profile ct
steady state, and show the direction of
‘ow ttom injectors (open circles with
‘atrows) fo producers open circles). The
Color boxes (gr cols) show transient pres-
Sure during preduction. The main value of
{his representation is that it shows areas
not swept and shows which Injectors are
supplying flow fo which producing wells.
For example, flow to the well atthe top lett
{s being suppiled by the injector atthe far
right, A more effective injection schome
wold have been to activate the injector
Just below the producer, which was not
‘active inthis simulation
28
you could, for example, follow a three-dimensional saturation
profile in time, IF you could do this, you would have a detesmin
istic system for reservoir simulation for that litle swath of area
that has been swept. You may nat have a precise reservoir
description, but you have something that allows you to precisely,
occurred inthe reservoir, giving you a suficient
duplicate
basis for prediction, Ta allow this to happen, we need research
efforts fr better well-to-well tomography and seismic reservoir
‘monitoring —for all the measurements that allow us to become
move deteeministic
SALERI
'
‘generalizing o the gigascale and integrating diverse types of
information will require even more expertise than exists today.
Actifcial intelligence systems were discussed ata recent SPE
forum.® The thrust ofthe talks was that, despite al the work that
has taken place, lobal Al packages today are no more advanced
than they were ten years ago. For the immediate future, | would
like to see more quality control, specifically, using more produc-
tion engineering data, For the longer term—on my wish list for
‘Christmas 2000—cross-well tomography would give you the
siga-scale picture ofthe reservoir with actual measurements
‘And, of couse, we need harcwvare and software that can handle
the tremendous data volume involved in such a characterization.
BANSAL
‘What will be helpful in the next few years, | hope, is the devel
‘opment of three-dimensional projections of reservoir simulation
‘outputs, Now we have two-dimensional outputs, but to really
tundestand things like steam fronts and temperature fronts in
EOR, we need three-cimensional views,
MEEHAN
‘On my wish lst are more comprehensive models of how natural
fractures influence reservoir flaw. We now live with very simple
see the nonexpert handling very specific problems, But
00) 2160 5A). 278) 204). 1800
si
Oilfield Review15501450
psi
Pio Pie Whee Wier
i
1880
itwodimonsional cross section of simulated pressure distribution citer water injection, produced by an SSI program
(on a microcomputer. By comparing several such simulations, the reservoir enginoer ca plan whore ami wher {0 Max-
‘mize pressure support in tho zosorvor. In this caso, injector well WI20 was placed high in the stricture fo maintain.
‘pressure in producing well P3.A T4Jayer simulation ite thls would typically be redkiced to threo or four layers fo make
‘@ manageable input for tee dimensional modeling.
_models of fracture spacing, orientation, roughness, aperture, per
‘meabilty, capillary pressures and so on. There are many reset=
voirs known fo be naturally fractured that do not show dual
porosity behavior on buildup, drawdowns or fluid displace:
‘ment. We need more fundamental physis research before we
can say why this sso.
[think the applications of ractaP” and geostatistical modeling
for characterizing reservoir heterogeneity show some promise,
Conventionally, specific data poins for permeability, poresty
and saturation are sampled and used to produce a smooth contour
‘map. This technique generally leads to an optimistic fluid dis
placement prediction and does not aecount for reservoir hetero:
‘geneity. Geostatistical methods seek to capture reservoir hetero:
‘geneity and model flow simulations using mace realistic distributions
‘0 permeability, porosity, etc. Chevron’s fractal work shows that
as long as heterogeneity is retained, the flow simulation results
more closely match the actual performance.®
Volume 2 Number 2
SALERI
But even when you're 100 percent successful with fractal or eo:
statistical modeling, you have the problem of putting those data
into a reservoir simulator, limited mostly by computer capabil
lies. At Chevion, we're good at coming up with fractal model,
but we haven't yet had the same success in applying the fractal
‘model to the simulator. We succeed best in smal ields—1,000
{0 2,000 acres. Modeling larger fields would require tens of mil
lions of cells, which is beyond anyone's capability. That may be
another 10 years away.
(“Advances i Resevol Managenent an ik Applic,” SPE Forum See,
7m Fatal ad Rocks,” The Tenia Bee 3600.1 uty 1988; 32-26,
£8. Hen TA: "Pratl Diststior f Revo Heterogeneity an The nice on
Fu aro per 15380, reset the at SPE Atl Teche Cam
ference ad Edits, New Ores, Lotuans, USA, Orober 186
Hewat TA are Bivens RA “Condional unulaton esen0 Hesrogenesy ith
Fats" paper SPE TAS2, preset the teh SP Ara cteea Confrence
ted Ehibiton Howton, lets USA, October 5, 188
29HALLENBECK
‘We've done a lot of modeling with both single and dual porosity
simulators and found that dual porosity simulations require
‘much reservoir characterization that is cutently not availabe.
They also requite many impractical assumptions related, for
‘example, to fractureto-matrx interactions, matrix block height
land capillary continuity. We have been quite satisfied with the
results of single porosity simulations when directly compared to