SPE
-of l+tmfem~
SPE 18427
Implicit Compositional Simulation of Single-Porosity and
Dual-Porosity Reservoirs
by K.H. Coats, Scientific Software-lntercomp
SPE Member
Copyright 1989, %ciely of Petroleum Engineers
This paper was prepared for presentation at the SPE Sympoalum
This paper wee selected for presentation by an SPE Program Committee following review of information contained in en abstract submitted by the
author(a). Contents of the paper, as presented, have not been raviewed by the Society of Petroleum Englneera and are subject to correction by the
author(a), The material, as presented, does not necessarily reflect any position of the Society of Petroleum Engineere, Ite officers, or members. Papera
presented at SPE meetings are aubjacf to publiczflon review by Editorial Gmmittees of the Society of Petroleum Engineers. Permission to copy is
restricted to an abstract of not more than 306 worda. Illustrations may not be copied. The abstract should contain conspicuous acknowledgment of
where and by whom the paper is presanted. Write Publice~ ma Manager, SPE, P.O. Sox 833636, Richardson, TX 75063-3636. Telex, 730969 SPEDAL.
Abstract
This paper describeean iraplicitnumerical model for
com positionalaimulation of single-porosityand dual-
poroeityoilor gaa condensate reservoirs.A 3-component
equation-of-statecom positionalapproach ia proposed as a
desirable alternativeto extended black oil modelling,
requiringlittlemore computing time than the letter. The
approachiaillustratedforan actualneas-criticalvolatileoil
reservoir. A aimple method for reducing implicit
formulation time truncation error is described and
illustrated. A new bottomhole constraint function is
describedfor better preservationof productionwelltarget
ratee in com positionalmodels. A new matrix-fracture
transferformulation includingmatrix-fracturediffusionis
preeentedfor the dual-porositydescription;itsaccuracy is
examined in connection with severaltest problema where
correct reeulte are available from single-porosity
simulation. Results are discussed for a 3D 600-block
aimulationof a highly fracturednear-critical volatileoil
reeervoir.
Introduction
This paper describeea fullyimplicitnumerical model
for compositionalaimulationof m ultidim enaion~ threa-
phaae flow in single-porosityand naturally fractured
reservoirs. A general descriptionof the model is given,
followedby a sectiongivingmore detailregardingcertain
features. The model equationsare then preeented. The
major em phaaia here relates to the fractured reservoir
application. Therefore, the rem ainder of the paper
describesa new matrix-fracture fluidtransferformulation
and estimatesitsaccuracy in connectionwith a number of
exam plaor testproblems.
General Descriptionof the Model
The model iE fullycom poaitimml with a generalized
R&erences and illustrations
at end ~f paper.
on Reservoir Simulstlon in Houston, TX, Fabruary 6-S, 1989.
cubic equation-of-state
1 (EOS) for representation
of gas-oil
phaee equilibriaand deneitiee. The generalized EOS
repreaente the Redlich-Kw ong2 Soave-Redlich-Kw ong3
Zudkevilch-Joffe Redlich-Kwong .!95, and Peng-Robinsond
E OS. A tabular,pressure-dependent K-valueoptionprovides
an alternative
to EOS usage. EOS param etesaare obtained
using a regression-basedpVT program~. Different
parameter setsare ueed forrese~oir and surfaceseparation
calculations. This eases the burden of determining
parameters and increasesEOS accuracy at reservoirand
surface conditions. Viscositiesare calculatedfrom the
Lohrenz et al correlationsand interfacieltension ia
obtainedusingthe M acL eod-Sugden m ethodg.
The model simulates1-,2- and 3-dimenaiortelflow in
Cartesianor cylindricalcoordinates.Darey’sLaw modified
by relativepermeabilityand capillarypressurerepresenta
the viacoue.,capillaryand gravity forcee. Effects of
interracialtensionon capillarypressureare included. The
model applies to depletion,water injection, cycling (gas
injection),
and enrichedgas/solventinjectionoperationsin
reeervoixtypes ranging from black oil to near-critical
volatile oil and condensate to lean gas condensate.
Applicationsincludeaimulationof laboratoryexpenm enta,
cylindncel-coordinatesingle-wel.1
etudiasand are~ crose-
sectionalor 3D field-scale
studiee.
Implicit formulafions generally have a tendency
toward greatacnumericaldispersioneffectsthan the IX PES
formulation. A dispersion-control feature is described
which reduces sensitivity
of reaultato time step aiae in
some caees.
Production well rate is allocatedamong Layers by
pressure and mobility,includingan implicitbottomhole
constrainttreatment to preserveapecifiadtargetrate. The
well rate terms involved are implicitin all variables:
com positions,saturationsand pressure.A new formulation
for the im pw bottomhole target rate constraintgives
better preeenrationof specifiedrate for the case of
compositionalsimulation,
---
239
2 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROSITY ANO DUAL-POROSITY RESERVOIRS SPE 18427

Pseudo or extended black oil models are frequently
used to approximate com postinal effectsin volatileoil
resewoire undergoingdepletionandlor cycling. There are
advantages in many euch caeee in favor of using a
com positionalmodel with a pseudoized, tuned three-
component fluiddescription. These advantages include:
easier preparationof PVT data, increased accuracy in
representationof com positionalphenom ens, and only
m arginzlly
greatercomputing expense.
The code utilizesmapping so that storageis required
only for activegridblocks. For exam plea 49 x 30 x 56000-
block @d having 4200 active bloclw requiresstorage for
only 4200-block arrays. To our knowledge, this mapping
logicwas firstdevisedby Dr. A. D. Modine in 1971.
M&id com positionmay vary with depth throughout
the hydrocarbon column in the case of initially undey
saturatedresenroire. Ttdz depth variationmay cross the
criticalpoint with near-critical
condensate in the upper
portionsand near-critical
volatileoilin the 10w er portions tortuoeity,
of the initial
column.
The model offersthe option of internally-generated diffusion coefficient,different for eac
vertical-equilibria
m (VE)capfiry pressurecurveslO. These
curves vary in time and from block to block. They reduce about 100 times em allerthan those for gae-gaa diffusion.
the errore and occasionaldisruptive“steel-plate” effects Liquid-gascoefficientsare largerbut still
associatedwithuse of rock capillary
pressurecurves.
computed usinga method developedby da Silvaand Belery.
Their method includesdetailedequations for calculating
diffusioncoeftiientsfor each component from the static
variable group of (resemoir ternperature, component
molecularweightsand critical properties),
and the dynamic
group of pressureand com positions,Allth.ssevariablesare
presentin any com positionalmodel or are calculatedin the
aimulation.
Da Silvaproposedsome time ago thatdiffusion can be
importantin fracturedmatrixreservoirbehavior,especially
when injec(iongas com position(e.g.N 2) differsgreatly
from native reservoir gas. Computations chow that
diffusioncan act very rapidlyto nearlyeliminatematrix-
fracturecom positiondifference both in the liquidand gas
phaee. The diffusioncoefficientsand rate expressionsand
theirdenvativeeare coded implicitly
in the m odeL
Table 1 shows the expreseione
for matrix-fracturegae-
gas and Iiquid-gaediffusionrates. The term Tor is
A@/!?isthe diffusion“tranemiscibility”
reflecting
m atr& and @d block dimeneione,S g (Sgg)is a saturation-
dependent fractionbetween O and ?%, and D g (Dgg) is the
component.
Diffusioncoefficientsfor liquid-liquid diffusioncan be
lessthan gae-gae
coefficients.

liiatorymatching often entailsa largenumber of long The diffusionratesare illustratedin Table I for cells
rune w~h productionwell rates specifiedae ST B/d of oil.. 1 (matrix)and 2 (fracture).Tf both cells,remain 2-phase
Until a good match is approached, rune may encounter gas-oiland equilibratewith both the gas-gasand liquid-gas
erroneouslylargecalculatedG O R valueswith resultinghigh diffusionterms shown, then an equilibrium-state dilemma
com puti.ng
expense. A method of avoidingthisis described eriaea. The equilibriumrequirem ent that all matrix
below.“-1 gatheringcenterlogicl1 which has proven weful component K-values be equal is a contradiction. The
infieldstudy predictions
isalsodescribed. problem is resolvedmm ply by usinggae-gasdiffusiononly.
For am allmatrixblocks(e.g. 1-4 ft. cubes),diffusion is so
Th the dual-porosity case,the model nl.lo
ws reseqvoir rapid that gae-gaecliff
ueion alone reeultein sm all matrix-
descriptionewhere the matrix block size and shape vary fracturecom positiondifferencesboth in the liquidand gas
from block to block throughout the grid. Storage and phaaes.
computing time requirem enteare proportional to the degree
of fracturing. For a 1000 gridblockproblem with allblocks An estimateof the diffusiontransienttime isgivenby
dual-porosity, theee requiremente are proportionalto 2000. solutionof the diffusion
equation
W, due to regionalfracturing, only 100 blocks were dual-
porosity, then these requirem enta would be proportional to a2c
—.
1100. SD
axi
For regionallyfractured reservoixdescriptions, the for initialconditionC(XD,0) = 1 and boundary condkione
model treatsthe three cliff
erent types of interlock 50 w:
C(l,tD)= O and 3C/3xD = O at XD = O, where
matrix-matrixElow between blocksin unfracturedregions,
fracture-fractureflow betwe en blocksin fracturedregions,
and I.Iatrix-fractureflow between block pairs on the = x / (!2/2) = Dt / ~or(k/2)2
‘D ‘D
interfaceof such regions, The model can be run in single-
porositymode with virtually no lossof efficiencycaused by This corresporids to the case of a linearcore 8 feet long
the presenceof dual-porositycode. initiallysaturated with fluid of unit concentration and
exposed to zero-concentration fluidat x = O and x = E. The
The calculationsinclude matrix-fractureliquid-gae solutionforaverage concentrationia
and gas-gasdiffusion,
usinga method developedby da Silva k -A t
enll
and Belery12. $s oCdtiQ=2~ 1
~ 1 ~2
n
The linear-solver
code includesthe threeoptionsof D4 where 1 is (2n-1)~/2. * Using first-termapproximation,the
(reduce?band-width)directsolution13, iterative
block SO R time t* %ecessary for C to decay 90% from itsinitial value
(successiveoverrelaxation),and the vectorizedconjugate- ~
gradientESPID O method developedby Don Thurnau.
t* = .S5Tor(%/2)2/D.
Diac&ion of Features
“For gas-gas high pressure (e.g. 4500 p&) diffusion,a
Diffusion
representativeD valueis.001cm 2/see.For a l-ft.core and
a tortuoeityvalue of 3.5,t* = 8 daya. The same analyaia
Diffusionis calculatedbetween matrix and fractures -
performed in threedimewione givesa lower transienttime.
within a grid block but not between grid blocks. The
For practicalpurpoaeathisb inatantaneow.
matrix-fracturegas-gee and liquid-gasdiffusionratea are
---
i?w
SPE 18427 KEITH H. COATS
The Extended Black OilApplication
The extendedblack oilmodel generallyrequires3 and
sometimes 4 (non-aqueous) com ponent8. Itscomputing time
then reflectsthe solutionof 4 or 5 equations(per grid
block). Its PVT treatm ent frequentlyinvotvesmultidimen-
eional tables representingcertain rules establishedfor
presence/absence of various components in the various
phases.
Today’s EOS com positionaltechnology allows an
alternativeapproach. A black- or volatile-oillaboratory
PVT dataset can be used with EOS regressionto obtaina
pseudoized 3-component fluid description. This 3-
component descriptioncan representthe oilin respect to
reservoirvolumetric(expansion)behavior,multiple-contact
revaporizationunder gas injection,
and surface separation
behavior. If C 02 or N2 are to be injected,then this
approach entails4 com ponents$the same as the extended
blackOiL
The 3-com ponent EOS com positional approach offers
severaladvantages. First,the PVT treatm ent isboth easier
and more accurate. Frequently,extended black oilmodels
involve significanteffort to devise and code tabular
representations of PVT behaviorwhich become increasingly
complex. The resultsin some casesinclude:
a) distortedPVT behavior causing com mutational
runningproblems,
b) difficultyin understandingand preparingPVT
inputdata requirem ente,
c) difficultyin designing meaningful laboratory
teststo determinePVT parameters,and
d) difficulty in representing m ultico m ponent
surfaceseparations.
The EOS 3-com ponent approach is easier,involving
repeated multivariable regressionson a collectionof fluid
behavior data. The EOS approach should also be more
accurate. Whatever itscomplexity,the extended black oil
PVT treatm ent attempts to represent what are aimply
com positionalphenomena generallyhandled well by today’s
k O S compositionaltechnology. The EOS regressionon a 3-
component basis allows adjustment of a number cf EOS
param etere and introducesa PVT continuityor consistence
y oiland/or water-oilcontacts,VE Pc curvesgive the exactly
over all pressureasd com postions. All components are
presentin both hydrocarbonphasesin a manner continuously
dependent upon pressureand com positioq.
This is not to say that any collection
of oil(blackor
volatile)
data representingnumerous and differenttypes of
testswillalways be matched accuratelyby the 3-component
EOS description.How ever,in any case the data shouldbe
matched as wellor betterby the letterthan by the extended
blackoilPVT treatm ent.
A common argument in favorof the extendedblackoil
model b that the few pseudo com ponente give a faster
running m odeL Thisis not necesaacilytrue. For the same
number of components, the compositionalmodel frequently
will require littlemore co m puling time. An arguable
conclusionis that the com positionalmodel should have a
black-oiloptionfor the conventional2-com ponent black oil
case, and any extended black oil PVT behavior should be
addreasedby the 3-component com positionalaimulation.
The oil Reservoir A study discussed below M an
exam ple of 3-component com positional
re&sentation of a
volatileoilin a resetwoirsubjectedto depletionand gas
injection.Table 2 givesthe lo-component composition and
some bubble-pointdata for vofi Oil A. Also listedare
241
3
the pseudo 3-com ponent descriptionand correspondingEOS
regressed match of date at the 266 deg. F. reservoir
temperature. Table 2 includes the 3=tage surface
separationconditionsand the regressed match of data for
the 10-com ponent and 3-component descriptions.Figs.1
and 2 show the differential
expansiondata and the regressed
match forthe 10-and3-com ponentdescriptions
Multiple-contact oilvaporizationis a mechanism of
some importance in ReservoirA due to the 86 Z methane
content of injectedgas. Previousexperiencein composi-
tionalsimulationindicatesthat accuracy iu E OS-computed
multiple-contactvaporizationrequiresa splitof the heavy
C 7+ fractioninto several subfractions. In this case no
laboratory vaporization test data were available.
Therefore,the PVT program was used to generatea I&point
vaporization “test” using the tuned (regressed) 10-
component description.This description includesthe C 7+
fractionssplitinto 3 subfractions.The computed results
were treated as “data” and e~.~eredas part of the dataeet
forthe 3-com ponent regression.Fig.3 shows the agreement
between 3-and lo-corn ponent calculationsof multiple-
contactvaporization.
Overall, Table 2 and Figs. 1-3 illustratea good
approximation of Oil A available data using the 3-
component description. The tuned EOS param etere are
cliff
erentforthe resemoir and surfacecalculations.
Vertical-Equilibrium
CapillaryPressureCurves
The laboratoryor r~ck capillarypressure(Pc) curve
reflectsthe relationshipbetween capillarypressure and
‘;::;a?ei ;~~e%~~~% c atblOck
td the point to the
center). Theaverage
VE or
saturationover the gridblock heightintervalencom passing
the point. The VE curve is obtained by integratingthe
laboratorycurve over a capillmy pressureincrement equal
to the product of block height and wate~oil or gas-oil
densitydifference.Thus each gridblockhas a differentVE
Pc curve and the curve changes with time reflecting
changes in fluiddensities.The gas-oilVE P= curve also
reflectschanginginterracial
tension.
Used in equilibrationof rese~oirs having initial
gas-
correctinitialfluids-in-place.
Rock Pc curvesgive emore in
these quantitieswhich increase as the ratio of block
thicknessto transition
zone heightincreases.
U riderdynamic conditions,the VE Pc curves reflect
the underlyingequilibria
m state withoutassuming that the
dynamic fluiddistributionissegregatedor,in any sense,in
equilibria
m. If the reservou were shut-inand ellowed to
aproach equilibrium,the calculated equilibrium fluid
distributions
are More correctifVE ratherthan rock Pc are
used.
Table 3 shows differencesin initialfluid+-place
values calculatedusing VE and rock capillarypressure
curves for a 400-ft. column with gas-oiland wateroil
contacts. Calculations are given for the number of layers
(Nz) ranging from 20 (Az = 20 ft.)to 10, 5, 2 and 1 (Az =
400 ft.).The VE Pc resultsare correctand identicalfor all
Nz values. The rock Pc resultsare good for Nz = 10 and 20
and are only a few per cent in errorfor Nz = 5. The rock
curve resultsexhibitsevereerrorforN z valuesbelow 5.
The VE Pc cumee give better accuracy for a given
number of layersand in some cases allow fewer layersby
preservingaccuracy in regard to definitionor transition
zone saturation”
distributions.They alsoprevent the “steel
4 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROS~Y AND DUAL-POROSITY RESERVOIRS SPE 16427

pLste”conditionwhich occasionally sriaesduringaimulation resultsif only bottomhole liquidcontributesto surface oiL

withrock tunes. Thisterm refersto the case where a grid But in generalthe bottomhole gas phase contributesalsoto
block pair interface is te m poratilysealed to all flow surfaceliquid. Choosing F as the totalbottomhole molar
because (e.g.} a) gas in the upper block dOe8 not want to productionrate w orkareasonablywellin many casesbut can
flow down and oilin the lower block does not want to flow resultin rate deviationsor in additional
outer iterationsto
UP, and b) oilmobility is zero in the upper and gas mobility reduce them ifcom position{~i}ischangingsignificantly.
is zero in the 10wer block. This phenomenon is most likely
to occur when permeability is high, viscous pressure The m ultietagesurface separation results exactly
gradientsare low and rock Pc iElow. It can act in a very satisfythe equivalentsingle-stage
fish equation
destructive or destabilizingmanner in im plici.t
formukiona.
Nc Zi(l- iii) Nc Zi
In the dual-porosity
case,capillzrypressureisassumed 02 = =,x *=1 ~ai (4)
zero in the fractures.The resultingfracturegridblock VE i=l
L(I - fi~+ ki
Pe curvesare:
where
~ VE
= (1-2 Sw+ h(Yw -Ye)/2 (la) (5)
Cwo Ri > yi$/XiS ai E ili/(l-$)

PVE
=(2s ~f - 1)h (Y.- Yg)/2 (lb)
Cgo
Equilibration with the VE matrix and fracturePc curves
gives correct fluid distributions and initialgas-oiland
w ate~oil contacts which are horizontalthroughout the
fracturesyatem. Fracturesaturations are between O and 1
in gridblocksintemectedby the contacts.Also,withineach
gridblockthe matrixand fracturesare in equilibrium inthe
sense that a model run with zero we~ rateswillexhibitno
saturationor pressurechanges. The matrixand fractureVE
Pc curvesare used for equilibrationand interlock flaw but
play no role in the matrix-fracturetransfercalculations.
They are an optionin the program and may be deactivatedif
rock-curve use is desired. Uae of nonzero fracture Pc
curves gives erroneouz initialtransitionzones in the
fractureaystem.
Well Bottomhole ConstraintEquations
In black oil m odela the bottomhole constraint
equations ensuringpreservationof productionwell target
rates are easilydetermined. h compositional models,
effectiveor proper constraintequationsare more difficult
to determine, The compositionalmodel equationsexpress
productionrateas m olsfd
= ~ ~gp$yi)k
p~QoPo x.+ ‘~wb)k
%k
where i and k denote component and perforatedV; -r,
respectively.For a given value of pwb, total w e~tream
molar rate q and com positionZi are easilyobtainedfrom
this expression. The multistage surface separation
calculation then givesthe surface8TB/d oiland M CF/D gas
ratesand com positions.U the welltargetrateiaspecified
in uNte of STB/d, then an iterativeprocedure determines
Pwb so that the surfaceoilrate equalsthe specifiedST B/d
rate at the beginning of each outer iteration. If no
bottomhole constraintis used, the well W.B/d rate at the
end of the iterationwillnot equalthe specifiedvalue due to
changesin q and Ziover theiteration.
The bottomhole constraintfora wellisan equationof
the form
6F = ~(;(a~sp~k) “ O
where Pj denotesthe COm positional m’odelvariableset { .
{YO P, S wv SO, Sg, and 6 Pj ie the new ~erate change, ~f
-. The outersum isover allperforatedgridblocks.1!he
4
objectiveis choice of an appropriatefunctionF euch that ,
invarianceof F over the iterationensuresinvarianceof the
ST B/d rateover the iteration.ChoosingF es the expression
for totalwellbottomholeliquidphase molar rate givesgood
L is mole of etocktank liquidper mol of feed and subscripts
denotes surface com poeitione. Multiplyingnumerator and
denominatorby w ellstreammolarrateq gives
Nc qi
(6)
F=E—= o
,=1 qL .+~q
and this is chosen es the constraintfunction F. qi is
w ellstrea
m molar rate of component i and q isstock tank
rate,mole liquid/d.Assu mptioneare: over tk e iterationas
q and {Zi}change,(a)the Cii vzluesrem sin unchanged and
(b)ifqk rem sineconstantthen the conesponding ST B/d oil
rate rem sinecmstant. Both aasum ptionshappen to be very
good ones. Thw the constraintequatin is
N
c aF
6F=E (7)
j=l ~ ‘qj = 0
The variablesin F subjectto the differentiation
are qiin the
numerator and q = Z qi in the denominator. qL and ~ are
known constants(latest&erate valuea). Use of Eqn. (7)
together with Eqna. (6)and (2)and chain-ruledifferencing
leadsto a constraintequationin the form of Eqn. (3).
This constraint’
equationgivesgood resultseven when
(2) the bottomhole gas phase isthe sourceof most or allof the
surfaceoiL In the case where specifiedwallrateM M C F/d
gas,a eimilarprocedureleadato a constraintfunction
Nc qi
F=T— (8)
~=1 qv + ~iq
where 8.i= 1/(1- ?i)and surfacemols/dgas rateqv and 8i
are heldconstant.
Tim e TruncationError Control
Fig. 4 ‘sho wa a resewoir% total GOR vs. time
calculatedfrom the fullyimplicitform@a&n with and
without the time-truncation error control method. The
reservoiris a 3D 8 x 13 x 6 single-poro~ grid represen-
tationof the fracturedReservoirA describedlater. The
(3) 4380-day emulation run was made using91-day and 45-day
time steps. The figureshows that withoutthe errorcontro~
calculatedG O R is significantly
higherusingthe larger91-
day steps.
The open and solidcircularpointsrepresentthe same
9l-day and 45-day step rune performed with the error
controL The sensitivity of resultsto time step size b
reduced ei@ficantly. Another point of compa~om is

242
SPE 18427 KEITH H. COATS 5

average fieldpressure,reflectingdi.fferencea
in c&uleted The occurrence of thisevent is flegsedin the output well
producedgas,as follows: eum m arias(end of step and end of run) and the overall
impact of the featureis noted by printingan oil-deficiency
Final Final variablein each step summary. Oil-deficiency is the total
Tim e Outer cumulative shortage of oil produced (STB) due to this
Run s&s Ens. C%~ol
—— SCYS; B % featuredividedby totalspecifiedcumulativeoilproduction.
-i- 48 m No 15600 m
2 96 194 No 13000 4357 Until a descriptionis tuned to avoid emoneous high
3 4s 104 Yes 11900 4429 G O R valueswe have a choiceof procedure. We can not use
4 96 194 Yes 11500 4463 the Q G M A X feature and aimply let the wells blow out,
indicating the mismatch. This type of run can take (say)
T&LA fieldpressure was 6943 Paia. The error control hours of computing time due to very smalltime stepaandlor
featurereduces the 91 vs.45-day step run pavg cliff
erance step cute due to large changes accompanying blowout. Or
from 175 paito 34 paL we can run with the Q G MA X featurewith (say)20-minute
runa with high oil-deficiencyvalues. Both rune indicateor
This controlfeature relatesonly to im miscibleflow, tellw the earne information- our G O R istoo high- but the
reducing sensitivityof saturationdistributions, W O R and latterruns tellus that in faster,smoother rune. Aleo,the
G O R to time step aiae. It does not controlthe compoai- Q G MA X restraint actsto preventor reduce the phenomenon
tionalsmearing or numerical dispersioneffectsassociated of one w ell’ablowout destroyingother w ells’smatch or
with m iacibledisplace
m ent. behavior. Lf GOR isvery high,the reservoirvoidageeffect
of a wellon Q G MA X isnot much differentthan itapressure
A subtle im pl.ication
appears in the above tabular effectifit were producingspecifiedoilST B/d with correct
results. We prefer implicitover IMPES formulationsin GOR - provided Q G MAX is equal or near the actual gee
cases where maximum IMPES time step is small and rate.
computing expense consequently high. However, the
implicit formulation generally requires much more In the Resenoir A studydiscussedbelow, the Q G M AX
arithmeticper step than IMPES and to come out ahead (of values were obtainedfrom the historyE& for each walles
IMPES) we must use a largetime stepin the implicitmodeL the maxim u m surfacegas ratesproduced over allof history.
If,however, as shown here, the implicitmodal time step The end-of~tep wellsummary printoutgivesoildeficiency
requiresconstraintto reduce truncationerror,then we are by well,showing at a glance which wells are the major
defeatingour purpose. The 45-day step Nna requiredtwice culpritsin the Q G M A X constraint
- Le. which wellshave the
the computing expense of the 9l-day step rune. Thus a greateatG O R problems.
method of reducingtruncationerroris important in regard
to computing expenseinsome cases. GatheringCenters

The error control method requixeslittleadditional
storage and virtuallyno increasedcomputing time. It is
very aimpie, consistingessentially of a Crank-Nicholeon
treatm ent of relativeperm eabilities.
For each phase,the
upstream relativeperm eabilityused for the new i+l outer
iterationis
~ f.+1
= .5(krn + kr(Sg+l))
r
A violentinstabilitycan resultwhen kr issmall and S
is decreasing. Methods of protectingagainst this are
num eroua and of variouscomplexities. We simply use the
A
fullyimplicitk= = %(s ‘+1) when thatsituationexists.We
findno benefitfrom extandingthisprocedure to capillary
pressure and do not apply it to ~ in the dual-porosity
matrix-fracturetransferterms.
Maximum Gas Rate
Specification of ST B/d oilin historymatch runs where
G O R isa problem can resultin a semistableprocess. As a
well G O R risesabove itsfieldvalues,“blowout”can occur.
That is,as G O R risesabove observedthe I.erSe freegee rate
causes rapid pressuredecline- a pressuresink - which in
turnresultsin even higherG O R and highergas rate -and so
on. Also, excessivegas rate at one well can affect its
neighbors- reducing pressureregionally, resultingin their
erroneouslygassingout as w elL
The model provides for data entry of a maximum
surfacegas rate (QG M AX) for each productionweLL The
model then produces the specifiedST B/d oilrate for each
well unless the associatedgas rate exceeds the w en’s
entered Q G M AX. In the latterevent,the wallproducesthe
Q G M A X gas rateas ti targetin placeof the ST B/d oilrate.
The wells are assignedin groups to any number of
gatheringcenters(GC). A given G C generallyincludesa
number of productionwellsand gas andlor water injection
wells.Assignm ent of a G C productiontargetrate resultsin
allocationof that rate among the G C producerson the basis
of their current open-flow potentialsor deliverabilitiaa.
Alternatively,individualproduction well ratee may be
(9) specifiedwithom isaionof a G C productiontargetrate.
G C injectiontarget rates may be specifiedin three
forms: a) absoluteunits-MC F/d gas and ST B/d water, b)
fractionsof the G C total M C F/d gaa and ST B/d water
productionrates,and c) fractionsof the total reservoir
voidage rate represented by the G C producers. The
resultingtarget injectionrates are allocatedamong the
appropriatetype G C injectorson the basisof theh current
injectionpotential or injectivities.
In the case (c)above,
the R B/d voidage rate is converted to MCFId gas andlor
ST3/d water injectionrates which at current bottomhole
pressuregivethe desiredR B/d rate.
Ih the case of gas injection, the model provideafor
entry of sales gas, fuel gas, and makeup gaa G C values.
Entered salesand fuel gas values are subtractedfrom the
G C produced gas to determine gas avd.able forreinfection.
If the targetinjectionrate exceeds availablegas, makeup
gas is added to meet the target. Provisionis made for
injectingmakeup gas Erst and meeting any difference
between availablemakeup gas and targetrate withrecycled
produced gas. Injectiongas com positionis calculatedfrom
the com positionsand am ounta of produced and makeup gas
constitutingthe reinfection
etream.
If no G C productionor injectiontargetsare specified,
then all individualwell rates must be specifiedand the
asaignments of wells to gathering centers becomes an
exercisefor cosmetic (printou@purposes. The G C feature
—
243
 6 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROSITY
is option~ that is,the model can be run with no wells
assignedto any gatheringcenters.
The Model Equations
The conventionalsingleporositymodel equationeare
firstretiewed. The model consistsof Nc + 1 primazy
equations expressing conservation of mass for N c
hydrocarbon(Ii C) components and water in each gridblock.
The H C components include N 2, C 02 and H2S and are
insolublein w ate~ water b absentin the oiland gas phases.
An additionalset of N c + 3 constraintequationsapply in
each gridblock,includingNc equationsexpressingH C phase
eqdibriu m (equaloiland gaa phase component fugacities),
and three equationsexpressingunitsums of H C phase m 01
fractionsand phase saturations. The 2NC + 4 unknowna
co~espondin to these 2NC + 4 equationsare denoted Pj and
are in order fxi}${yi},p,So,SgsSw.
The time differencenotationhere is,for any quantity
or variableX,
8X z Xn+l - x“ (lOa)
6X. = x’+’ - X2 a Xn+l -x’ (lOb)
where superscript1 denotes outeriterationand subscriptn
denotes time step leveL The new iterateapproximationia
obtainedas
Xg+l . ~g + 2NC+4 (~ ~, ~pj
(11)
~ apj
A fulldescriptionof these equationa,theu linearization and
reductionto a set of Nc + 1 equationsin the Nc + 1 primaty
unknowna ~ b given elaewhere14. For each gridblock,the
Nc + 3 conatreintequationsare used to eliminate N c + 3
unknowns from the Nc + 1 primary equations. The
remaining set of Nc + 1 primary unknowns is denoted $, a
column vector of mol fractions,pressureand saturah.nns
dependent upon whether the blockisthree-phase, two-phase
H C-water,or allwater. The resultis an equationforeach
gridblockof the form
c&~ = A(TA6~)+& (12)
where bottomhole pressurevariableterms are omitted for
clarity,C and T are (Nc+l) x (Nc+l) matricesand P and R
are column vectorsof lengthN c + 1. Allphase m ob~ties ~
the transmissibility matrix are upstream. The well
injection/production
terms are implicitlyexpressedin C and
& For each bottomhole pressurevariable,an im p~cit
— constraintequationexistsas describedabove.
For the dual-porositycase, the grid block mass
balanceEqn.(12)is
C13~ + Cm Wm = A(TA6~)+B
+ cm (~mn - g;) (13)
where the term C ~ representsthe mass storagecapacityof
the matrix in the grid block. All terms with no m or f
subscriptare fractureayatem terms; matrix variablesand
terms carry thesubscriptm. No matrix-fracture transfer
rates belong or appear in this mass balance. The mass
balanceequationsforthe matrixinthe gridblock are:
- ‘D Sgg Dg ((~gy~m - (~g y~~
. A grid block of height A z containsa stack of n = A zlh
- T ‘aopoxi‘p + ‘gp#i (‘p + ‘ego))
-..
244
AND DUAL-POROSITY RESERVOIRS SPE 18427
= V ~ (@(PoSoXi+,Pg$#)m i= 1,2,.0.Nc (Ma)
-TAwbw(Ap-~cw~=V8($b S ) (14b)
wwm
where V here is A XA yA z (l-@f)and T is given by Eqn. (23)
below. On the leftaides,all mobility,0, x~ yi and bw
values are upstream - matrix or fracture,depending upon
flow direction.AP iS porn- pof. The diffusion area S g is
calculatedas the geometric mgan 2 Sgm Sgf / (S m + gf).
The pseudo capillarypressuresPc are chacussedbe?BOW.
These Nc+l primary equationsin 2NC+4 unknowns are
linearizedand N C+3 unknowna are eliminated using the
constraintequati.ons14. The result is a set of Nc+l
unknowns expressible
in matrixform as:
cm Wrn =T(6~-6qJ+g (15)
where T and C m are (Nc+l)x (N +1) m atnces. As previously
describedfor the black oilcase1 7, m u.ltiplying
thisEqn. by
the inversematrix C-k and inserting the resultingexpression
for 6~m into Eqn. (13) gives the finalsingle m ateri.al
balanceequationin terms of fracturesystem unknowns.
66g = A(TA6~) + & (16)
All coefficientmatricesor column vectors(excepting6~)
containknow n alem entsevaluatedat the lateatiteration.A
linearsolvergivesthe solution6~ and a matrix and column
vector saved from Eqn. (15) manipulation then allow
calculation
of the matrix ~~m unknown.
The Matrix-FractureTransferFormulation
Assureptioneand Definitions
Several authorsl5-17 give detaileddiscussionsof the
dual-porosityfracturedreservoirdescriptionaddreseedin
thiswork. As describedby Warren and Rootl5,the complex
fracturesystem is idealizedto a network of intersecting
horizontal and vertical fractures. The dual-porosity
assumption/descriptionpicturesthe matrixrock as an array
of discontinuousisatrixblocks in the continuousfracture
system. The effectivefractureporosityissm au the matrix
providesnearlyallthe fluidstoragecapacity.
Effectsof viscousgradientsin the fractureon m atrk
fracture transferare neglected. The calculatedmatrix
response to changing fracturesaturationsassumes phase
segregationin the fractures.There is no directflow Erom
matrix block to matrix block within a grid bloclq nor is
there any directflow from the matrixblocksin a gridblock
to those in an adjacentgridblock. Most of the published
dual-porosityformulationsl 6-23 assuree that all matrix
blocks in the grid block exist at the same maturation.
Litvak22 and Rossen and Shen23 are exceptions. The
formukion here assumes thae matrixblocksaturations vary
withverticalpositionwithinthe gridblock.
The capillary diecontinuitiesat the horizontal
fractureshave an adverse effect on ultimate recovery
com pared to a system having no horizontal fracturee.
Horizontalshale streakswould act in the same manner as
fracturesinsofaras the capillary
discontinuity
and effect on
recoveryare concerned.
If the fracturespaciugaalong the Cartesianaxes are
Lx, Ey, and J?.z,
respectively,then a gridblock of volume
Ax AyAz containsAx AyAz (1- @f)/&xAy $2 matrixblocks.
matrixblockswhere h is matrixblock height(Lz),and n is 1
—

SPE 18427 KEITH H. COATS 7

or greater. The complexity of the matrix-fracturetransfer Models of the type describedhere representa stackof
formuladon described below reflects en attempt to n matrix blocksby a singlegridor nodal point. There is a
accuratelyrepresentthe case where n isconsiderably larger limitto the accuracy obtainablewith euch coarse,one-point
definition.Test or example problema need to show the
than 1. A grid block is referredto as a stack of matrix
blocke withoutliteral im plication
that s gridblock has only inaccuraciesas well as accuraciesof a proposed transfer
one stackof matrixblocks.Obviously, thereare roughlyAx method. Ideally,a m ethod’s areas of accuracy and
inaccuracy should relate to dominant and subdominant
A y / Lx Ly identicalcolumns or etacksin one gridbloch
The term aspectratiodenotesthe ratioof block heighth to aspectsof resenfoirbehavior,respectively.Admittedly,it
its lateraldimensionEx. is difficult
to generalizein a definition of what the latter
are.
H any authorsstateor imply that capillarypressureis
The Shape Factor
zero and relativepermeabilityis linear(kr‘= S for each
phaae)in the fractures.Here, theee are assumed physically
Warren and Rootl5 introduced a shape factor U to
correct fracture properties. Fracture nonzero Pc or
relatematrix-fracture
pressuredifferenceand flowrateae,
nonlinear~ may be introducedfor methodologicalor other
purposesbut are not attributedas real propertiesto the
fracture8. q= c“ (Fm-pf) (17)

for the singla-phaeecase, where q is the flow rate in an

The VE Pc optionaffectsonly the interlock fLow aud
element V of bulk reservoirvolume and pm is volumetric
initial distributions;the matrix-fracture transfer iE
unaffectedby itsuse or nonuse. N onzero fracturePc may average matrix pressure. For a cubic matrix block of
dimensionL and quasi-steady-state(QSS)flow,they gave
be entered as data but it affectsonly the interlock flow.
N oul~nearfracturekr willaffect interlock flow and, ae
upstwa m values,the m atrix-fmcturetransfer. (18)

where N isthe number of normal eeteof fractures,1, 2, or

The term immersion denotes the subjectionof au oil-
3. Kaze mi et a116proposedfor finitedifferencesimulation,
saturated matrix block or grid block to a fixed fracture
environm erttof all gas or all water. The term partial
immersion refersto exposure of a grid block to a fixed (19)
fracturegas/oilor water/oilcontactlevelbetween the top
and bottom of the block. A frequentlyunstated dual- or 12/~2 forthe cubicbl~ck case. The valuerecoinmended
here is
porosity assure
ption is that the recovery rate from an
im m ersedstack of n matrixblocksequalsn times that of a u= 8(>+ ~)
++&2 (20)
singleim mersed matrixblock. x Y z

I There & littleinterestin the im m ereiontransientin More generally,this work aimply uses an exchange
cases where both the trueand calculatedtransienttimes are transmiasibil.ity
definedby
small - e.g.60 days or less. Inaccuracyof a formulation’s
transientcalculationwillhave littleeffect on aimulated (21)
reservoirbehaviorinsuch cases.
For a singlematrixblock,
This and othersimilarpaperspresentexam ple problem
resultsreflecting theirtranef
er formulations.These results
shed littlelighton the questionof accuracy unlessthey are
com pared to correct results. This is possiblefor eingle-
matrix block problems,for some singlegridblock studies, and forthe gridblockof bulkvolume A x A y A z,
and, rarely,for fullfield-acale problems. These correct ‘kx k kz
T =8(— +2+
resultsare obtainedby single-porosity modeling.usinggrids ~ ) Ax AY AZ(l -@f) ’23)
which subdividethe matrix and fracturesinto grid blocks. If t; z
Resultsfrom these single-porosity and dual-porositymodel
runsare designatedSP M and D P M results,
respectively.All For QSS single-phase flow,the diffusivity
equationcan
SP M calculations are performed usingrock Pc and kr data. be solved to give exact valuesof u for any anisotropyand
An effortis made to presentexample problems where the any matrixblock shape. Appendix A givesthiscalculation.
Rem.Ltafor an isotropic, cubic matrixblockof dineneionE
correctSP M resultscan be obtained.
for N= 1,2 and 3 aetaof normal Eractureeare:
Thomas et all7 describeda dual-porositymodel and
presentedresultsfor severalexam ple problems. In certain
d 2
cases, our model can duplicate their matrix-fracture
Warren Analytical Kazemi This
~ansfer method. Therefore,forseveralexample problems
our D P M refwitaare com pared to those obiainedusingtheir JJ & Root QSS-~~w et ai Work
1 12 4 T
method. The term D P M T denotes resultsof their method
generatedusingour m odeL 2 32 2S.45 8 16
3 60 49.58 12 24
Any calculatedeffect of injectionto or production
from the matrix in a dual porositygrid block is error.
Physically,a 6-inch diemeter wellbore can intersector The ahape factor and lengthy aseocieteddiscussions
penetrateat most one of the many matrix block stacksin contributeconfusionwith littlebenefit. Single-phaseQSS
the grid block. The only possibleeffect of the matrix is flow isseldom of interestin the resemoir simulationbut if
introductionof a skin factor. This is true even if or it were the above recommended and Piazemi’s shape factors
especiallyif horizontal fractures are absent and the are 2 and 4 times too 10w, respectively. TWO blocks of
w ell/verticel
fractureorientations
are viewed such that the differentshape can have the aam e shape facto~ but block
wdlbore intersectsno fractures. shape,especially aspectratio,can be a dom &ant parameter

‘- ’245

,.
8 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROSITY AND DUAL-POROSITY RESERVOIRS SPE 18427

in reservoirbehavoir. The shape factor definitionloses gas flow are three-dimeneionelb the matrixbut the oilflow
m caningaltogetherforanizotropic
matrixpermeability. iesubstantiallyverticaldownward.

The shape factorisneitherhelpfulnor relevantin the A key characteristic of this process is the virtual
matrix-fracture transferformulationdescribedhere. The absence of viscouspressuregradientin the gas phase. That
two matrix block transmiaeibiWes of imterestare T (Eqn. is,gee phase pressureessentially obeys 8PI 8z = Yg where
(22))and gee exists,and oil mobility controlsthe dtainage rate.
7= = kztx~+tz (24) C apilleryforces act to retain oil in the matrix and at
equil.ibriu
m (largetime), the gee saturationdistribution,
The former givesttearly correctresulm for the imbibition curve ABC in Fig. 5 M the section ABC of the capillary
process. Calculationof T from the shape factorand Eqn. pressurecurve, The volumetricor integratedaverage gas
(19)essentially
givesimbibitionrates correspondingto an saturationat equilibria
m isdenoted sge~
effectivematrix perm eability1/2 itstrue value. Tz is the h
transmissibilityconnected with the gxavity drainage Sg(z)dz (25)
‘ge=~o
procesf5. J
At equilibrium,capillarypreeeure is zero at the
The transientdecay time for single-phase,mat+ bottom and h Ay at the top of the block,varyinglinearlyas
fracture flow ia generallyvery sm elL Aa developed in
Appendix A; the ratio(~m-pf)/ (pi-p?isabout .03at a time PC=(h-z)AY (26)

t = .1 P ($C k2 / .00633k where Ay iey. - yg~the ga=oildensitydifferenceexpressed

in pai/ft.This Eqn.‘givesdz = -dPc/ AY at equilibriumand
where matrix initiallyat pressure pi ie subjected to a Eqn. (25)becomes
hAY
constantfracturepressurepf on allblock facea and FM is .& Sg (Pc)dPe
volumetricaverage matrixpressure.For properties ‘ge (27)
!o
I.l=lcp $= .3 L=lofio k=lmd C=10 -5 p~-l which is“the area Aabcd of Fig. 5 dividedby h AY. If
thresholdcapillarypressurePce exceeds h AY, Sge ie zero
Thisgivesa 97 Z transient
decay at a time of and no drainageoccun% If capillarypressurevariea with
interfacieltensionthen Eqn. (27)is
t = .1(1)(.3
)(10-5)(100)
/ .00633(1)= .0047days hAY/ur
1
Sg (Pc)dPc (28)
‘ge = mr
Equilibnum Saturations
1
where Pc in this Eqn.Oia the invariantrock (laboratory)
Any matrix-fracture transfer method has two matrix capillarypressurecurve measured at tension@ and
fundamental properties. The fket is the grid block q is U@. Sge varies with pressure and com positionin
equilibriumcurve, the relationship - between grid block accordance with theireffectson the value of AY/~ For
fractureand equilibriummatrix saturations.The second is pressuredepletion,both increaseae pressuredecreases,but
the nature of ths transientby which the equilibriummatrix ~ increasesfasterthan AY. The rasultis that Sge can
saturationia reached under conditionsof partialor full decreaseeigticantlyas pressuredeclines.
im m ereion. The gridblock equilibriumcurve depends upon
values of.the matrix block equilibria m saturationsSge Fig.6 shows drainageand imbibitionmatrht water-oil
and Swe. Sge @ we) iZ the matrix block equilibriumgaa capillarypressurecurves. we defer diacueeionof the_case
(water) saturationreached by fullim m ereion of an oil- of a matrix block in the initialtransition
zone and”consider
saturated matrix block in gas (water). Matrix block an oil-saturated matrix block with Sw = SWC and Sg “ 0.
ultimate oilrecoveryin pore volumes is equal to Sge and Im m ereionin water resultsin oilrecoveryby imbibitionand
‘we - SWc for the gas and water im m ereion caaes, gravitydrainage. Imbibitiondominatee the earlytransient
respectively.The matrix block equilibriumsaturationsare with fracturewater flowinginto allsixfaceaof the matrix
defined firstand then used in expressingthe grid block block. This process slows se matrix water saturation
equilibria m curve. approaches the value where imbibitioncapillarypressure
equals zero (point C on Fig. 6). Gravity drainage then
Thornas et al proposedthe use of pseudo or VE matrix continueswith water enteringthe matrixat the bottom and
and fracturecapi31ary pressurecurvesin the drivingforces laterallyalong the aides in much the-same fashion as
for matrixfractureexchange. That proceduregivescorrect describedfor the gas/oilcaee. During thisI$Jtedrtia.ge
matrix block equilibriam gas and water saturation for the process,oilflows out of the block only at the top; none
im m ereioncase. The equilibriummatrix block saturations flows laterallyout the aides. At equilibria m, capillary
Sge and Swe describedhere shouldequaltheirvalues. pressureia zero at the top and - h;: at the bottom Of the
block? where AY ia Yw - Yo, watepoil density
The gasfoil gravity drainage process has been difference,peifft. The average matrix block water
iescribedand studiedby many authors for decades. In saturationat equ.ilibriu
m iZSwe)
particular,van Gol.f-Racht24gives a detailedanalysisfor
:he case of drainageaccompanying the im m eraionof an oil- Ih Sw(z)dz (29)
s we -5
Jaturatedmatrix block in fracturegas. Fig. 5 illustrates ii-
{
thisprocess.G se flowsintothe matrixblock at the top and Capillary-gravitatio%considerationssimilarto those of
laterallyalong the sides. The lateralgaa entry rate is the gae-oilcase give
highestat the top and decreaeeeto zero at a heightPee/ AY o
rnbovethe bottom of the block. Lateraloilflow rate iazero sw(Pc)dPe (30)
s
at the block’sverticalzideaand oilleavesthe block only at we = h ! .hAy
thebottom. Due to the viscouspressuregradientinducedby
the downward oilfLow, the point of zero lateralgas flow which ia the area Aabcd of Fig. 6 dividedby hAY. At
intothe blockisabove pointC in Fig.5 in earlystagesand equilibrium,water saturations
at the top and bottom of the
moves downward toward C as time increases.Both oiland blockare thosenoted by pointsC and B,respectively.

248
SPE 18427 KEITH H. COATS 9

The grid block equilibriumcurves are now discussed. where T is given by Eqn. (22). The ~c terme are pseudo
The gridblockisa stack of n matrix blocke where n is 1 or functionsfor approximate representationof 3D unateady-
greater. Litvak22 pointedout that partia.im m eraionof a state gravity and capillaryforces active in the matrix-
grid block resultsin drainage (imbibition) only in those fracturetransfer. Most publisheddual-porositytransfer
matrix blocks above (below)the fracturegas (water)leveL formtdations are equivalent to these equations; their
Thus for the gas-oilcase, the grid block fracture and exprgasionaof capillarypre~ure and.gravityterme define
equilibria
m matrixgas saturationsare rebted by the Pc terms in Eqna.(35). Pcgo and PcW. are functioneof
● (Sgm~ ?gf) and (Sw m ? S Wf)s respectively. The phase
‘ge = ‘gf ‘ge (31) are upatream valuee. The fracturem obilitiee
m obilities are
calculatedfrom linearcurveskr = S for each phase normally
Thisrekionahip appliesforlarge n and isapproximate to a but nonlinearfracture~ data may be entered.
data-dependentdegree for n = I. As an exception,for zero
capillary
pressureitia exact for alln, includingn = 1, and The con~on qw + q. + qg = O wows e~mination of
‘ge = 1 - ‘erg - ‘WC* A p from Eqna. (35)to obtain&the3-phase transferratesin
terms of the pseudo functioneP c
For the w ate~il case, the grid block equilibnu m
b +~ ;
curve can be determtied exactly for any value of n. We qw=- T@o+x ‘t
g Cwo g Cgo) (36a)
firstconsiderthe case of n =‘1 or A z = h, the case of a
singlematrixblockand itsadjoiningverticalfractures.The
VE capillarypressurecurves for matrix and fractureare q.= T+ ( aw 6CW0 - Xg ~ ~go) (36b)
easilydetermined. The fr~ure VE curve isEqn. (la). For
each SWf, we calculatePcf from this Eqn. and find the
equilibriummatrix,blocksaturationfrom the matrix VE Pc qg’.~(kwfiewo+(a w+ao)tCgo)
A (36c)
curve at thisPc value. The resultis shown by the upper
curve on Fig. 37, for an example problem discussedlater.
The data are those of Table 4 and matrix block dim ensiona
where A = Aw + X. + Ag.
sre 10 x 1“0x 30 ft.
For the water-oilcase Sw = Swc, ,lW= O and mass balance
For the case of intermediaten, n = 3 for illustration
gives,
here, the grid block curve can be constructedusing the
upper (n = 1) curve. The resultis the step-functioncurve ‘wf bm ~
qo’T— = v~m (isWm / dt (37)
shown on Fig. 37. For large n, the grid block equilibrium Cwo
curveis ‘wf + km
s*
we
= SwfSwe + (1- Swf)
Swc (32) For the gas-oilcase,Sgm = Xgm = O and mass balancegivee,
Our D P M formulationuses thisEqn. in a form coreplicated
by considerationsof transitionblocks and hysteretic * $ .V$m@gm, d~
qo=-T (38)
behavior. M kdle Eqn. (32) appliesonly for large n, it is
~f + brn Cgo
alteredas followsfor improved accuracy at small n. ‘She
Swf valueisreplacedby
Integration of these equations gives the saturation
(recovery)transientin analyticalform for given fracture
;Wf = (Swf + s; f)/(l+ S;f) (33) saturations.
swm
where S~f isa displacementwish a defaultvalue of .1. The
kWf + ~m ~ T
resulting
gridblockequilibnum curve =—
wm t (39)
\ ‘$m
S:e = SWf Swe + (1 - iw+ Swc o ‘wf %m CCWO
(34)

for S~f = .0856isshown on Fig.37 and agrees more closely I s

withthe correctcurve forn = 3. ‘m%f+%m ~ .: ~ (40)
\
The Tran8fert%quation8 o ~f %m ~cgt ‘m ‘@m
Ifthe forms of the pseudc functionsare known then the grid
The pseudo capillarypressures in Eqne. (14) are block equilibriumcurve giving m atrik Maturation as a
defied here in the contextof coatrixco fracturephase flow
functionof fracturematurationis obtained by settingthe
ratesat constantpressurewith no mass transfer.This&la
pseudo functionsto O.
clarityand d0e8 not affecttheirdefinitions.Thisaimpl.ified
fiame work is similar to that presented by Rossen and
For the gas-oilcase,the well-knowngravitydrainage
Shen23. The phase transferratee expressed in reservoir
equation describing the initialofi rate response to
volume unitsfora singlem atri.xblockare,
immersion is24
= T kg (Ap + ;Cgo) (35a)
‘g q. = Tz kom (hAy - Pee) (41)

q. ‘TAO AP (35b) where AY b Y. - Yg. The overalltransientreflectsa

complex interactionof matrix Pc, hog And block height
variable8. The baaiaof the method describedhere is the
qw = T Xw (Ap - $Cwo) (35C) observationby van G elf-Racht24 that many researchers
h?~e found experimentally and com mutationallythat the

247
10 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROSn Y AND DUAL-P OROS~Y RESERVOIRS SPE 18427

transient in m ereion response often approtimates the The sign of the term may be positivewhen oil is
exponentialform actuallycontinuingto drainto the fractures.A gridblock
-x*t, may exist at some time near the equi.libnu m condition.
s = Sge(l-e (42) Then a decrease in Sge and an increaee in S f may
13
eimultaneoualyoccur in relativedegreessuch that tt e term
where X* isconstant. Our SP M calculations agree with this signis potive. Let Sgfi be the fracturecontactlevelat
observation. We adopt this form becauee if gives better the beginning of this situation. Then above Sg~ no
resultsover a wide range of data.sets
than any other method exchange takes place. Below the contact in the vertical
we have tried. intervalSgf - S fn, matrix blocksare becoming exposed to
,.. fracturegas wti
“% drainageresulting(m atri.x-to-fractureoil
~ pcgodefin.ifionissoughtwhich exhibits the fonow~g flow) in that interval related to the new lower Sge.
beha=o~ Additionalcom plexitieacan be descibedat greatlength.

a) The im m ereiontransientobeys the exponential For brevity,the equationsof additional logicare given
form Eqn. (42). without lengthy identification of terme with phenomena.
b) The gridblock equilibriumsaturations obey Eqn. Two arrayaare carried.The firstisS fmx. Sgfmx isreset
(31). equalto S f at the end of every step w‘$ere the term signis
c) For partialimmersion, the tranaientrem sins negative ?norm al drainage is occurring). The second S*
fivafint with Sg/Sgfva time independentof Sgfi variableis resetas matrix S mn only when (a)the end-o#
d) The initial oilrate response obeys the form of etep term rngnispotive, an5 (b)S fn c Sgfmx and the new
Eqn.(41) sBf > s .fm*. Two .aaee are co~ered for a positiveterm
e) Im m ereiontransientaccuracy for a given rock ~. #he firstcase entailaa poei.iive term sign together
type (kro , Pc data)is presened ae block height with Sgf < Sgfm x. Imbibitionoccurein thiscase. The term
and interacid
! tensionvary. iD Eqn. (43)u multiplied by (Sgfmx - Sgf)and the fo~w~g
additivecapilhmyterm appears:
Skippingderivational
details,
the reeultis
(sgfmx
-‘gf)
‘Pcgo (s
g -sgf sge )-Pee)ur (44)
A T hAY-P The im bibi.tion
capillary
pressureisused here.
z
P
Cgo=r
b
g sge
Ce (l+!!W- ) (Sg-SgfSge)
l“o~gf The second case entailsa positiveterm signwithSgf>
(43) Sgfmxo The terms X and Y are defined
X5 S*-S
g gfmx ‘ge
togetherwith the “constant-~”contion thatkrog = 1 iI’Ithe
calculationof k.m in Eqn.(35b). 6g is a parameter easily (45)
determined by comparing SP M and DP M single-block YE S-s s-x
g gf ge
im m eraiontransients.Itisa singleparameter foreach rock
type,not a differentparameter foreach gridblock. Default The term in Eqn. (43)is replacedby Y. If Y is negatbe~
valueis 1. .~ the constantA optionisdeactivated,the D P M drainage is occurring and no additional terms are
im m eraion traneientexhibitatoo much curvature and a introduced.If Y ispositive,imbibitionisoccurringand the
larger ~g value is required. The for-m ~43) s~tiefies(a) followingadditiveterm appears:
because the left-handaideof Eqn. (3S)~ hear UI s~ The (Pcgo(Y)- p )u ifX~O
coudition8(b)and (c)are exactlysatisfied.Condition(d)is ce r
satisfiedby inspection. The degree of satisfactionof (46)
condition(e)is generallygood but is only approximate and (Pcgo
problem (data)-dependent. ‘Sg - ‘gf ‘ge)- ‘Cc)or if X < 0
The capillarypressuresare drainage or imbibitionvalues
Real fieldproblems exhibita complexity of gas-oil depending upon whether Sg is increa~ng or decrea~g~
exchange behavior which seldom appeare in test or respectively.
conceptualexam ple problems. The fo11owing describedlogic
represents only one of many possible approaches for The w ateeoil case is now addressed. For clarity
approximatetreatm ent of thiscomplexity. The signof the subscript m is omitted from all matrix saturalione,
term Sg - S f Sge in Eqn. (43)isimportant. A negativeaig: m obil.ities
and capillarypressures. Fracture saturations
indicates ~rainage proceeding toward equilibrium. carry the subscriptf. Subscript w is omitted since all
positivesign may or may not indicate oil flow from saturationsare water saturations.Fig. 7 illustrates the
fracturesto matrixin the gridblock. With no additionsto initialwater saturationdistributionin a grid block lying
Eqn. (43)a poeiIivesign will. resultin such fracture-to- within the water-oiltransitionzone. The block’eupper
m atrb.oilflow. Whether thieis correctdepends upon the intervalbetween pointsC and E containsmobile oilwith an
source or cause of the positivesign. A positivevalue can average water saturationSb at Capil.lavpressurePCW The
result from S increa~g andlor from Sgf andlor Sge 10wer intervalbetween pointsE and H containsimmobile oil
decreasing. I!?the positivesign only reflectsincreasing with an average water saturationSa. The gridblock@l
interf
aci.eltensionand 10w er Sge then oilflow to the matrix averagematuration is
shouldnot be allowed. The upper block region,above the
fracturegas-oilcontacc leve~ containsthe m aerixblocks Si = SFi Sa + (l-sF#~ (47)
drainedto a Sg value largerthan the new (decreasing) Sge.
But there is no oilin the fractureaoppositethese blocke where S .is the fractional distanceof the pointE up from
avdable forim Libition.Without additional logic,E qn. (43) the bot%m of the block is not fracture water
‘Fi
willreeultin falseimbibi.~onof fractureoilfrom below the saturation.Sfiis zero for the case shown. The pointSb,
gas-oilcontact into the matrix blocks above the contact. Pcb lies on the grid block matrix VE capitky pressure
Thisin turnresults in model G O R valueserroneouslylarge. curve. Tn terms of areas,the saturationare

24s
SPE 18427 KELTH H. COATS 11

=A ~big/Az AY Sa ‘sFi=O
‘i
% = Aabed/(ze - Zc) AY (48)
= Si = S:c = Swc
‘b
‘b = Adei~(zh - ze)AY

where Ay isyw - yo.

The maximum grid block-watersaturationattainable

I s = ‘fs+(l-sf)swc

(55)
correspondsto recovery of the mobile oil in the upper s:=
‘Fwe + ‘1 - ‘Gswc
intervaland is

s = sFi Sa + (1 - SF) (1- Sorw)

Wro
The block may have an initialsaturationSi near or above 1-
(49)
s* =Swe+ —
s: -
s (Swc - Swe)
s: - Swc
Sorw and yet hava a significantamount of mobile,
recoverableoiL The entered kro” curve is stored es a The matrix-fracture
water-oiltransferE qn.(37)above is
functionof normalized water saturation(S~ - Swc)/(l-Sorw
- Swc). For blocks where Swro exceeds 1 - Sorw, bow
(56)
valuesare calculatedusing(Sw - Swc)/(Swre- Swc) for this
norm alizedsaturation. This resultsin oil mobilityat all
A
water saturationsup to S“ro.
The capillary
portionof P cW. is
A-
Only the matrix blocks in the upper interval of
average saturationSb wti imbibe water. As fracturewater ‘0+ ‘Wf ‘Wf
; ~Pe(6) (57)
entersat the bottom of the gridblock,no responseOCCUrS. Cwo = A. +~wf
‘his non-responsecontinuesuntilfracturewater levelrises
to pointE. As the levelrisesabove pointE, imbibitionand
The term ~ Wf is how Sf/p w where ~ow is matrix ~w at
gravitydrainage occur as an increasingnumber of upper
imbibitioriPc = O. Thisreflectsthe effectivecorrectinflow
intervalx atrixblocks become exposed to fracturewater.
value of Thomas et SL The imbibitioncontinuesuntilPc =
The imbihkion rates of the upper intervalmatrix bIocks
O. Thereforeany negativevalueaof the imbibitionPc curve
reflectrock imbibitioncapillarypressureforces (not VE
are set to O. Effectsof any positivevalue of imbibitionPc
capillarypressure). These rock capillaryforces are
at 1- Sorw are alsoeliminated.
associatedwith the initialcapi&wy pressure Pcb of the
blocksat saturationSb, not any totalgridblock saturation
The gravitydrainaseportionof ?CW. is treatedas an
or Pc value.
additiveeffectand stems from the bazi.c
equation

For Sf > SF~ the gridblocksaturationat any ti.m

e is ~. = Tz (3wAoh AY, (58)
s= sFi Sa + (Sf -sF/ s+(1 -Sf)Sb (50) analogousto the gas-oildrainageequation(41). The term
$W is a parameter of defaultvalue 1. The gravitydrainage
where S k-the average saturationin the verticalintervalSf portionis
- SFi and S i = Sb. The imbibitioncapiJLsry
pressurecuwe is A
storedas a functionof normalizedsaturation = : ~whAYY(i) (59)
P
cWo
(51)
i= (sw - Swe)m - Sorw - Swc) where Y isa straightlineequalling~l
at S* = Sb and O at S*
All scanning qurves are the same functionof norm aliaed = Swe. The totalwater-oilpseudo PcW. reflectingcaPfirY
saturationbut S iscalculatedas and gravityforcesisthen

; ❑ (s* - S:c) / (1- Sorw-S;J (52)

=~
Ao+Awf if
F 4PC(3)+7 ‘z f3WhAYy(~) (60)
The value of S~c for each scanningcurve is calculatedso Cwo ~o + bf Awf
the curve passesthroughthe pointSb~ pCb as ShOWn on Fig.
6. The matrix block equilibrium saturation Swe is For norm al cases where imbibitiondominates gravity
Adjka/hAy. The corresponding gqoilibriu m grid block forces,~w remains 1. For significant gravityeffects,SP M
saturationforany Sf corresponds to S = Swe and is im m eraionresultsare used to determine a @w valuegivinga
good D P M match. @w is a singleparameter for each rock
(53) ,type(saturation table),not a separatevalue for each grid
s: = ‘F~sa + (Sf‘sFi)‘we +(l-sf)~ block.

The term S* is The above treatm ent is considerablymore complex if

a featureallowingtrapped water ti activated.~ fracture
S:-s water advanaes then recedes, the temporarily exposed
s* z Swe +— (Sb - Swe) (54) matrix may retainitsimbibed water. Thisdepends upon the
S:-si slopes of the drainage and drainage scanning capillary
pressurecurves. In this case the model treats two grid
SettingS to Si givesS* = Sb SO imbibitionPc equalspcb at block matrixsaturations,S and S2 (haverun out of symbols)~
time O,satisfyingthe equ”libriu
m requirem ent. withEqn. (50)appearing

For the simpier case of a gridblock lyingabove the

s= ‘~i ‘a + (Sf - ‘~~ g + (1 - ‘f)‘~ (61)
initial transition
zone

249
,,
12 IMPLICIT COMPOSITIONAL SIMULATION OF SINGLE-POROS~Y
[did S2 illSb. S2 is recalculatedby maze balance
considerationseach time step in a manner dependent upon
whether fracturewater saturationincreasedor decreased.
The resultis thatif a gridblock were to stabi.kzeat some
S~ the matrixsaturationabove Sf k not Sb but .some larger
saturation S2 reflecting earlier temporary periods of
exposureto and imbibikionof water.
All calculationof Sge) Swe, Sa! Sb$etc.in the a~~ve
equationaare internallyperformed in the m od~ requmng
no externalaimulations
or calculations.
Discussion
Dean and L021 describe sever~ formul.etione,
&Iclud@ duel-porosity.For the ga~il caae they generate
a matrix block pseudo capillarypreseurecurve using full staticbaaically
im m eraionSP M resultsin Eqn. (38). This givesthe co=ect
Sge value. They then generatea pseudo fracturecapfiry
pressurecurve which givesconect equilibriummatrixblock
saturationafor partialim m ereion. They apply the same
procedurein the wateroil case. They statethat Thomaa et
al used a matrix pseudo P= whilethey use pseudo Pc curves
for both fractureand matrix. In fact,Thornse et al used
both fractureand matrix pseudo Pc curves. Thornas’curves
give the same correct matrix block equilibriumsaturations method. They illustrated
for fulland partialim m eraionas obtainedby Dean and Lo.
m addition, Thomas’ method does thie automaticalLy problem16. This problem is treatedin detailbelow and is
(internally) for matrix blocks of any size or shape and for
changing Pc o (with tension),requiringuone of the SP M
matching efFort of Dean and Lo. An advantage of the Dean
and Lo procedureliesin itstransientaccuracy which should
be exactin the fullim m eraioncase.
Both of the above methods representthe gridblockby
a singlerepresentativematrix block,similarto other duel-
poro~y formulationa16S19. The partial im m eraion ~
transientsand equilibria m saturationsare those of a single
matrix block ratherthan those of the gridblock (stackof
matrixblocks).
Rossen and Shen23 specifieda fracturepseudo Pc of
SgfhAY (essentblly), then generated the matrix pseudo Pc
usingfullimmersion SP M resultein Eqn. (38). Throu8h a
aimpie, clever transformation they obtain matrix and
fracturepseudos which give exactLycorrecttransientsand
equilibriummatrix saturationsfor allSgf on a @ b~ck
basis<e. for a stack of matrix blocks. They appliedthe
same procedureto the wate~oil case.
Our difficultiesin use of a method similarto Thornas
et al were as.follows. Reduced gas-oiltransientaccuracy
was obsened as h, kro,andlor Pc data were changed. As a
aimple exam pie, coneider the case of negligiblegas-oil
capillary pressure.The m alrixVE krog curve ~ a str~ht
linein thiscase,independentof the rock curve. The dual-
porositytransient isthe same for m hog curv:s. Kowever$
the SP M transientahows wide variation for ~f erentbog
curves. 7n the wate=iL case,the SP M imbibitiontransient
reflectsthe rock capillarypressure curve regardlessof
block height. However, the vertical’ equilibriumpseudo Pc
cutie @e gives a dual-porositytransient rate which
bcreases with height. Loosely speaking,the errorin the
dual-porositywater imbibitiontraneientis proportionalto
the differencebetween the pseudo and rock Pc curves. If
theix method b appliedto the water-oilgravitydrainage
case (negligible rock Pc)sthe transient emor can be ~r8 e
simply because the fullmatrix blocktranzmiaeibility
isused
(when water is present)es opppaed to the gravitydrainage
z-directiontramsmieaibility.A reasonablepresum pti.oniz
that they intendtheirwater-oilformulationfor use in the
norm alcase where imbibihn dominatesgravitydrainage.
ml!”
Zml
AND DUAL-P OROS~Y RESERVOIRS SPE 1S427
W e tried m ethoda aimilar to Roazen and Shen,
generatingpeeudo kro andlor Pc curves from SP M results.
Different h/~o/Pc com binationz gave different pseudo
curves,requiring~in general differentpseudo cuzvss for
each grid block. Even if only one or a limitednumber of
blockhedghtawere allowed,the pseudo curvez changed with
time due to density and interracialtension changes. In
addition, pseudo curves were different for different
positions in the i#tieltransition
zone(a).Finally,for some
dataseta,pseudo Pcgo curves wkh negativealopeoccurred,
leading to com mutationalinstability.The pseudo curves
apply to the drainage(gae-oil) processbut do not represent
oilimbibition.
Sonier et el19 emphasize the need for dynamti m ode~
ratherthan previouslypubliehedstatk m odelz. Their term
referato the need forexternalgenerationof
some parameter(s). In particularthey referred to the
Tho m aa et al model aa static. In fact it is dynamic.
Thornas’ use of pseudo capillaryprezzure curves reflecte
both capillarypressureand gravityforces with no external
calibrations or parameter.
Soaier et al pointed out deficienciesin previous
formulations’ gravityterms and presented theirimproved
the accuracy of theirmethod in
connection with the Kazemi et al five spot waterflood
briefly sum m arizedhere. The vertically fracturedrezervok
is a 600 x 600 x 30 ft.five-spotquadrant with 10 x 10 x 30
ft. matrix blocks and an 8 x S x 1 DPM areal gridis used.
Therefore matrixblock and gridblockheightsare equaland
there are no horizontalfractures. Data provided(Table4)
include matrix block perm eability, porosityand ~ and Pc
data and nonzero fracturePc and nonlinearfracturekr data.
Here and below we use zero fracturePc and linearfracture
Thishas no effecton,theobservations and conclusions.
Sonieret al showed that theirnew gravityterms are
so correct that with capillarypreaeureezeroed and only
gravityforces active,their model closelyreproduces the
nonzero P= Kazemi et alresults.Thisissurp-g becauze
the nonzero Pc Kaze mi problem is imbibitiondominated;
gravity forces are insignificant in the m atxix-fracture
tranaf er. Results with capillarypressurezeroed and only
gravity forces active differ widely from those with the
givencapillary pressure.
Since we may be in errorin understandingthe Sonier
et al method, theirequationsare reproduced here se used
for the zero Pc water-oiltestproblem describedbelow. For
w ate~oilmatrixto fractureflow they write
% = -Tkc (AP + Yo(zwf - zwm))
(62)
~w = T Aw (AP - Yw(zwf - zwm))
where fora singlematrixblock
(~= .08)
T = .001127 LX !tYkz u k
(63)
and
z ~f = (Swf - Swf$ t= (64al
s -s
zwm = 1 :~orwm wmi - Swmi ~z (64b]
In this problem oilflowa from matrix to fractureand watel
in the oppositedirectionso for claritywe omit subscripts m
and f on nobilities.EliminatingA p from theirequatimu
usingq. + qw = O gives
 KEITH H. COATS 13
SPE 18427

(65)
20m parieonwith Eqn. (37) shows their equivalentgravity
lrzinagepseudo functionis
; = (Y. +Yw) 2Z (Swf -Swfi
Cwo
s -s . (66)
- Iw-msorwwmm: sWmi)
For the im merzioncase, Swmi = Swc, Swfi = O and Swf = 1
n the above equations. The above equationsreflectonly
:hei.rgravity ~erms. Capillary pre~ures are zero &
~ccordancewiththeixapplication to Kaze mi’sproblem.
The zero Pc immersion transientequation for their
method ia given by equatingq. from Eqn. (65)to (V/5.6146)
im 8Swm/8t where V is matrixblockvolume ~x~y~z. We
;olvedthisequationwith resultsshown in Fig. 8, usingthe
iataof Table 4. Zero Pc SPM resultsusinga S x 5 x 10 grid
iedcribedbelow are shown forcomparison. The error(h our
:esults usingtheirmethod) is eo largethat we alteredtheir
method a~ follows.In a laterpaper Sonieret e120statethey
?eviaed theix gravity terms by using phaae density
differencesrather than phase densitiesin front of their
iepth terms, as describedby Litvak. They alsostatedthis
:hangeresultedininsignificant differencesfor practicaland
realistic problems. We don’tunderstandwhat is m cant but
sasume it may somehow resultin Yw -Y. appearingin Eqn.
:65)in place of y w + yo. Resultsusingtheirmethod with
they w - Y. substitution are”shown in Fig.8. Again,their
ae w gravity term exhibits largeerroron the”highside. Fig.
B alao shows the zero Pc D P El resultsfor the method
flescribed in this paper. Their (revised)gravity term
significantly overestimatesthe true gravityresponse(Pc =
0). One source of errorin theirgravityterm isthe use of
total-block T in place of the smeller,correctz-direction ~z
Eor gravitydrainage. In additiontheirnew gravityterm
results in gravitydrainageeven when oiland water densities
areequal(seeEqn. (65)).
s
Tim e 5X5X1O 2 x:: 10 2X1XI0
Days krw7w32 kw~”z %Wrl
m .2750 .2753
1200 .3003 .2993 .3001
2000 .3326 .3301 .3319
3200 .3786 .3726 .3754
The SP M resultsare essentiallyindependentof ~wf
valuesrangingfrom .2 to 1,a factorof 5. For drainage,a
DP M formulationshoulduse krwf = Swf and,in the gas-oil
case,~gf = Sgfi
W u and Preuza25 presented a duel-porositymethod
allowing matrix block subdivision(MIN C) and com pared
resultswith conventionalD P M results.Their MIN C resulta
compare very closelywith SP M results. In part, they
concluded that for single❑ atrixblock studies,theirMIN C
method gives more reliablebehaviorthan the conventional
D P M. They ahowed that MIN C (matrixblocksubdivision) is
neceaaary or desirable when fracture water saturation
change ia rapid,when block size or oilviscosityare large,
and when matrixpermeability ‘ialow.
We considera case where fracturewater saturatioti
change is rapid.- the im m ersion case of instantaneous
change from O to 1. The data are the Table 4 data withthe
10 x 10 x 30 ft.1 md UIatrixblock. In the gravitydrainage
case discussed above the saturation gradients in the
5 x 5 x 10 SP M gridat 500 days are large. Near the block
bottom, water saturationsrange from facialgrid block
valuesof .37 to .53 to the W .25 at the interior block.
The D P M one-pointrepresentationof such gradientsby a
I
m“ngle average value might be expected to show error
indicating “a need for matrix subdivision.Howevar, Fig. 8
shows the D P M and SP M resultsagree w elL
For the imbibitioncase (Pc = Table 4 values)Fig.36
shows SP M and our DP M resultsagree welL W u and Preusa
alao made. this calculationand showed moderately poorer
SPM - DPt4 agreement. Fig. 9 shows our SPM and DPM
resultsfor the case where matrix permeabilityia 10wered
tenfoldto .1 m d, The significantly greater accuracy of
MIN C (SPM here)k. evident. However, a practical question

With capillarypressures included in their pseudo

function(6.6),
two additionalerrorsappear. First,setting
the pseudo function to zero doea not give correct
e@librium matrix block or grid block aaturationa(Sw e}
S we). Second,the fracture~wf value for imbibitionisnot
constrainedto krw m (pcWO”O) as suggestedby Thornes et ZL
ariaearelatingto the shape of the D P M curve. Uncertainty I
in actualmatrix permeabilityand historymatching resultin
adjuatm ents to match reservoirbehavior. The questionis
whether a reasonableD P M permeabilityadjustmentaimply
gives a curve crossingthe true curve with no overall
improvement in match or whether the D P M curve shape is
such thatsignificantlygreateroverallagreem ent OCCUTX.

A subtleaspect of the above zero Pc gravity drainage The,open circleson Fig.9 chow D P M resulta%r km =
SP M calculationsreLetea to the @roper value of the .14, a 40% adjustment. The agreem ent is significantly
upstream krwf value. By analogy to the Thomas et al better.Large-timeagreement isgoo~
recom m endation for gas-oil gravity drainage, a D P M
transferformuletionshould use the matrix ~wro value for Tim e s wm
upstream ~wfy or (presumably)SWf x kr~ro for the partial D aye 5X5X1O DPM km = .14 md
immersion case. Our gas-oildrainagecalculations to date
e .5488 .5593
show the gP M grid resultsare independent of whether
5000 .5S12 .5912
upstream ~gf is&go (theirrecom m endati~ or 1.0. The
10000 .6230 .6306
reasonfortluaisthatgas entersthe blocklaterally over the
20000 .6536 .6608
aides as well as from the top and the effectiveentry
transmissibilityis much largerthan a 1D verticalpictureof
The questionof need for matrixsubdivisionisone that
the processwould indicate. The aam e situationexistshere
can be argued endlesslywith littleresolution
or agreement.
in the zero-Pc water-oilgravitydrainagecaee. Following
Meaningful answers are problem dependent. Nevertheless
are SP M w atex=oilgravitydrainageresultsforthe 5 x 5 x 10
we have seen littleneed for subdivision, especiallyin
and 2 x 1 x 10 grids with ~wf = krwro = .2 and for the
respectto other errorsourcesin dual-porosity modeling. Tn
2 x 1 x 10 grid with ~wf = 1. These resultsare for full
the case justtreatedoilviscositywee 2 cp, the block was
immersion of the 10 x 10 x 30 ft. block w~h Table 4 dsta
10 x 10 x 30 ft. and permeabfiy was .1 md. The Reservoir
except thatPc = O.
A describedlater hae .23 Cp o% .5 m d permeabilityand

251

I
14 IMPLICIT COMPOSITIONAL SIMULATION OF 81NGLE-POROSITY
maxim u m block size of 4 ft.cubes. Withoutbelaboringthe
variousscalingcriteria, there is aimply no need for sub-
divisionin thiscase.
To some extent, their conclusionsregarding D P M
inaccuracymay be affectedby the D P M model they used.
For exam pie, they pres.mt D P M imbibitionim m ereion
resultsfora 10-ft.cubicblockwithdata givenby Thomas et
‘sL They show ratherpoor agreement between MIN C (SPM)
and their D P M results. However for the same problem
Thornas et al showed their D P M matched SP M “results However, for the high-ratetestat W = 1200 CC,observedL
virtually exactly.
CornparisonWith ExperimentalData
Kleppe and M oree26 reportedimbibitionoilrecovery
data for a ed.ngle cylindrical matrix block with an annular
verticalfracture.The matrixblock was .324 ft.in diem eter
and 4 ft. in height. Their data are given in Table 5.
Constantwater injection rateat the bottom of the core was
3.3 and 35 cc/min for their low and high-rate tests, recovery. ICIfact,there isbasically
respectivdy. These re8ultedin roughly constant rates of
fracturewater levelriseof about 11 and 140 ft./dfor the
two tests.
Beckner et a127 eimulated these expenm ents in
single-porosity and dual-porositymodes using a numerical
aimulator.TheirEine-grid SP M resultsagreed wellwith the
low-rate test data but showed almost none of the rate
dependency necessaryto match the high-ratetest. Our SP M
resultsshown in Figs. 10 and 11 exhibitsignificant rate the velocityof the water flowingthroughthe fracture.
dependence; agreem ent with data is good at low rate and
moderate at high rate. A 5x12 cylindrical-coordinate grid
was used and isdescribedin Table 5.
There”is no point in pursuingthe SP M match of the
high-ratetestdata because Figs.10 and 11 indicatethe data
are in error. The SP M calculatesrecovery and water level
both lower than observed. But this cannot be, since if
recovery is 10w er than obsezved then water level must be
greater than observed. Thus a materialbalance erroris
presenteitherin the data or the SP M. The latter%m aterhl
balancesare 1.0 to fivedecimal places. We returnto this
matter below.
Beckner et al performed dual-porosity mm ulatormns
and reported “The inabilityof current double-porosity stack of n matrix blocks equals or approximates the
.simulationto model imbibitionwith an advancing water
levd was clearlyshown..2’.Fig. 12 compares our DP M and
SP M resultsfor the 10w- and high-ratetests.Agreement is
good at the 10w rate. The high-ratecase com pariaon
exhibitsa differencesimilarin type (but much less in
magnitude)to that which they reported.
Returningto the questionof high-ratetestdata error,
we firstqualifyour usage of the term “data”. The data
shown here were interpolated from figuresin the Beckner et stack approximatea that of a singlematrix block initially,
al paper,obviouslywith some attendanterror. Moat cases
involving comparisons of theory (calculation) and
experimentel data involve an unresolvable question of
whether the data might be wrong and the calculations flows to the verticalfracturesabove the gas/oilfracture
correct. Fortunately in this case, the nature of the
expenm ent and reported data allow a material balance
check. Defining
w= Curnulativewater injection, cc
o= Oil recoveryfrom the core (matrix), cc
. Annularfracturearea = 8.2743cm 2
;= Heightof waterin the fracture,cm
yields a aimple m aterisl balance requirement before
breakthrough,
AilU DUAL-POROSITY -.. RESI?RVOIRS
—--- . . . . ... ----- liV197.-.
.QPR
1
o- W-AJt (67)
Since the water level and oil recovery data are nearly
straightlines,the materielbalancecheck may be performed
at any one point priorto breakthrough. For the 10w-rate
testat W = 1400 cc, the data are Z = 99.3cm and O =
575 cc. Eqn. (67)gives
o . 1400 - 8.2743(99.3)= 578 CC.
and O are about 102.5cm and 440 cc,respectively, and Eqn.
(67)gives *
0= 1200 - 8.2743(102.5)= 352 CC.
The discussionof rate dependency and the manner of
plottingoil recovery in this experiment are somewhat
misleading. They give the impressionthat there is some
meaningful adverse effect of water injectionrate on
.noeffector,arguably,
a favorableeffectof rate on recovery. Ultimaterecovery
of course is independent of rate. Plottingrecovery vs.
cum ulativewater injection shoWS, for any fixedvalueof the
latter,decreasingrecovery with increasingrate. However,
a plot of the same recovery vs. time would show, at any
fixedvalue of time, higherrecovery at higherrates. The
basic recovery process(imbibition) isindependentof water
injectionrate; a given portion of the matrix willimbibe
water,once water isavailable to it,at a rateindependentof
The matrix propertiesof thisexperiment providean
exam ple of the type of problem in which the nature of and
ability to match the transient are of no’ practical
consequence. The transient imbibitiontime in thiscase isa
sm all fractionof a day, which for practicalpurposes is
instantaneous. Any conceivable error in the calculated
kaneient would have no effect on aimulated resetvoir
behavior.
The Cascade or Reinfiltration Effect
The publisheddual-porosity matrix/fracture exchange
formuletione reflect an implied assure ption that under
im m ereiongravitydrainageconditions, recovefjrate from a
recoveryrate of n singlematrixblockssubjectedseparately
to im m ereion.
Du preY28 and Festoey et S129 present SP M
calculations for a stack of matrix blocksshowing that this
assureptionis wrong. They show that oildrainingfrom the
bottom of one matrix block imbibes into the top of the
matrix block immediately beneath and does not enter the
verticalfractures.The resultisthatrecoveryratefrom the
and for a long periodthereafterthe totalstack recovery
may rem sin much less than n times the single block
recovery. Further,for the partialim m eraioncase,no oil
contact; oilfbw to the verticalfracturesoccurs at the
firstm atrbtblockbelow the contact.
These resultshave two damaging consequences for
m odela incorporatingthe aaaum ption. First,the true
recovery rate of a grid block is much less than that
calculated.Second,and perhaps more serious, in the case of
m ultilayergridsthereisno longerany relationship between
availableoilrecovery in a gridblock (oildeliveredto the
verticalfractures) and gridblock fracturesaturation.For
exam pie,considera reservoir150 ft.thick modelled’w~h
SPE 18427 KEITH H. COATS
three 50-ft.layerswith a matrixblockheightof 10 ft. Let
the initialfracturegastoilcontact of O be lowered to and
held at 75 ft. The top gridblock willexperienceincreaainE
matrix gas saturation but none of the corresponding
displacedoilappearsin thatgridblock. Rather,itflowsto
the verticalfractureain the second (middle)grid block.
Thus one gridblock’sdeliveryof oilto itsverticalfractures
dependaupon othergridblocks’conditions.
To examine thisreinfiltration effect,we performed
gravitydrainageSP M calculations for a single10 x 10 x 10
ft.matrix block and for a gridblock containinga stack of
sixof these matrixblocks. Data are givenin Table 6. For
clarityhere we refer to the horizontalfractureaas the xy
fractures.First,the singlematrixblock was m odelledusing
the 2 x 1 x 7 gridshown in Fig.13. The calculatedrecovery
curve is shown by the solidlineon Fig. 14. Second, the
stackof six matrixblockswas modelledusingthe 2 x 1 x 37
gridof Fig. 13. The recoverycurve isshown by the dashed
‘he in Fig.14. The inkialrateisindeedabout sixtimesleas
than the eingIe-blockrecovery rate, as reported by the
above-mentioned authors,
Allauthorsacknowledge that the cartesiannetwork of
verticaland horizontalfracturesis an idealizationof a far
more complex reservoirdescription.The xy fractures,if
they exist at all,are not preciselyhorizont~ just as
vertical fractures are not exactly verticaL A third
calculationwas thereforeperformed with a .5 ft.downward
verticaldisplacem ent of each fractureintersection(sub-)
block inthe2xlx37 grid. This correspondsto xy
fracturesslopingat an angle of about 11 degrees from the
horizontal The resultingrecovery curve shown by the
triangularpointsin Fig. 14 agrees closelywith the single-
blockrecoverycurve.
In sum m ary, with sloping xy fractures,calculated
recovery from a stack of six matrix blocksvirtually equals
six times the recovery of a single m atdx block. A
conservativeconclusioniathat SP M resultican be obtdned
to argue in favor or againstthe asaum ption,indicatinga
need for experimentaldata.. The experimente of Saidiet
a131indicateour sloped-fracture SP M resultsare erroneous.
Some sensitivityrune were performed for the sloped
xy fracturecase. Tbe base case reported here used five
(sub-)gridblocksvertically for each matrixblock with A z =
3, 3, 2, 1, 1 ft. The xy fracturegrid block permeability
correspondsto a contributionof about 50 m d to x- or y-
directionefEective fracture system permeability. The
fractureslopeis 11 degrees. A run was performed usinga 2
x L x 49 gridwith each matrixblock Az = .5,1, 1.5,3,2, 1,
1 ft.to obtaina small gridblock at the top of the matrix
block. In addition,a verticaldiaplacement of .2 ft.lowered
the dip angleto only4.6 degrees. Thisrun gave a recovery
curve 10w er than but close to the base case curve - a
maxim um of 2.7 saturationpercentage pointslessover all
time. Another run used the 2 x 1 x 37 base case gridand dip
anglebut 10w ered the xy fractureperm ea~i3ity by a factor
of 10 to a contributionof only 5 m d to fracturesystem
effectivex- or y-directionpermeability.The resultwas a
lower recovery curve with the m a~mum difference‘~from
baae case)of 4.2saturationpercentagepointsat 600 dayn.
In ~ theee sloped-fracture stack rune,the saturation
distributions
in the individualmatrixblockswere esaen~y
@dependent of positionin the stack and very similarto the
distribution
calculatedforthe singlematrixblock caae.
Ih these gas/oil.calculations,
capillaryforces tend to
prom ote the imbibition or reinfiltrationof oiL The
aaeum ption might be more validin the water/oilcase since
26s
15
oilis the non-wettingphase case and P=w o acts to prevent
reinfiltration. This can be argued againstby considering
negative valuea of imbibitionPc W. at each matrix block
bottom, but at thispoimtwe tireof futiherspeculation.
Gas OilExamplea
The firstexam pies presentedhere com pare SP M and
D P M results for constant-pressuregravity drainage.
Calculations are performed for the data of Table 6, and for
the data of Table 7 which are roughlyrepresentative of the
Resemoir A propefies. The SP M gridis2 x 2 x Nz for the
symmetrical 1/4 element,actuallyrun es a 2 x 1 x Nz with
x-directiontranam iaaibilities
doubled.
Figs. 15 and 17 compare SPM and DPM 10- and l-i%.
block resultsfor constant-pressure drainageat bubble-point
pressureusingthe data given in Table 6. The equilibrium
gas aaturationeSge at bubble-pointpressureare .498 and
.1258 for the 10 and 1 ft. blocks,respectively. These
values reflectthe gas-oildensitydifferenceof .1477 pa
used in Eqn. (28). The calculatedI,arge-tim e or stabilized
Sgm values are .4998 and .4974 for the SP M and DP M
calculations,respectively.The firstvalue isatabilize~,the
latterisstillincreasingvery S1Owly at 10000 days.
Fig.15 shows good agreement between SP M and DP M
results for the 10-ft.block using a ~g of 1.14. The
agreem ent with SP M resultaahows that the recovery curve
is closelyapproximated by the exponentialform, Eqn. (42),
fortheseparticular kr and Pc data.
Thornas et al reportedbetteragreement than we show
fortheirmethod in Fig.15. Thisisbecauae they performed
theircalculationswith pressuredepletion- about 750 psi
pressuredeclineover 1000 days. This depletionmasks the
accuracy or evaluationof the tranefer form ulationsincethe
increaseof matrixgas saturation isa combined resultof the
matrix-fracture transferand gas evolutionor liberationdue
to pressuredecline.Thisgas evolutionreducesthe tranehit
time and im provea the apparent accuracy of the tra~fer
formulation.
As previouslystated,one difficult objectivein D P M
development is constructing a transfer or exchange
formulation which at leaat approximately preserves
accuracy under time and spatialvariationof reservoix/fluid
properties. Fig. 16 indicates the senaithiky of. D P M
accuracy to variationinthe bog curve. The curve of Table
6 was reduced by a factorof 3 except forthe entry~og = 1.
The D P M with f3g= .485 only approximatelym atchea the
SP M resultsfor the 10-ft.block case. The method of
Thomas et al givesa higherrecovery curve. The D P M and
SP M recovery curve shapea show that the SP M recovery
curve does not obey the exponentialform as well wifh
krog/3 as itdidforthe originalkzop
Fig. 17 shows SP M and D P M resultefor the conatant-
pressuredrainagecaae for the 1 ft. matrix block. The
calculations usingTho m as’method agree very well with the
SP M results;the agreement isessentially the aam e aa they
reported. The reason for similarityof their and our
com pariaonein thiscase (as opposed to the Fig. 15 lo-ft
case)ia that theirdepletionwas minimal in the 1 ft.case,
havingleesthan 100 days to act due to the festertransient.
The D P M gives a recovery curve slightlyhigher than the
SP M and D P M T results.The Bg valueof 1.14determinedfor
the lo-ft.block waa uae~ the SP M/D P M curve differences
of Figs. 15 and 17 thus indicate the approximate
preservationof accuracy of the DP M formulationfor a ten-
foldvariationof blockheight.
16 I14PLICITCO MPOST1’IONAL SIMULATION OF SINGLE-POROSITY AND DUAL-POROSITY RESERVOIRS SPE 18427

The resultsjustdiscussedrelateto the case of a single from 600 to 1200 days. Gas evolution accompanying
matrix block. For the case of a grid block containinga
stack of many matrix blocks,the drainagerecovery curve
for partialim m ereion(sridblockSgf c 1)eho~d be identic~
to the single-matrix blockcurve providedSg ❑ /Sgfisplotted
in ‘lieu of Sgm. Our D P M gives thisresult. Our run using
the Tho m ae et al D P M method for Sgf = .5 gave a stabilized shut-in at 1200 days, the depletion/evolution
Sgm/Sgf of .446 for the 10-ft.block case and an S m of O
forthe l-ft.case. The .446 value reflectsthe equ!libtim
Sgm value of .223 for a single 10 ft. matrix block as
opposed to the value of .249 for the stack of blocks
representedby the gridblock. The O value of the 1 ft.case
reflectsthe influenceof the thresholdcapillarypressureon
eqflbrium Sgm of a single1 ft.matrixblockforSgf = .5.
The proceduresuggestedby Roeeen et al was used with
the SP M resultsshown in Fige. 15 and 17 to calculate evaluated at the average eaturati.onvalue of this
peeudo matrixgas/oilcapillary pressurecurves forthe 10 ft.
and 1 ft block cases,respectively.Fig. 1S show.sthat the
curves differgreatlyfor the two differentblock heights. distribution.
The implicationof thisis that in caees of variableblock
heightover the reservoixgrid,each gridblock would need to
have input and storage of a differentp8eudo curve. The
slopeof the pseudo curve issatisfactory (positive) for the 1
ft. case but ia negative for the 10 ft case at saturations imbibitioncurve along with the S: drainagecurve. Results
below .3. A negative slope can cause stabilityor
convergence problems.
Tn this and other unreported cases,the unorthodox
integrated curve shapes obtained lead us to question
whether we are in error either in understanding or
implementing thisproposed procedure. For thisreason we
includeTable 6 Vhich givesthe SP M Sgm vs. time curve for
the 10 ft.block. These are the only data missingfor the
purposeof performingthe integration.
Figs.19 and 20 compare SP M and D P M resultsfor the
gas/oildrainageprocessforthe data of Table 7. A 1x1x4 ft.
matrix block containing saturated oil is im m creed in
equilibria m fracturegas at constantbubble-pointpressure.
Fig. 19 shows that the SP M and D P M resultsagree fairly
W~ with 13g= 1.3. As previouslystated,the D P M should
givean invariantplotof S&m/Sgf for the partialimmersion
case. The cixculerpointsm F%. 19 are a plotof Sgm/Sgf
from the DP M forthe case of a constantSgf = ,5.
The approximate preservationof D P M accuracy for a
fixed~g and differentblock heightsisindicatedin Fig.20.
SPM and DPM resultsare shown for Sgf = 1 and ~g = 1.3for
a matrixblockheightof 1 ft.
Simple pressure depletion may be a dominant
m echaniem during periods of production with no gas or
water injection.Two cases are consideredfor a singlegrid
block subjectedto pressuredepletion.Data in Table 7 are
used. The gridblock of.height24 ft.isa stackof six1 x 1 x
4 ft. matrix blocks. In the firstcase, gaa evolved in and
escaping from the matrix percolatesupward through the
verticalfracturesleaving the grid block’sfracturesoil-
filled.Jn the second case,caprock or gas-filled fracturesin
an overlyingblockpreventsthispercolation and the evolving
m atb gae enters and accumulates in the fractures, largerSge correspondsto more gas in the m atxixand lessin
res~ting in an increaaimgSgf or declininggas-oilfracture
contact, The SPM and DPM gridsare 2 x 1 x 24 and 1 x 1 x
2 respectively.A productionwallproduceea constanttotal
RB/day rate from the fracturesfrom O to 1200 days and is
shutin from 1200 to 2500 days.
Fig. 21 shows the SP M matrix gas saturation,Sgm ?
plottedva. time. Sgm is the volumetric average of gas
saturationsin the 24 matrix sub-blocksof the 2 x 1 x 24
SP H @l. Sgm increasesfrom O to 600 days and decreases
--.
m
pressure decline tends to increase Sgm w~e caP~
forces along with weaker gravityforcen tend to decrease
Sgm by forcingflow of matrix gas to the fracturee.Theee
opposingtendenciesbecome equal at about 600 daye with
the capillary-gravity forces dominating thereafter. After
effect is
absent and the capillary-gratiy forcesact alone to reduce
Sgm. The rate of Sgm declinediminishesas Sgm decreaaeep
due to decreasing~gm and Pcgom.
The D P M resultsshown by the open circularpointsin
Fig. 21 match the SP M results welL The growing
discrepancyat large time reflects the verticaldistribution ?
or variationof gas saturationin each matrix block in the t
SP M calculation.The DPM uses ~gm and Pcgom values .-
distribution. These values are lesethan the SP M effective
values which reflectvalues integratedover the vertical
For the resultsjustdiscussed, a singlematrixcapillary
pressurecurve of S; was used. The triangular-point DPM
resultsin Fig. 21 nulicatethe effect of use of a .1*S2
are identicalto 600 days Ace Sgm is inCreSSkig and o~y
the drainage curve is applicable. After 600 days,,Sgm
declinesand the D P M calculationswtiches to imbibitmn
curve values. The lower im bitdfion curve resultsin a 10war
rate of gas expulsionfrom the matrixor @“eatermatrix gas
retention.
The SP M and DP M calculations give the same matrix
pressureand interracial tensionvs. time cuwee~ shown in
the lower portionof Fig. 21. The plottedIFT is ratioof
tensionto initial tensionat the initiAI 5553.7 peia bubble
point pressure. The value of nearly6 at 1200 days m eana
thatgas-oilcapillary pressuresincreasesix-fold from bubble
poihtto 4200 peia.
For the 8econd depletion case, gas entering the
fracturescannot percolate out of the grid block. Oil
productionrate at the bottom of the block is constantfor
1500 days and O thereafter.In thiscase gas simultaneously
flows from fractureto matrix and from mat& to fracture
at differentpositions in the gridblock. In the upper partof
the block,gas flows from fractureto m atti. In the 10wer
part,fracturesare oil-filled and gas flows from matrix to
fracture. After shut-inat 1500 days,a circulatory type of
flow occurs; gas flows from matrix to the fracturesin the
lower region,percolatesupward and enters matrix blocks
above the fracturegao-oilcontact.
The D P M cannot represent simultaneousfbw of a
phase in both directions; it can only approximate that
situationby a net flow in one direction.Fig.22 compares
SP M and D P M resultsfor thissecond depletioncase, with
and without~ T effects.U the latterare ignored,lessgas
appeam in the fractures which, in terms of reservoir
performance, translatesinto 10wer G OR. The reason for
thisis that Sge islargerwhen IFT effectsare @nored and
the fractures.Fig. 22 ahowa only moderate agreement of
SP M and D P M resultsforthissecond depletioncase.
Fig.23 shows pressureand matrix gas saturationvs.
time for the gP M and D P M calculationwith and without
IFT effects. The curves are the same for allfour cases.
The interracial tensionratiorisesfrom an initial value of
1.0to 4.5at 4430 pt& at 1500 days.
 Fig.24 comperee SP M and D P M resultsforthe case of existedat earlytime with a downward verticalmovement
fracture oilimbibitioninto a gas-filledmatrix block at over time of a nearly piatodike gas-oilinterface. The
constant,bubble-pointpressure.This processmay occur in inifialS e value was .5664at the initial
IFT (rat5.o
of te.tin
the reservoirif oilin pushed upward into a gas-cap grid t. d te~~ of 1.0. Rough averages of IFT, Sge and
block or a previouslydrainedg-idblock. The 1 x 1 x 4 ft. Sgm valuee over blocks where Sgf = 1.0 at 7 years are es
matrix block initial
saturationsare Sgm = .9,Sw m “ Swc “ follous:
.1,and the fractureoilsaturation isconstantat 1.0. Table IFT
7 g$ee other data. An imbibitioncapillarypressurecurve C see Inj.Gas Diffusion Fracture Matrix Sge ‘gm
of Sg was used.
1 Separator No 7 1.4 .53 .54
2 Separator Yes 2 2 .50 .53
Fig.24 shows closeagreement between SP M cnd D P M ~
Nitrogen No 60 5 .37 .37
single-matrix-blockreeultaforthe case Sof = 1.0. However; 4
Nitrogen Yes 40 40 .23 .50
the D P M formulationdoes not exhibitthe properinvariance
to partialfracturesaturation.The correct(dashedline)and
The average Sgm of .53for Case 2 includeseeveralblocksat
actual(triangtirpoints)D P M resuh for the case of Sof = and near the gasinjector where Sgm was .9due tO matrixOfi
.5are shown, The correctD P M Sgm valueiS 1/2 the sum of
vaporization With diffusion, matrix/fracture gee
●9 and ‘he ‘~M ‘gm value calculatedfor Sof = 1.0. The com positionswere nearlyidenticalas were matrix/fracture
D P M formule~on does givethe correctasymptoticor large-
time valueof Sgm of .45. oilcom poeMons where fractureoilexisted.In Case 4, the
large differencebetween Sgm (.5)and Sge (.23)reflectsa
combination of early-tire e drainage of oil when S8e WJM
The above discussionand Figs. 15-24 examined the
largerand vaporizationof matrixoilby the high-N2 content
accuracy of the D P H formulationin connectionwith single
matrix gas. No complete matrix oilvaporizationoccurred
m at~ block or singlegridblock behavior. The sensitivity
in Case 4. No grid block matrix oilsaturationat eeven
of D P M full-gridaimulationresultsto inaccuraciesin the
yearewas lessthan .3.
formul.etion is of equal interest. Fig. 25 shows the D P M
single matrix block, constant-pressuredrainage curve
Water/OilExa mplee
calculatedwith (a)f3g= 1.3(asin Fig.19),(b)13g= l.% and
(c) Bg = 1.0 and the constant-~option deactivated. The
Single-Block Imbibition
cases(b)and (c)give moderate and largeerrorein the D P M
resultsforthe rnngle-m atrix-blockcase.
SP M and D P M resultsare compared firstfor water
imbibitioninto a singlematrix block. The accuracy of the
The D P M param eteraof the three cases (a)-(c)
shown
D P M resultsisexamined for changesin two variables-block
fi rig. 25 were used in 12x5 x-z cross-aectiona~ constant- heightand the irnbibition capill.aty
pressurecurve. The two
pressuregas injection,D P M aimulationrum. The crose-
blocksizesare 10 x 10 x 10 ft.and 10x 10x 30 ft. The fkvt
sectiongridand welldata are givenin Table 7. The injected
Pc isthat givenin Table 6. The second,denoted %w PC”,
separator gas com positiongivee a vaporizinggae drive.
given in Ref. 30, is identicalexcept for the fol.lo wing
Average reservoix pressure during the seven yeara of
entries:
simulationis about 5600 psia for all runs diacusaedhere.
Gas injectionra!e is constant and the productionwell is Pcwo
placed on deliverability at a bottomhole pressure,of 5500 ‘urn Table 6 Low Pc
psia. 7 50. r
.25 9. .5
Fig. 26 shows that the moderate and large .3 2. .3
inaccuraciesin the D P M single-block
behaviortranslateto .35 .5 .15
am alland moderate inaccuraciesin the 12x5 cross-sectional .4 o* o.
results. ThiE tendency of a change in a given model .45 -.4 -.2
(reservoiror fluidor M echaniam) parameter to cause a
significantly
smallerchange in overallsimulationresultsiE Obtain@g the correctSP M resultsfor these cases is
well known. Nevertheless,(as,indeed,~here perhaps)we not entirelystreightforward. A l/4-elementNx x Ny x Nz
frequentlybelabor the accuracy of a model parameter or SP M gridincludeehorizontalfractureblocksat k = 1 and N z
mechanism without examining its impact on the overall and verticalfractureblocks at i = N x and j = N ~ The
aimulationreaulta. choice of krwf value in these wate~filledfractureblocks
for upstream use in water imbibitioncan affect the SP M
For several years, da Silva12 has gathered and resultsby a fractionto severalsaturationpercentagepoints.
presented evidence that diffusion may play a very In theory,the facial~wf value is determinablefrom the
significant role in dual-porosity reeervoir behavior, boundary conditionSWf -= 1. Firstthoughtleads to a value
especiallywhen injectiongas com poeidon differsgreatly of krwf = kflwm = the matrix ~w at Sum = S$m = water
from that of the naturalreservoirgas. Fig.27 ahous the saturationat imbibitionPcw o = O. Thisvalue is .03 for the
effectof diffusionin the D P M cross-sectional
aimuletionfor data here for both Pc curves. However, the block
the case of separatorgas injection.With diffusion included experiencesgravitydrainagein additionto imbibitionwith
in the calculations, the results show somewhat higher finalequilibriumwater saturationequallingS% m at the top
recovery with significantly 10w er G OR. Fig. 28 showa a and the Sw of pointB in Fig. 6 at the bottom of the block.
greater effectof diffusionwhen nitrogenis injected. Oil The maxim urn possiblevalue of the lattersaturationforany
recoveryiaroughly60% greaterwith diffusion whileG O R is block height is I-Sorw and the maxim urn possible
significantlylower. A crossplotof GOR ve.recovery would corresponding~wf valueiskrwro, which iS .23 for the data
illustrate
the effectof diffusionmore dramatically;at 16% here. These values of .23 and .03 are over seven-fold
recovery,calculatedG O R is 3800 Scf/STB with diffusion differentand at varioustimes and block positions, the true
and 25,500Scf/STB withoutdiffusion. fac~ krwf upstream value for water flow tito the m atdx
block may apan thisrange, SP M resultsobtainedusingfixed
These cross-sectionrune with the two different
%tif = .23and ●O3 v~uea are sis~an~y different
injectiongasea and with and withoutdiffusion
exhibitedthe
followinggeneralcharacter. A strongfracturegas override
---
I
18 IMPIJCIT COMPOSITIONAL SIMULATION OF SINGLE-POROSITY
The proble❑ isreso’vedin the followingmanner. For
a givenblockheightand P= curve,the value of Sw at point
B of Fig.6 is determined. The value of ~w ~ at thisSw is
calculated.The proper~wf value cannot exceed this~w
and it b used for facial upstream krwf in the SP M
calculation.The resultingkrwf values for the four cases
are:
MatrixBlock
Sise Pcwo krwf
10 X1OX1O Table 6 .069
10 X1OX3O Table 6 .108
Lox loxlo Low .06S
10 XIOX3O Low .104
The value of krwf onlyaffectscalculatedflow intothe first
matrix grid block at the m atrixffrac ture interface.
Thereforeuse of a sufficiently finegridwillgivevery small
dimensionsof thisfirstblock and the SP M resultsbecome
insensitive to the value of krwf. We used a 5 x 5 x 12 grid
for the l/4-element(x = O to 5 ft.,y = O to 5 ft.)with AX
and A y matrixspacingsfrom centerto fractureface of 2.5,
1.5,.7,.3ft.
Sensitivityruns with other grids and ~wf values
indicatedthe above procedure givesaccurate SP M results.
The truncationerrorof Nz = 12 vs Nz = 22 issmalleven for
the 30 ft.blockheightand isin the oppositedirectionof the
s m d error of using the maximum possiblekrwfi The
verticalspacingfor Nz = 12 is .001, 10 * (h/lO),.001 ft.
The SP M calculationisactuallyperformed as a l/8-element
which reduces computing time. The aimpleatgridisa 2 x 2
x Nz with the matrix grid point at x = y = 2.5 ft.and is
equivalentareallyto thatused in the D P M formulati.on
-also
equivalentto the shape factorEqn.(20).The 2 x 2 x Nz grid
is actuallyrun as a 2 x 1 x N ~ with x-directiontransmis-
sibilitydoubled. The resultingSw m vs. time curve ranges
from a fractionto about 3 saturationpointslower than the 5
x5x Nz grid.
Fig. 29 shows SP M and D P M resultsfor the 10 ft.
block and Table 6 pcwo. Thornas et al presentedSP M and
D P M resultsforthiscase. Our SP M resultsand resultsusing
their method show about the same agreement se they
reported. The two D P M methods give virtuallythe same
transientforthiscase.
Fig.30 shows that forthe 30 ft.blockour D P M results
are slightlyhigh and our curve using Tho m as’ method is
higheryet.
Fig.31 shows forthe 10 ft.blocksnd low Pc case good
SP M agreement using our D P M and a low transientusing
Thomas’ method.
For the 30 ft.block and low Pc case, Fig. 32 shows
good accuracy of our DP M and a significantly
more rapid
transientusingTho m as’method.
For both 10 ft.and 30 ft.block heights,the low Pc
case requixeda f3wvalueof 2.7 whilethe higherTable 6 P
case requiredno change from the default1.0 value. Ti-&
reflectsour experiencethat imbibition-dominated(highPc)
casesgenerallyrequireno adjustmentof f3
w.
Figs.31 and 32 show fairpreservationof accuracy of
D P M resultsusing the same 2.7 8 w value with a 3-fold
change in blockheight.
If matrix block heightis significantly
less than grid
blockheight,then a plotof (Sw m -Swc)/((Swf* (Swe - Swc))
for the partialim m ereioncalculation
shouldgivea transient
AND DUAL-P OK OSITY RESERVOIRS SPE 18427
curve independent of Swf. Fig. 33 shows SP M and D P M
resultsforSwf = .5 and 1.0forthe 10 ft.block and Table 6
P c“ The curvesare nearlyindependentof Sw f value.
Three-DimensionalWaterfloodExample
Resultsare com pared to those of Thomas et al for
theirlinedrive water injectionexam pie. Their 10 x 3 x 5
griddescribesa 2000 x 2000 x 250 ft.resemoir with water
injectionand totalliquidproductionratesspecifiedat x = O
and x = 2000 ft.,respectively.Pc and ~ data and fluid
propertiesat 6215 paie formation pressure are given in
Table 6. Matrixblocksare 10 ft.cubes. The remainingdata
are not reproducedhere.
Each layer of thisgridis 50 ft.thick and containsa
stack of five matrix blocks. Fig. 29 shows that both
Tho mast and our D P M methods give good agreement with
fullim m ereic
n SP M resultsfor water imbibitionintoa single
10 ft. m atr”xblock. However, Fig.34 ahows the different
gridblockequilibrium curvesforthe two methods. Equating
the VE (pseudo) matrix and fracture capillarypressure
curves for various Sw f values gives the two solidcurv..
shown on Fig. 34. The upper curve is obtainedby applying
this procedure for a single 10 ft. matrix block and
representsTho mas’method. The step-function curve isthe
correct resultobtained by applyingthe procedure on the
basisof the 50-ft.gridblockheight. The dashed lineisthe
relationshipused in our method ifS~f = O.
The expected impact of the differencesin these
equilibriam curves on 3D resultsis difficultto deduce.
Confusionarisesin trying‘:0reconcilethe curve differences
with factorssuch as (a) vhe magnitude of the imbibition
transienttime (Fig.29, about 1 year)com pared with flood
displacementtimes, (b)the time a grid block spends in a
partial immersion etate,com psred with imbibitiontransient
time, and others, If a problem were designed whera a
dominant aspect of resemoir behavior was a long-term or
stabilized establish
m ent of partialfracturesaturation(say,
.5),then reasoningalone might deduce from Fig. 34 that
calculated recovery would be significantly higher for
Tho m as’method than oura. Tn any event,it seems safe to
inferfrom” Fig. 34 that Tho m as’resultsmight reflectfaster
earlyimbibitionleadingto 10w ar W O R and higherrecovery.
However, all blocks eventuallyexperiencepermanent SWf
values of 1.0 at which recoveriesfor both methods are
identicaL
Fig. 35 com pares the two methods on the basis of
water cut calculatedfrom the 3D simulations.The methods
give very similarresults.Our water cut iahigherearlyand
10w er at latetimes. Thomas reportedbreakthroughat about
1.5years,and water cut and recoveryvaluesat 10 yeara of
92% and 35 % OOIP, respectively.Our breakthroughtime is
about 1 year and ill-year water cut and recovery are 88.4%
and 34.6% O 07.P,respectively.
Our run was made using an initial5-day step with
autom atic step selection limited only by a specified
maximum step sise and a maximum increaseof 50% from
step to step. For a maximum step of .5 years,the run
required27 steps and 63 iterationsfor the 10 years. A
maximum step of .25 yeara resultedin 45 steps and 91
iterations.Tim e truncationerroris small as the fasterrun
gave lo-year water cut and recovery values of 87.4 Z and
34.4 % OOIP. Resultswere insensitiveto whether the VE Pc
optionwas used forthe interlock fractureflow.
SPE 18427 KEITH H. COATS
Five-SpotWaterflood
Kaze mi et al presenteddata and resultsfor an SX8X1
areal grid simul.ationof a waterflood in a five-epot
quadrant. Theirresultewere closelyreproducedby Thomas
et ei. The data are in Table 4. Matrixblock dimensioneare
10x1OX5O ft.and reservoirthicknessis 30 ft. Therefore
matrix block height equals grid block thicknessand there the y-direction permeabilityassignedto the matrixsegment
are no horizontalfractures.
Several purposes are served in connection with this
exam pie. First,the D P M isevaluatedin the case where its
accuracy is leaat- the case of n = 1 or equal matrix block
and grid block heights. Second, an equivalentline-drive for the data of this example problem. The fracture
waterfloodcase is defined. The interwell distance, total permeability is kxl = kzl E 10000 mm
volume and ratesare the aam e as in the five-spotquadrant. permeabilityof 1 md giveskz2 = 1 md, and kx2 = O. ThiE 3D
The correctlinedriveresultscan be obtainedfrom a three-
dimensionalSP M grid. Therefore,the D P M resultsare
com pared to correctreaulta,not just to those of another
m odeL
The com pariaonbetween line-driveD P M and correct
(SPM) resultsisextended to the case of formationthickness
= 90 ft. where n = Az/h = 3, a case where D P M accuracy
would be expectedto be greater.Finally, the areal8 x 8 x 1
D P M five-spot
resultsare presentedwith an estimateof the
correct five-spotresultsprovided by the DP M-SP H line
drivecom parisone.
Our calculations utilizedata differing
from Kazemi et
al in the fcllowingrespects. Our injecdon and production
rates are 200 R B/d. With theirdata in parentheses,shape
factoris .16 (.0S)and fracturekr and Pc are linear(non-
linear)and zero (non-zero), respectively.‘theextended time
period of 3200 (1200) days allows more meaningful
comparisons. Fro m che data, we calculate effective
fractureperm eabilitiea of 50 and 100 md in the x and z
directions, respectively. Thomas et al used an effective
value of 500 md. The effecton W O R and oilrecovery of
effectivefracture permeabilityvalues over thisrange is
essentially absent and isotropicvalues of 100 and 500 m d
are used here forlinedriveand five-spotrune,respectively.
b the 3D SP M gridsdescribedbelow forthe linedrive,wells
were completed in alllayers and a PI of about .06 R B-
cplday-paiper ft. of com pl’etion was used. R esulteare
unaffectedby PI valuesrangingfrom thislevelto 100 times
larger.
The Ehre-epotquadrant is 600 ft.on each aide. The
equivalentline drive element having the same interwell
diztanceand totalvolume is 848.5 ft.long,424.24 ft.wide
and 30 ft. thick.To account fordualporosity, an 8 x 2 x 10
3D single-porosity gridia used with A z = 3 ft. The two
adjoining8 x 10 cros~ections, denoted by subscriptsj=l
and j=2,representfracturesand matrix,respectively.For
the givenfractureand matrixporosities of .01and .19,
AY1 = 4.2425 ft. @l=1.
(68)
AY2 = 420 ft. ~ = .19(.99)=.1881
The proper value of k for y-direction(matrix-
fracture)flow iscalculatedas !?
ollows.A skgle 10 x 10 x 30
ft.matrixblockhas a totallateraltram miaaibility
acroesits
fourverticalfacesof
‘K= ‘h ~=18kh (69)
%% + 5X14)
x 600
In a gridblockof dimensionsA x A y A z thereare
AX Ay Az(l- of)/ 2X Ey h (70)
‘-.257
19
of these matrix blocks. Tn the 8 x 2 x 10 gridthe @irection
tranemiaeibilitybetween two gridblocksistherefore
Ax Ay AZ(1- ~)
(71)
5 = ‘Y2 W2 = 16kh %$yh
where Ay is the total424.24 ft.?AYZ ie 420 ft.?and kYz iS
j=2. The value of kyl’for the fracturesegment isinfinity.
Eqn. (71)gives
8 Ay2 AY
k = 14254.5md (72)
~y2 = %x .4Y
the matrix
SP M gridrepresentsa fracturedformationwith a normal set
of both x-z and p verticalfractures.The D P M equivalent
gridis 8 x 1 x 1. Both gridsrepresenta line-drive or one-
dimeneionalflood.
The 8 x 2 x 10 grid representsthe matrix-fracture
flow usinga singlegridpointor gridblockforthe 10x 10 ft.
xy area of the matrix block. Thizrepresentation isexectLy
equivalentto a 2 x 2 x 10 3D grid representationof a
(symmetrical)quadrant of a singlematrix block with kc
associatedverticalfractures. Thus we can aeeees the
accuracy of the 8 x 2 x 10 field-scalesimulationresultsby
performingim meraionrunefor 1/4 element of a singleblock
usinga 2 x 1 x 10 xz grid with x-dixection
Cranemieaibililiee
multipliedby two.
Fig. 36 com pares the single-block, water imbibition
(immersion)curve calculatedusinga (1/4element)finegrid
of 5 x 5 x 10 and the 2 x 1 x 10 grid.The lattergridgivesa
transientcurve somewhat low. However, increasingthe
matrix permeabilityby 20% gives a 2 x 1 x 10 @ result
virtually identicalto the correctcurve,as shown in Fig.36.
Therefore,ky2 = 1.2x 14254.5= 17105.4md was usedin the
8 x 2 x 10 field-scale SPM @d. The 5 x 5 x 10 finega
used A x and A y spacingsfrom fractureto blockcenterof C,
.3,.7,1.5,2.5 ft.and verticalspacingof A z = 3 ft. Both
gridsused krwf = .2at Swf = 1 forflowintothe matrix.
Alao shown on Fig.36 are the single-blockim m eraion
reeultausing our D P M and Tho m as’ method. The latter
m ethod’ause of VE capillary
pressureses drivingforce gives
an excessivelyrapid traneientfor the 30-ft.block height.
Our DPM read% are somewhat low but compare wall with
the correctSP M results.The DP M runeuse the 1 md matrix
perm eabilityvalue.
The implicationhere is that SP M-D P M agreement for
the sing%e matrix block full im m eraion case ensuree
accuracy of the field-scaleSP M 8 x 2 x 10 simulation.One
might object that the full im mereion com pariwn is
inconclusivesince all grid (matrix)blocke in the field
aimulationexperiencevanoua statesof partialim m ereion.
We thereforerepeatedthe singlematrixblock 5 x 5 x 10 and
2 x 1 x 10 (with km x 1.2)rune for the partialimmersion
case SWf = .2. The followingreaulteindicatethat the full
im m ereion accuracy is retained under partialim m ereion
conditione.
Tim e s“m (swf =.2)
l?S#— 5X5X1O 2X1X1O
.2997 ~
.3610 .3607
1200 .3944 .3953
2400 .4326 .4361
3200 .4493 .&539
20 IMPLICIT COMPOSITIONAL SIMULATION OF 81NGLE-POROS~y
Fig.37 #hows the gridblock equilibriumcurvesforthe
caseshere of n = 1 and 3. The upper curve iscorrectfor the
caae n = 1 and isobtainedas discussedpreviously.Tho m as’
method reflectsthis(upper)curve in the matrix-fracture
drivingforce expressionfor any value of n. The correct
curve forthe case n = 3 is shown by the etep-functionsolid
curve of Fig. 37. The D P M curves(Eqn.(34))are ahown by
the dashed lineefor S~f = O, .0856and .314. W e used S~f =
.314 and .0S56 for the casesn = 1 and n = 3, respectively.
Thisreduces the D P M error for small n. The value of .463
for equilibriumgrid block saturationat SWf = O is the
equilibnu m saturationof a 30-ft. matrixblock exposed to
water at itsbottom face.
Fig.38 com paresSP M end D P M oflrecoveryand water
cut resultsfor the field-scaleline-drivewaterflood with
formation thickness= h = 30 ft. The DP M recovery ia
moderatelyhighup to 1500 days and too low at largertimes.
The SP M and DP H recoveriesare 37.9 and 39.1% OOIP,
respectively,
at 1200 days, and are 56.? and 53.2Z O OLP,
respectively,
at 3200 days. The ultimate recovery for all
three calculationsis 62.6Z OOIP when the fracturesare
100Z w ater-Eilled.
The high DP MT recovery curve aimply reflectsthe
high im m eraion transient of Fig.36. If the matrix,
permeabilityisreduced 65X in thatrun,agreement with the
SP M resultsis much better.
For the case of reservoirthickness= 90 ft.,the well
ratez are increasedto 600 R B/d so that rate per ft. of
thicknessis unchanged. The SP M gridis S x 2 x 30 with 30
3-ft.layersand zero z-direction tranamissibilit.ies
between
layers10 and 11 and between layers20 and 21 in the matrix
cross-aectinj = 2.
Fig. 39 compares SP M and DP M recovery and water
cut resultsfor the case where resenoir thickness= 90 ft.
and n M Az/h = 3 for the 8 x 1 x 1 DPM calculation.The
D P M resultsare more accurate in this case than in the
previouscase of n = 1. The D P M T resultsfor thiscase are
the same es the DPMT resultsfor n = 1 ehown on Fig. 38.
The SP M and DP M recoveriesare 36.1 end 36.4Z OOIP,
respectively, at 1200 days, and are 53 and 51.4% OOIP}
respectively,at 3200 days. Recovery for this 90 ft.
thicknesscaee isroughly2 to 3 recovery percentagepoiute
lower than forthe 30 ft.thicknesscaee.
Fig. 40 shows five-spotoil recovery and water cut
D P H resultscalculatedusingthe 8 x 8 x 1 areelgridand the
30-ft.reservoirthickness.We believethe correctfive-spot
recovery curve would compare to the 3PM curve on Fig.40
in about the same fashionas exhibitedby the SP M-DP M lime
drivecomparisonof Fig.38.
The five-spotD P M mu requixed28 timeatepsand 34
iterationsto 1200 days and 44 eteps and 50 iteratkme to
3200 daya. Materielbalances were .99998. Tim e steps
ranged from the initial2 days to 180 days. Tim e step
controlswere maximum saturationchange per step of .15
and maximum step size of 180 daya but both were rarely
invoked. Rather,output times of every 200 daya to 1200
days and every 400 daye thereafter together with a
maximum increaseof 50% from step to step determinedthe
time steps. Rune with varioussmeller maximum step has
indicate that time truncationerroris minimeL
Applicationto VolatileOilReservoixA
Reeemoir A isa fracturedmatrixreeervotiwith about
16 yeare’history. The reeervoiroilie an undereaturated
m.-
Zou
AND DUAL-P OROS~Y RESERVOIRS SPE 18427
volatileoilwith B. of about 2.2 RB/STB and solutiongas of
1966 SCF/STB at the 5553 paia bubble point. original
reservoirpressure is about 7000 peia with a water-oil
contact about 1200 feet balmw top of structure. The 3-
component EOS representationof the oil wae discussed
above in connectionwith Figs. 1-2 and Table 2.
During the lest13 of the 16 years’history, first-stage
separator gas (86Z methane) w es injacted. When gas
injectionbegan, rese~oir pressure wee 6S00 peie,about
1300 PM above bubble point. Reservoir pressureand oil
recovery at the end of historyare about 4000 paieand 12%
of OOIP, respectively. GOR hms risen to about 9000
Scf/STB. Queetiona concerning future recovery include
continuedseparatorgas injection, N 2 injectin andlor water
injectiow
The reservoirstudyhas proceeded far past the status
described here. The originalhistory match dateaet is
retained as a fixed reference for the purpose of testing
model changes or enhancem ente. Table 9 gives some
res.ervofr descriptiondata. Since the reservoiria highly
heterogeneoua~ only average values are given for
permeability, porosity, etc. The effective fracture
permeabilityisthe order of 100 times matrix permeability
with the latteraveraging about .5 m d with variationsto
much 10w er values.
Of interestare the sm all mattix block size (prolific
fracturing)on the order of 1x1x3 ft.?and the 10w fracture
porosityaveraging.0005or lees. A consequence of the first
isa large matrix-fractureexchange coefficient the order of
107 (RB-cp/da~ei). The 10w fracturepororntygivesam all
fracture-ayetempore volumes and high throughputratiosfor
large time steps. This requireeimplicit model aimulation.
The large transfer coefficientsresult in com putatkmel
difficultysincethe matrix-fracture darcy flow ratesare the
products of large coefficients and small potential
differences(.01to .001 and lower psi). This combination
can cause round-off and com putationel convergence
difficulties,especiallyfor large time eteps and large,
rapidlychangingiajection/producticm rates.
Fige.41 and 42 show totalfieldoilproductionrate and
gas injectin rate ve time as 9l-day averages. The rates
change every 91 days throughout the dataset,giving a
maximum 91-day step size. Fig. 42 ahowa the extreme
variationsin gas injectionrate from one 91-day periodto
the next+ About 6 of the 47 wellsare gee injectionw ew
only one well injectswater, during the lest 270 days of
history.The gas injection rate changee might be expected
to resultin oscillatingfractureoilsaturationsfrom step to
step,especiallyat or near injectionand/orproductionwells.
Fig.43 ahowa calculatedfractureoilsaturationvs time in a
gridblock adjacentto a high-rateinjector. Calculatedofi
saturationisvirtuallymonotonic with time in spiteof ‘large
injectionrate changes.
The S e value is important in this study. Its
internally-c3 culeted value changea wish time and from
block to block. Historyruns were made for the three came
of constantSge (with spatialvarietion)~ de~fiependent
S e, and dermt~ and tension-dependentS e. Rune were
a%o made with and without diffusion. T%e o~d S e
velueaare about .4-.5forblockheightsin the range of 3 f0
4 ft. The model historyrune show calculatedgrid block
matrix saturationsb close agreement with the Sgf x Sge
product~ reflecting the Lsrge matrix-fracturetram mw
aibilitiee.Two exceptionsoccur. First,forrock type 2,Swc
is large, oil saturationis low and oil easily becomes
immobile due to trappingby gas and watar in accordance
with the Stone 2 km method. Thus a number of the type 2
SPE 18427 KEITH H. COATS
blocks show matrix gas saturation< Sgf x S ~ simply
because the matrix oilis immobile and cannot 8rain out.
The second exceptionis that gridblock equilibriumoccurs
at some pressure,followedby significant pressuredecline.
Oildoes not want to (and does not in the calculations)
drain
out of the block to matrix oilsaturationsbelow equilibria
m cutsoccurredand totalcomputing time was about 2.5VAX
valuesbut the oilsnrinkein accordance with the constant-
volume expansion. Also, depending upon the nature of
injected gas, oil can vaporize, further reducing its
saturationor,equivalently, increasingmatrix gas saturation
above itsequil.ibnum value.
Fig.44 compares fieldtotalobserved and model G O R
vs time. The model resultsreflectdiffusion and density-and
tension-dependentS e. Calculated G O R is avout 9400
Scf/STB at the end oF history.Withoutdiffusion, calculated
G O R is about 14Z higher. For the time-constantSge case,
calculatedG O R is 7700 Scf/STB at the end of .Gtory. S
and diffusion are the controllingparameters in tl%
calculated G O R behavoir. Runs with depth-dependent
initialcom position(bubble point decreasing with depth)
show significantlylower G OR.
Fig. 45 indicatesthe strong diffusioneffect when
injected and native reservoir gas compositions cliff er
significantly.Restart runs were made from the titme-
constant Sge case,injectingN 2 rather than separatorgas
during the 3472-5753 (end of history)day period,with and
withoutdiffusion.Fig. 45 shows that calculatedG O R is
extremely higher without diffusionthan with it. Table 10
shows the effectof diffusionon matrix-fracture oiland gas
phase compositionfor thisN 2 injectioncase in a gridblock
adjacentto a gas injector. Without diffusion, virtuallyno
N 2 appea~ in the matrix phases and large composition
differences exist between matrix and fracture phase
com positions. W ifh diffusion(only gas-gas diffusionwas
used),matrixand fracturephase compositionsare identic~
both in the gas and oil.
Table 11 shows calculatedmatrix and fracturegas
saturationsat the end of historyfor the constantSge run
with diffusion. Saturationsare shown for several cells,
includingactiveand shut-ingas injectionblocks. As stated
above,eqwbti m matrix gas satxationieSgf x Sge. Table
11 shows that the calculated matrix gas saturationsare
close to equilibria m, due to the large n atrix-fracture
exchange coefficients. Table 11 shows a matrix gee
saturationexceeding Sge for the activegaa injectionblock
4-8-1. This reflectsthe vaporizingeffectof the separator
gas, reducing matrix oil saturation to lees than its
equilibria m value.
In all model rune no computed matrix oilsaturations
leasthqn .2 were computed. That is,complete vaporization
of oildid not occurin any case/block.In fact,the minimum
matrixoilsaturations over the gridwere closeto .2 both for
N 2 and separator gas injectioncases. These minimums
occured in gas injectioncellsand matrix oil eaturatior.a
were significantlyhigherthroughoutthe non-injection cells,
obsyingthe equilibrium raldionship.
Maxim um gas productionrates(QG M A X) were entered
in the data fileforallproducingwells.They were obtained
from the historyfileas the maximum gas M CF/D rates
produced throughout history. The model produces the
specifiedST Bid oilrate for each v allunlessthe associated
gas rate exceeds the w en’sentered Q G M AX. In the latter
event,the wallproducesthe Q G M A X gaa rate, The overall
impact of thisisnotedby printingtotalcum dative ehortage
of oilproduced (STB) dividedby totalspecifiedCUm ulative
On production.This fractionis .0145 for the run shown in
Fig. 44.
259
21
The 5753-day historyrun required60 time stepsand
167 total outer iterations,an average of about 3 outer
iterationsper 9l-day tine step. 91-day steps were used
throughoutwithlarger,up “to20*day, stepsused duringthe
first1280 days before gas injectionbegan. No time-step
780 CPU hours.
An iterativeY Z planar SO R was used with reeulting
omega valuesof about 1.4and totalinneriterations at 5753
days of about 1700,or an average of 10 SO R iterations
per
outeri.tmation.The linearsolvertakea about 20% of total
computing time. Allcomponent materialbalances’ were the
order of .999X .to .9999X throughout the runa with the
“sensitive”norm aldenominatorbasisof cum dative injection
or production.
Computer time with 4 components (3+N2) active is
about 60% greaterthan that for 3 components. All rune
were made with a 4-component set of PVT data entered
with N2 the lastcompol.ant.SinceinitialN2 mol fractionis
O, the model autornatically com putea with only 3
com ponenteuntilthe firsttime step(ifany)of N 2 injection.
It then automatically expands to the 4-tom ponent
calculation.This can save considerablecomputing expense
in problems where certaincom ponente(s)are absentinitially
and throughoutpartor allof historybut become activedue
to laterinjection.
Sum mary
An ,implicitcom positionalmodel has been described
for con positionalemulation of single-or dual-porosity
reservoira. The model simdates uneteadptate three-
dimenaional,three phase flow in heterogeneousreservoirs
rangingin type from black oilto near-crkicaloilor gas to
lean gas condensate. Applications includedepletionand gas
andfo~ w? .:er injection.Single-porosityapplications
incurno
10SS ,.’ ef :iencycaused by the presence of dual-porosity
code. . code is mapped so that storageis requiredonly
for active grid blocks. This can ai@ficantly reduce
machine storagerequirem enta for reaervo~ having highly
irregubr geometry. Dual-porosityapplicationsinclude
regionallyfracturedreservoirswhere unfracturedaingle-
porosityregionsexist.
Advantages of 3-tom ponent com positionalas opposed
to extendedblackoilm odellingare describedand illustrated
for an actualnest-critical volatileoilreservoir.A aimple
method for reducing time truncation error in implicit
formulationeisdescribedand illustrated. A new bottomhole
constraint function is presented for presezwation of
productionwelltargetratesin com positional m odela.
A new matrix-fracture tranaf
er formulation ia
deecxibedfor the dual-porosity
case. Matrixblockhe ‘and
shape may vary fro m block to block over the grid. The
formulation accounts for matrix-fracturediffusionand
effects of changing gas-oil density difference and
tension on gravity drainage recovery. The
interfaci.sl
approximate accuracy of the formulationis shown for a
number of testproblems where correctresultsare available
from single-porosity
simulation.Resultaare given for a 3D
600-block simulationof a highly fractured nearmitical
volatile
oitreservoir.
Nom encleture
~ Form ationvolume factor,RB/STB
b l/B
%j EOS binaryinteraction
coefficients
 22 IMPLICIT CO MPOS~ONAL SIMULATION OF SINGLE-POROSITY AND DUAL-POROSITY RESERVUIKS SPE 18427

c~ Matrix rock com preaeibility,

I/psi P Phase viscosity,
cp
Cf Fracturecom preseibility,
l/p8i ~a, Qb EOS parameters
c Fluidcom presaibility,
I/psi 4 Porosity,
fraction
D Diffusion
coefficient,
cm 2/see P Phase m olerdensity,m olslvolume
GOC Gas-oilcontactdepth,ft. c1 Shape factor, l/ft2, or interfacieltension,
h Matrixblockheight,ft. dynesfcm
~j, k Gridblockindicesin x,y, z directions 00 Initial
interfaciel
tension
k Permeability,
md % Interracial
tensionratio,afOo
kr Relativeperm eabil.ity T Matrix-fracture
transmiezibility,
RB-cpfd-psi
K K-value,Ki = yi/xi Tz H atrix-fracture
drainagetranamieeibility,
Itzkx
!zX,gy, kz Matrixblockdimensionz,
ft. Ey / kz, RB-cp/d-pai
n Number of matrix blocks in stack wixhinone T Interlock tram miscibility,
RB-cp/d-prn
gridblock ‘or l!ortuosity
Nc Number of hydrocarboncomponents

NXS NY* Nz Numbers of gridblocksin x,y, z directions Subscripts

P Pressure,psia f Fracture
Ap Porn- Pof go Gas-oil
Pc Capillarypressure,
psi i Initial
pce Thresholdgas-oil
capillary
pressure 9. (auperacript)
outeriteration
number
PI Wellproductivity
index,R B-cp/d-psi m Matrix
Pj Co m positional
model variables n Tim e level
q Flow rate w ,O,g Water,oil,gas
s Saturation,
fraction Wo Water-oil
Swc Connate water saturation XPY9Z x,y,zdirections
‘orw Residualoilsaturation
to water
Reaidueloilsaturationto gaa Acknowledgem ent
‘erg
S$ Matrixwater saturationat imbibitionPc W. = O The author acknowledges the extensiveassistanceof
Swes Sge Matrix block equilibrium water and gaa c“ A. Ellis in connection with model code and
docuG station and of M. A. Martinez and C. A. Elb in
saturationa preparationof thispaper.
s$fe~S$e Gfi block equilibriummatrix water and gas
References
aaturationz
S:f m ent
Fracturewater saturationdi$plece 1. Coats,K.‘H.:“Simulationof Gas Condensate Reservoir
Performance:’J. Pet.Tech. (Och 1985)1870-86.
T Temperature,oF
v Bulkreservoirvolume, cu.ft. 2. Redlich, O, and Kwong, J. N.S.Z “OU the Thermo-
dynamics of Solution V. An Equation of State,
woc w ate~oflcontactdepth)ft.. Fugacitiesof Gaaeouz Solution;’Chem. Review (1949)
x,y, z Cartesiancoordinate 44, 233.

Ax, A y, A z Gridblockdim enaione,

ft. 3. Soave, G.: “EquilibriumC onetantaP~om a M edified
Mol fractionof component iin oilphase Redlich-Kwong Equation of State;’Ghem. Eng. SCL,
xi
(1972)27,1197-1203.
Yi M 01fractionof component i in gas phase
Zi W ellstreamcomposition 4. Zudkevitch, D. and Joffe, J.: “C orreletion and
Prediction of Vapo~Liquid Equilibria with the
z Depth, m eaaured verticallydownward, ft. or , Redlich-Kwong Equation of State,” AIChE J. (Jan
fluidsupercom preasibility
factor 1970)16,112-19.

“5. Joffe, J., Schroeder, O.M., and Zudkevitch~ D.:

llVaPoeLiq~ E q~rie with the R edlic~K w ‘ng
Greek
Equationof State,“AIChE J.(May 1970)16,496-98.
~g Dual-porosity model gravity drainage
param etera 6. Peng, D. Y. and Robinson,D.B.:“A New Two-Constant
Equation of State;’Ind. Eng. Chem. Fundam. (1976)
Y Fluidspecificweight,pzi/fL 15,59.
A Pheae mobility,
kriv
I I

/
SPE 18427 KEITH H. COATS 23

7. Coats, K.H. and Smart, G.T.: “Applicationof a the 1987 SPE Amual Technical Conference, Dallast
Regression-BasedEOS PVT Program to Laboratory TX., Sept.27-30.
Data;’SPE Res.Engr.(May 1986)277-299.
24. Golf-Itacht,T.D. van: Fundsm entele of Fractured
8. Lohrenz,J.,Bray, B.G.,and CLsrk,C.R.:“Calmdating ReservoirEngineering
& ElsevierScientificPublishing
Viscosities of Reservoir Fluids From th~k CO~ 1982.
Compositions/’J. Pet. Tech. (Oct. 1964) 1171-70;
Trans.,AIME, 231. 25. W u, Yu-Shu and Kareten Preuzz,“A Multiple-Porosity
Method for Simuhtion of Naturally Fractured
9. Reid, R. C. and Sherwood, T. K.: “i’hePropertiesof Petroleum Reservoirs”,SPE Reservoir Engineering
u t~d edition,M cGraw-Hill Book
Gases and Liquids, (Feb.,1988)327-336.
Co.Inc.,New York City(1977).
26. Kleppe, J. and Morse, R.A.: “Oii Freductionfrom
10. Coats, K.H., Nielsen,R.L.,Terhune, M.H. and Weber, Fractured Reserb’oirs
5y Water Displecemcut:’paper
A.G.: “Sire
ulationof Three-Dimeneion~ T we-Phase SPE 5084 presentedat the 1314 SPE Annual T $chnical
F1OW in Oil and Gas Reservoirs/’Sot. Pet. Eng. J. Conference,Houston,TX., Oct.6-9.
(Dec. 1967)377-88.
27. Beckner, B.L., Tahimoto, K., Yamaguci@ s.,
lL Marathon Oil Cornpany, Littleton,C o1o., Personal Firoozaba@ A., Aziz, K.: ‘Ymbibidon-Dominated
Communication, 1985 M atrkFracture Fluid Transfer im Dual Porosity
Simulatoreflpaper SPE 16981 presentedat the 1937
12. da Silva,F. and Belery,P.: PersonalCornmunication, SPE Annual Technical Conference,Dallas,TX., Sept.
Petrofina,Brussels,
1987. 27-30.

13. Price, H.S. and K.H. Coats, “Dfiect Methods in 28. Lefevbre du Prey, E.J.,“C aecade drai~ageof blocksin
ReservoirSimulation”,
Sot. Pet. Eng. J. (June,1974) a fissuredrese~oir~ Revue de l’institut Franceiedu
295-308. petrole,
Janv.-Fev.(1976),p. 173-178.

14. Coats, K.H.: “An Equation of State C om positional
Model;’Sot.Pet.Eng. J.(Oct. 1980)363-376.
15. W erren$ J.E. and Root, P.J.: “The Behavior of
NaturallyFractured Resenroire;’Sot. ‘Pet. Eng. J.
(Sept.1963)245-44;Trans.,AIM E, 22S.
16. K. and Zeman, P.:
Kazem~ H., Merr& L.,Porterfield,
llNumeric~ sim~a~n of Water-OilF1OW iIIN atura~Y
Fractured Resemoire/’Sot. Pet. Eng. J. (Dec. 1976)
317-32Q Trans.,AIME, 261.
25.. Festoey, S., Van Golf-Racht, T.D.: “Gas Gravity
?)reinagein Fractured ReservoirsThrough New Dual
Continuum Approach”, paper SPE 16980,presentedat
the 1987 SPE Annual Technical C onferance, Dall.cw,
TX., Sept.27-30.
30. Thomas, L.K., Firoozabad& A.:“Tenth SPE Symposium
on R eaervoirSimulation Dual Porosity Comparative
SolutionProject,u to be presented at the 1989 ‘PE
Symposium on Reservoir Simulation,Houston, TX.,
Feb. 6-8.

17. Thomas, L.K., Dixon, T.N. and Piereon, R.G.: 31. said& A.M. and Tehra~ D.H.: “Mathematical
“Fractured Reae~oir Simulation/lSot. Pet. Eng. J. Simulationof FracturedResezvoirPerformante,Based
(Feb.1983)42-54. on PhysicalModel Expenm ents,”Paper P D 10(3)on the
Developm ents in Reservoir Engineering,Proceedings
18. Gilman, J.R. and Kazem~ H.: ‘Improvements k of the 10th World Petroleum Congresa,Bucharest,
Simulationof NaturallyFractured Reservoir/’ Sot. 1979.
Pet.Eng.J.(August1983)695-707.
Appendix A
19. Sonier, F., Souillard,P. and B~skovich, F.T.:
“Numerical Simulation of N aturelly Fractured For quasi-steady-state (QSS),single-phase,m at~fracture
Reservoirs:’paper SPE 15627 presented at the 1986 flow in a bulk volume element V of reservoir,
the ftow rate
SPE Annual Technical Conference,New Orleans,LA, ~
O ct.5-8.
~= -’<*C aft (Al)
20. Sonier,F. and Eymard, R.: “A New Simtitor for
Naturally Fractured Reservoirs:’paper SPE 1600S where @c iseffectiveporosity-compressibility
of the m atrix
presentedat the Ninth SPE Symposium on Reservo& in V. In the cliff
ueivityequation
Simulationheldin San Antonio,TX., Feb. 1-4,1987.
kfi+Kd+kti= l.l~ca~t (A2)
x~xz YaY2 Zaz2
21. Dean, R.H. and Lo, L.L., Ilsim
&tio~ of N atur.zlly
Fractured Reservoirs”,SPE Reservoir Engineering a p/at isindependentof positionwitti”the matrixfor QSS
(May, 1988),638-648. flow so thatsubs~ution from Eq~ (Al)gives

22. Litvak$ B.L.: “Simulation and Characterizationof a2 ~~+=

k~+k~+ = o (A3)
Natwlly Fractured R eservoire;’
Proceedings of the
‘ax Yay2 Zaz2 v
Reservoir CharacterizationTechnical Conferences,
Dallas(April29-May 1, 1985). Academic .Presa,New where V isthe m atrk blockvolume. Ifthe matrixblock h-
York (1985). dimenaiona a, b, c in the x, y, z directions,
respectively,
then Eqn. (A3) can be written
23. Rozsen, R.H., Shen, E.L: “Sim~&n of Gas/0~
Drainage and W aterlOil Imbibition in Naturally &+r~+s_?.!$+Q=O (A4)
Fractured Reaeiwoi.rs/’
paper SPE 16982 presented at ax; ay~ 3ZD

261
 24 IMPLICIT CO MPOS~ONAL SIMULATION OF SINGLE-PORO~y AND DUAL-POROSITY RES5RVOIKS SPE 18427

where
= xla yD = y~ ‘D = Zlc
‘D
(AS)
=$~~=>a~Q .&
r
X b2 Xe x Where tD is ktlIJ~ ca2 and sums are over odd postive
Eqn. (A4) can be solvedfor p usingthe Fourierfinite integera. For the isotropic, cubic block case thisratiois
sinetransformand the resultintegratedto obtain about .03 at tD =.1and fortD >.1 the ratioizapproximated
by the firstterm of the summation,
1
;m
- Pf = Q 83EZE
ijk 2222
(A6)
—~—
- Pf *3
e
-31!
2tD Fm
‘i ‘j y~ (Ai +
Pi-Pf
rB2 j
- fi6
+ w:)
(A16)

The term ~m is volumetricaverage matrix block pressure Thus a 90+ Z decay occursat tD ‘.1or
and each EWm m ationisover odd,positive
integersonly,and

‘i
= iw $j =jfi yk=kfl
(A7)
t . “AL@sL2 (A17)
.00633 k

where unite are CP, psi-l,ft and md for U, c, g and k,

respectively. For many fracturedreservoirs,this decay
time isverys m ~ the orderof .01days or less.
where ~ ~ ~ denotesthe triplesummation of Eqn.(A6).

For the twe-dimensional x-y case (no horizontal

—
— fractures,
N *2),the corresponding
~esultis
=-
—

and forthe one-dimensionalcase (vertical

to the x-axis,N=l),
fracturesnormal
I
(A1O)

where, again,each sum m ation is only over odd, positive

integers.

Warren and Root definedtheirshape factorby I

u= qll
(A 11)
kV(~m- pf)
vhere
(A12)
u= 4N(N+2)/L2,

k iE a representativematrix block dimension,and N is the

number of normal eetsof fractures,N=l, 2 or 3. Kazemi et
alsuggestedthe shape factor

(A13)

and here we use I

(A14)

A tabular com pariaon of U values for the isotropiccase

a-b-c for N=l, 2 and 3 from the analyticalEqna.(ASXA1O)
and Eqne. (A 13) and (A14) is given below Eqn. (23). The
sum a in Eqne.(A8HA 10) can be calculatedto giveu values
for any anisotropyand any set of unequal matrix block
dimenaiona. Warren and Root give expressions for k in
terms of unequala,b, c valuesforthe casea N-l, 2,and 3.

The transient time associated wtih single-phase

matrix-fracture flow can be examined by solvingEqn. (A2)
forirkialconditionp = pi and boundary conditionsp = pf on
all matrix block facw (where p here is matrix pressure).
The aoludon forvolumetric-averagematrixpressureis

ZU
 18427
69

?AILE
1 TASLE 6
mrnms (MAXUX+IACXL7ECI

L ~‘CG‘cc(cc,.
~.&# Wauwis pm8mre 3959.89+
’00 Da, .
CC*)
Co.preni6&iy ef rock, fiamre, ●nd water 2.0x 10-6Ilpd
onCo.psudbiuk, 10+ 1/+
‘LC ‘LG At ,anwokr 91WIKU
-A ~ k!*C,, V2?.odty .5 Cp
‘LG ~ “L1 - CC2) on ViU.mlf”
or 2 op
n“ .9SS12 BB/ST1

CL . OL. cc - Pcy a.
M .c*r ‘ledfy
.96o6 EB/STB
- Wm.
022detuizy .3611 9d/fL
71it& ~tte,billty 1 *d
cc EOUIL2BR1U71 D2LEH)5A, M,U2x g0?&2y .19
ccl “ C.32 Sbqle Factor .08

~-rl
51.tr& bbck &e iJx10x30f2,
CL1 - CG2 Fraetum pm.e.bi7icy 100DO -d
1 2
CL1 “ ccl Framdre pamizy .01
‘L1 ‘i “ Dc, Yi
mid 8X.3X1
yilxi . ;:. Ay
‘i- ‘L1lPG1l ‘ii 7522.
3052.
LO
w U- Lajectiorl rate 200 Srnld
Tota2 Ziquid pmd.c!iori rate 210 SWd
EAcn CELL
2-PHASE GAS-02L 71.2zix k, ,md PC da2a 17

S= krw k
. Pc”o
5A1LS 2
75 r = r
VOLAY2LS OIL MSBIVO1l A PLOTD COk2?0231201S ASD DAfA .02 .705 2.95
:: .055 .42 1.65
T -266° F .26 .85
:: ::45 .11 .3
.7 .2 0 0
RE8C S.
V017(FLUID mxzcnon GAS

M ~ ~ ~ ZAILX 5
cl .5898 .5898 PC1 .8601 .S601 KL1?P1 AmD k70UlE 51? XL262X8TAL BAZA
C02 .0093 .1926 Pc 2 .1243 .0113
.0757 .2176 Pc 3 .00$6 .071s
:: .D409 .0283 21@ix Wme.bi!iz, 290 md
C4 .0300 .0115 M.tl’2Y p+r.adq .225
C5 .0192 .0039 care .Jh..t*, 9.87 en
C6 .0175 .0D17 Mid. dia.tit. .f tube 10.39 c.
?7 .07812 .0053 Core hei2bt 122.8c.
F8 .09517 .0003 W .ter density 1.02 ~W
F9 .04431 %7+ “ .8578 0i3d.~, .811 dc.
wate, V&ity 1 :p
Okzticc.dq 2.3 cp
Eos Pare V.2U. * of core 2114 cc
DATA ~ ~ Vdu.e of beture 1017 cc
BUS8LE PT. Ssm 5333.7 5s53.7 Low 5nJcc2ioa rat. 3.3 CC1.ia
00 ●t B.P. 36.11 36.11 36. IL Ib,lcu. ft lush&jec&?. “m 33 Cel.iu
a at B.P. .2716dpnwlc~
k= ●nd P= data 27

3-cPT EOs B.P.RESULTS k k

-z% ~
~ =* -L .3 0 .75 3.1
Pcl .5898 .13507 .352 .0005 .675 2.3
PC2 .1926 .1670 .S67D .35 .001 .605 1.83
PC3 .2176 .0’364 .1672 .375 .00;5 .533 1.52
.6 .002 .67 1.29
. 11z4pbvR
A%. ,423 .003 .4 1.D8
.45 .Owl .33 .96
.475 .D07 .27 .82
3-STAGE SURFACE SZPARAT30N
.5 .01 .19 .7
~ Iis@4 T& .55
.01s
.027
.12
.06
.61
.52
1014.7 150
: 264.7 80
3 lk.7 6D TABLE 6

TEOMA8 H AL s2E@LE-sLOCC DATA

~ ~ ~
STAGE 1 OOR1 1057 1057 1057 Co.preni6iUCy c.f cock ZM.tucg, ●nd “’tu 2.5.104I/&
M ●Q3x bbck dimwims, Lx, k,y, !X Ift.orloff.
STAGE lYG .6910 .6819 .6708 Permesb@, k=, ItY, k% 196. *m ..3
STAGE 3YL .8363 .8363 8ubb2e-@at -m 5360 C&
.8324 -!. . . . . –J,-
C& dam . ----
C07L2 1489 1510 1489 W.tcr den,ZUY2YZU 5560 m X;’#-*m
0DL3d.wlc, at 6215+ .2344 @m2.
no 1.8530 1.8530 1.8530 .“ .-.. -4-.
0. . ...’., . ..= -d.
.“.-, . - .2213 es
W.ter vkadsy .3s cp -
1) SCF S2.U.1 Cu/BSL u.at 1 liqti OiJ d.+ .t $560 @a ,25075 PcIJ~
2) T.ttiSC?ISTB G“ der,dz, .C 5560 @a . . . . . .. .. .
.,”>”, pm/m.
CU vL!c.c+ ● 5560+ .0274 Cp
Oil viac~ ●t 5560 w .21 C9

H’frh k ,sd P D.m,

XY~CC’2 0? ~OCK v&~ IT2CA2AQ023JB82UM

P
cA?2LLA1Y P116sU1S
090211VXS k
~ ~
CALCOLArBD 162V26L ?L727M ~-?LAOE 1.00 30.0
.66 9.0
400 FT. VEM2CAL COLLIIIR .3 Xi’ .723 2.0
.35 .02 .6
Goc - 835L woc - 285ft. .4 .03 .692 0::
.45 .045 .392 -c6
P - 3D(1-C”)3 = 10s3
.“0 :7”0 .5 .C4 .304 -1.2
LjY - .1662 - .11$1 p4i/5L .6 .11 .154 -4.0
“0 a. .7 .18 .D42 -10.0
.75 .23 o. -40.0
....
,.. ....
,..

z
0. -120.0
rR33TAL F2, DID9-IN.PLAcG lDOOVSM

VE P= ROCK P
c
‘2.-1-ssh k
peso
w
—. . —
L ~ 07L Y2LF.EGAS 21ATE21 OIL ?RSE GAS .2 0. 1.0 1.0
.45 0. .42 .386
2747.1 3319 1226.4 1423 5947 .3 .0227 .31 .255
: 2747.1 3319 1224.4 3518 3727 : .6 .11 .19 .145
2747,1 3319 1324.6 2623 3441 1233 .7 .25 .103 .11s
1: 2?47.1 3319 1224.4 2737 3326 1207 .8 .45 .05 .095
20 2747.9 3319 1224.3 2771 3323 1196 .9 .7 .015 .08s
1,0 1.0 0. .075

-m
 ?ABL6 7 TAIM 10

GAS-232LDATA FO1 1s2s6 ~?. MA?kIf uLOCK

lwKc? OP B31POU01 01 51A712X-~lACfUll COUP061?20WS
Tim. -5753 Dtya (Ed Of EhnT)
;ock CO. pressibMty 5.5x10+ lIPcI
122 Ioj@~d 2472-57$3 DayT
r“cture
co
mF.redbiuk. 5.5x1O* 11P16
ce2t3 -8-l
22.t.r c.* reuibCif2Y p:; Il,a
Permsblh?y, k=, Ity, kg 1.000000.729 NO Dif2ud0. Difl%dm
0.744°4 0.512
M.txix bfmk dirnen.dor.h L=, L ,
YE
122.
Ilk
0.53351 0.343 ~ ~“ ~ ~ ~
z 0.&3675 0.2698
Matrix p-may .3
Effective 2nctum per. eabiUy 60 md 0.35137 0.2078 on, =
c1 .4424 .0805 .2899 .2897

?r.ez.nn VKic? ,002 0.27687 0.1561 x 222 .!3032 .2696 .1306 .1307
Iaf2ia2 preuure 5553.7 F& 0.212760.1138
Ini&2 .** ,.tlxaciot?s 0.15b49 0.0799
022 .9 0.11351 0.05355 c1 .0208 .1769 .s510 .S567
w*tcr .1 0.07721 0.0337 GAS y
GM 0 0.0489k 0.0195 U2 .0071 .7846 .3176
Y .3181
Cd dencizg Aiffer.nce at 5553.7 Pti .1181 pd.% 0. 7i462 0.57603 0.028020.010
Sze u 5353.7 Pda .5722 0.83846 0.67466 0.013650.004215
0.89231 0.77635 0.004950.0012b9
Diftin t0r2u0& 3.5 0.000880.0041S6
0.946150.88606 TABLZ 11
Gtid 12X1X5 1.008Q0 1.04000 0.000000.0
Ax 100 EC. CM 8ATu RAT20#S b~ 5753 DAIS
As 500 ft. 1mbi60ion Pcgoi - .1 ● Dmim8e PcgO
As 50 ft.
CELL Sfxs
cu bjecdon weU completed in ti 5 fayera ●t 1- L k k
.2621
~
PI Wr fqer 1.23 8B-cp/damci 2+-1 .4998 .2624
tijecciam race 2500 MSCPld 3-8-1 .8625 .4796 .4131 .4136
4-8-1(1) 1.ODDO .4627 .5369 .4627
2-10-2 .239 .5044 .1217 .1206
7-’+2(2)
.9796 .4961 .4863 .4860
c1 .8621 0
C 2-6 .1305 0 (1) Activ. foje.cor
P 7-9 .0074 0 (2) Sh.efn 2ojector
N2 o 1.0

Pmdudm we22 comp2eted in 2qers 4 and 5, 1-12.

PI per Llyer 1.23 RB-cplda~ Psi
P20wLw bottomhole pressure 5500 ph.

TABLE 8
EE.uELL

31ATL23 GAs SATURATIOW 1S fIMC 10s T~I S?f2 10 ?I. BLOCK CA22Z RESERVOIR FLUID A
(la p3meAiu7ig. 15)
DIFFERENTIAL EXPANSION
Time Tim e
D ‘y,
. L
Days
— %-
‘“”~ 2“
0 0 300

‘.5kc’
.3173
5 .0069 400 .3690
20 .0273 500 .4020
50 .0670 60D .4263
100 .1300 800 .4531
150 .1878 1000 .6704
200 .2383 7000 .4998
250 .2S17

00 SCF

I.Y
~ .9
RB/STB
TA3.LZ9
R~
BE8ERV02E & DsscM?r309 .8
2.0 “-~ )’ 1000 ~
Rock CO.
p=ti~y 5.3 x 10+ I/@i -.Q
*.

r
w .t.r co mprenLb53Lcr 3.0 x ID+ llpcl .7
w ,ml Vim@ .22 cp “%..
w ok2ticc.ri23. .23 Cp -.
e.
1.06 RB/STB
B“
W●ce. denciw 63.65 lbs/m.5L ‘%.. .6
Q

AZJ 3691 2461 16h0 2 e 1230 3, 1640 20S1 2&61 2661 f~

B.
~v 4101 14 ● 1640.4 3.2461 4921 Ec. 1.5 500 .5

L,yer thkktieia 200, 100, 150 k

Sm.mture 9500-11000 t% subsea
22.* Pc.c. ●.bilicy .5 ~d
M,cdx pmdcy .3
22●zria b20ck size (., y, s fr.et.t’=p..inm) 1.4z.SX45L k \-l
Effective Cr.ctut’. per mC&+Licy 50 md
?C,cture p+mlicy .0005
Ctititud amk-kactute
exchange coef2tc2ent 9 x 107 Ef.%91d.y psi
~ Fescure at 105t39 53. 7000 +
1065022. P, 1000’S PSIA
D DIP h.3 z 10-p STS

2s4
I
S!% 18427

FIGURE 2
FIGURE 4

RESERVOIR FLUID A TIME TRUNCATION ERRoq

DIFFERENTIAL EXPANSION IN 4380-OAY RIJNS

DATA —
—
‘--%
41J STEPS, M
STEP& HO TTEC
TTEC
I
● ● 4S STEPS WITH TTEC
10-CPT CALC’D ● ● EOS 00 9S STEPS WITH T?EC I
3-CPT CALC’1) O 0 Eos .8
1400C

1. 1

1.05 12000
.6
z GAS FIELD GOR,
PO
SCF/STB
1.0 CP
10000

.95 4

.9 8000

.85 , I 1
,-%,
2
1 2 3 4 5
P, 1000’S PSIA 6000
o
FIGURE 3
TIME. DAYS

RESERVOIR FLUID A FIGURE 5

MULTIPLE -CONTACT VAPORIZATION GAS-OIL GRAVITY DRAINAGE

EOS CALCULATED RESULTS

4415 PSIA 266 “F

10-CPT —
hA7
3-CPT O 0

1.0 Pcgo
5a
-\
.9 -\
%?!
\ t
o; 1
Q Sti Swc+sorg Lo
.8 - \
s LIO
\
.7-
\ GAS FLOW

Z=o a + ~ GAS
.6 - b _ FLOW
\
S9
\
1 * r 5b
.5 , i
o pM\AY
I 2 3 4 5 Z=h I
B
+
PV SEPARATOR GAS INJECTED

214
 flGURE 6

WATER-OIL CAPILLARY PRESSURE CURVES

U-J
m. o
0 0
E

A
m
w. A
0 o
P A
Cwo
A
/ —SPM 5x5x1O GRID
lME191TtON swm~. A
0 0 SONIER ET AL *9

D A SONIER ET AL 19
o - -. — --- — ‘e 0
c A
(YW-YO REPLACES YO+YW)
-hAY m
A m.
0 A •l OPM BETAW=1,5
)

A
F
u-l
I 1 w
0 Sw - 1.0
TIME (OAYS)

FIGURE 7
HATFR MBJRIT’1ON CURVF iOx~x30 FT. BI OCJ(

GRID BLOCK IN INITIAL

km=.l md TABLE 4 DATA
WATER-OIL TRANSITION ZONE
m
ul
0 — SPM
-- OPM
A OPM km=.14 md

-DRAINAGE ~
I
P I
. I
\ \
Cwo
I I

s urn

GRID
BLOCK

*l!k+2k!L
U!
m
0
o I-sorw Lo
Sw 4

m
N
9/
0 1
200 400 600 aoo *000 1200
TIME [DAYS)

. . w
.-..
 FIGURF 12
COMP~N OF S PM AND EXPERIMENTAL RFSUl TS ~ R“

KLEPPE AND MORSE EXPERIMENT

0 —KLEPPE ANO MORSE DATA g — SPM

s 0 SPM LOW RATE m o DPM LOW RATE
A SPM HIGH RATE A OPM HIGH RATE
1 1
-1
,4””0

:.

CUM OIL
PRODUCEO -
FROM CORE
(cc)
z.
ml

0-
0 500 1000 1500 2000 o 500 1000 iioo “ 2000
CUMULATIVE WATER INJECTION, CC CUMULATIVE WATER INJECTION, CC

FIGURE 11 FIGURE 13

COM PARISON OF SPM AND EXPFRIMENTAI RFS L!JJs

SINGLE-BLOCK AND STACK-OF-BLOCK

SPM GRIDS FOR GRAVITY DRAINAGE

0
m —KLEPPE AND MORSE DATA
0 SPM LOW RATE
A SPM HIGH RATE , * O STACK OF SIX BLOCKS
1 2xIx37GRID .-
g

0
“

rl

F
m- SINGLE BLOCK
2x1x7 GRID

H- l-B. I
0
ln- ‘SLOPED FRACTURE
lfATER 2xIx37GRID
LEVEL IN-
FRACTURE 7 ““
(cm) c+ .5 FT

- FRACTURES

XfE
Az=q~3,2,1,1,e FT
%-
AX*5,EFT .5 FT
@*5F~
0 ~ 3,001667F~
1
0 500 1000 1500 2000
CUMULATIVE WATER INJECTION, CC k =1500

.. .
227
 FIGURE 14 FIGU R F 16

OVER “{ FOR SINGI E-BI OCK ANQ

@sLu+wL~
-- INAGE, h=iO FT,
W.LBLQM

%og /3

In —SINGLE BLOCK
0 - – STACK OF 6 BLOCKS
HORIZONTAL XY FRACTURES
A STACK OF 6 BLOCKS
SLOPEO X’f FRACTURES
T
0

m
0

‘gm

/
/ / ‘4
/
---
/
/ o
— SPM
0
/ O DPMT
/ . A DPM BETAG=.485

/
/
/
/
0
200 400 600 800 1000
TIME (OAYS)
TIME (DAYS)

FIGURF 15 FIGURE 17

$AS/OIL OR AINAGE RFCO ‘4FRY -GAS/IJIL ORAINA6F RECOVERY

Oxloxfo FT. MATRIX CK 1x1 FT. MAT R IX BLOCK

sgf = 1.0

u-l N A
0 0 A
AO
A
o
0 A
.
w o
0 0
0
A
— m o
0.

m 0 A
0

‘gm ‘gm~-
—SPM
w 0 DPMT
a“ — SPM A DPM BETAG=l.14
O DPMT :-
A CIPM BETAG=l.14 0
6
-1
0“

0
,,.,,00 :L
00 200 400 600 30 60 90 120 150 180 2i0 240 270 300
TIME (DAYS) TIME (OAYS)

2U

..
 SPE 1842 ?

PSFUDO P. FROM SPM GAS/OIL ORAINAGE RESU s

U!NJBF
?Q

VITY DWAGF FOR A i FT. MATR~

~1 /u
OATA OF TABLE 7

w
0

m
0

%:’
— SPM
–- DPM

0“

0
‘i 20 40 ‘ do
,
100 t20
,
140
I
160
TIME *?OAYS)

$ i
,
,0.00 0.04 0.08 0.12 ’716
‘gm
.!2,QJRF 21
QNGIF GRIR_RIQGK Depiction

AZ=24 FT. h=4 FT.

Sgf=o.
0
0

OATA OF TABLE 7

1
m
0 r%.
0 A
In
0
A
(0 0“ A
0 o A
0
o
In e — SPM \
0

v
0“ i=
2000 2500
TIME (OAYS)

‘gm~- z
w
Ii-1.
——. ___ ___ -a
0
‘SPM Sgf=l.O
all -m
p 9-
0

I
–- DPM Sgf=f,f) PSIA IFT
0.
0 -w
O OPM (SQm/Sgf)Sgf=.5 g —PRESSURE
. ‘- IFT
/ .m
0
g /

w. /
-w
/
0
0 /
+ , , (
g<
, ,
‘o 200 r
400 600 800 1000 o 500 1000 1500 2000 250;
TIME (OAYS] TIME (DAYS)
 EwBE22’
J3AS-FIIJFD MATRIX WQCK
DATA OF TABLE, 7
AZ=24 FT. h=4 FT.
‘gmi=g
la
0
—SPM

--
0 RPM
SPM
EILJEEm

O!!F
o c o
w
1’ —5PM

O OPM
So f=l.

So f=l.
O

O
In A DPM OFF
0
o ‘gm

N
=? 0
0 0
n l-)
0 0
~+-----
0

I-u
0
0
A
I A OPM Sof= .5

-- ‘CORRECT’ OPM Sof=.5

I
:IA
‘gm \ A

\ AA
m --~
0“ -A-3— - - - - . - .
:~ A
A
400 aoo 1200 1600 2000
TIME (DAYS)
:~
o 200 400 600 000 1000
TIME [OAYS)

FIGLIRE 23 FIGURF .%

I f)CK GAS/Oil OF Pi FTION FFFFCTS OF PARAM.TFRS ON DP~

~Q ~

4 FT. MATfiIX !3LOCK

p ‘ND‘gm ‘s ‘rhfE
DATA OF TABLE 7

0 m
0
10” ——_ 0
m
/
/ 0
—0
0 /
0. /
; / A
/ A
0
0. m!
z 0

— PRESSURE
0
P; ‘gm
PSIA- /“

1/
/
0 N
0
/
m-
w
0 0 A —BETAG = 1.3
/ 0 BETAG = 1.0
/.
b A BETAG = 1.0 LAMBOA=CONSTANT
0
0
.
/
~- 0

0
w
, , g------
‘o 400 800 1200 1600 2008 130 200 300 400 500 600 700 aoo 900 1000
TIME (DAYS) TIME (DAYS)

270
 .

-SENSITIVITY O DPM CROS!j SFCTIIJNA1 FFFECT OF OIF FUSIOtIi oN op~

~ SULTS
CONSTANT PFIESSURE SEPAt3AT0R-GAs
INJECTION D1FFUS1ON = O. CONSTANT PRESSURE N2 INJECTION
—BETAG=l.3
‘ 0 BETAG-l. o 0
m ‘NO OIFFUS1ON
A BETAG.l. O LAMBDA+ COt.lSTANT

1
‘1
0, OIFFUSION
o 0
0
0
OIL o
RECOVERY
% OOIP

c.-

GOR %
GOR
SCF/STB o
SCF/STB
2-
m

0 A 0
“-
0.
1% A
g
0
I o~
-0 1 2 5 6 7 1
0 1 2 5 6 7
TIM~ (y E;Rs)
TIfE (YEiRS)

FIGURF 27

FFFECT OF DIFFuSION IJN DPM

EIGURF .%

CONSTANT PRESSURE SEPARATOR-GAS INJECTION

Pc OF TABLE 6
—NO OIFFUSION
0 O OIFFUSION
e-l
I o s~e=.475
0 u!
0
lv- 0 0
0
OIL
RECOVERY n m a
% OOIP %-

0-
T
0
0- , , 1
s
Wm

m — SW
0
GOR O dPMT
SCFA3TB A OPM

W
7 I
00 200 400 600 800 iooo
TIME (DAYS)

27””
 ~E 18427

UGl!MJQ
ITION CURVF 1=30 FT. Bl OC K WATFR TION CURVF IOX1OX3O FT. B! 0~

Pc OF TABLE 6 LOW Pcwo CURVE

swe=.544 Swe~.546

(n
0

0 0 o 0
0 A
o
A 0
A U-I 0
0 A
o o
A
OA 0
o
A s urn:- 0
o
0

1/
— SPM —SPM
O OPMT OA O OPMT
A DPM m A A DPM
0

r
I , , I
200 400 600 aoo 1000 00 200 400 600 600 1000
TIME (DAYS) TIME (OAYS)

FIGURF 31 FIGURF 33

TION CURVF 10XIOXIO FT. BI OC K INVARIANCE TO FRACTURFWATERSATURU

~
LOW Pcwo CURVE

sHe=.4Ell PC OF TABLE 6

DPM RESULTS
a
4
A 4
A
A A

{“
A A
m
v
0
A
0
A

%m
—Swf=l.o
m —SPM
0“ O DPMT A Swf=.5
A DPM BETAW=2.7

N
, 1
00 200 400 600 600 1000
TIME (OAYS) 0
0 #
200 400 600 600 1000
TIME (3AYS)

,. - —..
I ?72
 FIW+F 34 FIGURE X
RIUM MATRIX WATFR SATUR WATER~BITION CIJ~ 10X1OX3 O FT. BI 0~
~li
OATA OF- KAZEMI ET AL (TABLE 4)
AZ = 50 FT. h=lOFT

m ~
0 —1 MATRIX BLOCK CALCULATION 0
•l CORRECT
– – DPM
0 0 0
m
m
0 0
v

g
0
s s wm
nme - 6EI.QMEzHOJl
—5X5X1O SPM
m - - 2X1X1O SPM
v
0 o 2X1X1 DPMT
m
0 A 1x1x1 DPM
n 2X1X1O SPM kIfl.2

/ In
N lx
I
0.2 0.4 0.6 0.0 1.0
Swf TIME (OAYSI

FIGURF 35 EL!WIF 3Z
3D IINF-ORIVE WATFRFIOOO FXAMPI F BIOCK FQU~IuM CURVFS . 0 FT.

OATA OF TABLE 6

——.. y.4z.41
3
0 -- OPM
0

-1
—OPhfT
–- DPM (O+OIL RECOVERYI
In
la.
G-
0

/
m /’. /
0
(n” u-l- /“
0 /
,,/’
s Wm
MATER ,...-
CUT U-1 /
(m% / -.
0-
/ 0 /
/ /
0
0
tll-
In
m. y
0
0/ <
/ /
u-l /
(IJ
o- I I
0 2 8 10 OO.O 0.2 0.4 0.6 0.6 i.o
T;ME (YEA&)
Swf

m
 Y AND WATEFIQJJ

Bx8xI GRID
AZE30 .FT, h=30 FT. q=200 RB/D RESERVOIR THICKNESS = 30 FT

0
LO-
a
A --”-
//”
l%- A
C&

.-
0 A 0 /
w“ =r-
/
% OOIP ~- % UOIP 0
m- —SPM m-
—DPM
-- DPM –- OPMT
0 0
tu - m!-

-IL,,,,,,,,, ---
—
0 0
*-

0 0

0
m-
% WATER
CUT
i%-

0
w-

01 1 0
~
I
I
‘o 400 800 1200 1600 2000 2400 2800 3200 ‘o 500 1000 1500 2000 2500 3000 3500
TIME (DAYS) TIME (DAYS)

~uRF 39 FIGURE 41

I TNF-DRIVF OIL WVFRY AND WATFRC~ TOTAL FIELDA OIL RATE

(91- DAY AVERAGE)

AZ=90 FT. h=30 FT. q=600 RB/O

300

i%-
C+ ‘

% OOIP
l%- —SPM
- – OPM

r
0
ru- 200 -
0.

0 MMSTB
DAY

0
m-
100 “
% WATER
CUT ~

L
ln-

~.
-1
z~ 1200
TIME
1600
(DAYSI
2000 2400 2800 3200 0
74 76
, t ,
76
& t

60
,

82
,

84 86
, 1
88
 SEE 18427

FIGURE 44

TOTAL FIELD A RESERVOIR A GoR

GAS INJECTION RATE
f91-DAY AVERAGE]
— OBSERVED
400 r n 1 000 DpM

I.n\
Sge -DEPENOENT UPoN

[1
DENSITIES AND TENSION
WITH DIFFuSION

3001- lull I
u
MMSCF
DAY
HI n nil

u
GOR
SCF/STB

I
TIME
II
86 68

N2
FIGURE 45

INJECTION STARTING 1981

FIGURE 43 I (1 1

15000 ;:
NO
CALCULATED FRACTURE SO VS. TIME DIFFUSION+ \
I 1:

GAS INJECTION WELL 39, LAYER 2

. ;
I 1000 - !
I.c ‘::
/ 1, ,4’
%-1
:
9000 - I
.8

7000-
.6
GOR
SCF/STB
sOF
5000 -
.4

3om -
.2

0 1000
76 78 80 82 84 86 * , ,
74 76 78 80 82 84 86 [
TIME
TIME