You are on page 1of 16

e

Wety

SPE 39745

EOS

Simulation

and

GRNN

Modeling

of the CVD

Behavior

.,

of Petroleum Engineers

of Retrograde-Gas

Condensate Reservoirs
Adel M. Elsharkawy and Salah G. Foda, Kuwait University, P.O. Box 5969, Safat 13060, Kuwait

m~m~frn~
~~inm~wbmmod

bymsPE-cOmmw
fdtOwhlgmvbwd
bytiwtimr(s).~da
m~,n

~*NJlh~&ti~c+~mEtitimw~b
~bywmtiil.hn-.u~,
*nOtmmnmily mtbct8ny
-dhWd~m~t&-a~ti
~~ti
=-w~ti~byE-%miausofthg~d
~
E~
Uochonb
~rm.
~
.--~ww~~m
k~
pulpMnwmuutfhO~tilh,
$Ocbtyd~
~~bnb~ln~b
m8&ktml@m
~dti~*M-wk
s~mtibmpbd.
h~mu,tcontdn~
~dwhN88ndbnt10m
wp9pbrm~.
wM~,
wP.o.
13cll
m, Rklurdmm,
1%~,
u.aA,,ralol-e72-r3e2.w
.

,.

that the proposed generaI regression netlraI network; l~lodels


are accurate, valid and reliable. Tl~ese models can be used to
forecast CVD data needed for many reservoir engineering
calculations in case laboratory data is unavailable. Tlle
GRNN models save computer time involved in EOS
c:dculations. The.study also SIIOWS
tl~iltonce diese models are
properly trained tl]ey can be used to cut expenses of frequent
sampling and laborious experimental CVD tests required for
gas condensate rcsemoirs.
Introduction

A reliable prediction of tf~e performance of gas condensate


reservoir during pressure depletion is necessary in evaluating
new fields and determining tl~eir reserve. Tile predicted
performance is i]lso used in material balance calculations and
making economic decision regards planning future recovery
processes. SUCIIprediction of gas condensable performance
can be obtained from experimentally measured data on
constant volume depletion (CVD) test using bottom Ilole
samples or St]rfi]ce recombined samples collected where the
reservoir pressure is above tl~edew point pressure, In cases
wl~cre CVD data is not available. equation of stnte and
empirical correlation are used to predict tl~e depletion
performance of gas condensate reservoirs.
Cubic equation of states (EOS) are used extensively in the
petroleum industry for studying volumetric and phase
behavior of petroleum reservoir fluids and to cl]eck \l\e
validity of Ihe PW data and measured saturation pressure
EOS Models are also used for separalor design, gas plant
design. compositional simulations. and vertical perfornlance
calculations of gi]s cot~dcns:](cs.
A numero((s ntlmbcr of cquatiott of states Ilave beeii pllblisllcd
in tile literature to model reservoir fluid plu~se bel~avior in
general and ~iis cot]dctlsi]tc in pilfiicldi\/-*. Generally. tile
performance of tlie EOS is good for simple hydrocarbon
systems, predominantly oil. However, it deteriorates for plulse
bchavtor modeling of complex hydrocarbons such as volatile
oils and gas condensates especially in the retrograde region9,
An important step in any meanin~lt use of tllc EOS based

Abstract
Currently, two approacl]es are beij~g used to predict the

changes in retrograde gas condet Iwqte composition and


estimate tl]e pressure depletion bell:~vior of gas condensate
reservoirs. ~le first ilpp~OitCh uses the eqltation of states
whereas the second uses empirictil correlations, Equi][ions of
states (EOS) are poor predicti~ e tools for complex
bydroearbon systems, Tlie EOS needs adjilslment against
phase bel~ilvior data of rescwoir fluid of knolvn composition,
The empirical correlation does Itot involve numerous
numerical computations but their accuracy is limited.
TIds study presents t\vo general regression neural ne(work
(GRNN) models. The first model, GW
1 is developed to
predict dew point pressure and gas compressibility at dew
point using initial composition of numero{is samples while
tl~e second model, GW2,
is developed to predict tl~e
clmnges in well stream e~uent conl]}osition at any stages of
Ciin also be used 10 dc[crmine
pressure depletion. GW~
tl~e initiaI reservoir fluid conlposition using dew point
pressure, gas compressibility ilt dew point and reservoir
temperature. Tl~ese models are based on analysis of 142
sample of laboratory studies of constant volume depletion
(CVD) for gas condensate systems forming a total of 1082
depletion stages. The dati] bi~se represents a wide range of
gas condensate systems obtiiined worldwide.
fl~eperformance of the GM
models has been compared to
simulation results of tl~e equation of state. Tl~e study SIIOWS
i 87

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE SP~ 39745

,.
,=.

facior colild result in significant differences in EOS


predictions. It is well known tlu~t a typical proble!n \;ritl]
usii~g tile EOS is tl]e difficulty in describing petroieun]
frictions constituting C7+. Tl~ese fractions significantly affect
EOS predictions. Rcstllts of WI~itsons work sugge~ted ll}al
C7+ CllilrilCtcriZ:ltiOllII;IS significant influence 011 EOS
predictions oftllcrcscrvoir tluidbel~avior. However. W1li[son
fo[ll~dtl~iltit is di~lclllt to n]ilke objective collcllisiol~s~~bollt
~il~icl~C7+ cllilracterization istllebest. H&-reporrEdIlle
tclnpernture i)nd con]positional dependence of t[~e .bfi~i]~
interaction, Wllitsoa found tlult saturation pressures may be
ovcrestinulted if Cl- C7+ interaction n~atcllcd :It, one
ten]pcrilture is used to estinliite tl~e saturation pressllreat
lower tcnlpcrillure.
Recent years l~ave witnessed a steady increase in tl}e
applications of artificial neural networks in, .petrolellnl
been
Engineeriltg2q-J4. Neural net\\ork teclllliq~le:-~~!v~-~~lso
-75.3it
,
used in stlldying reservoir fluid pl~i~se behavior
in[crprcti]tion of well testing data 39-4Z.deternlination of pore
press~lre43.
.,
.
-.

com@ti6nal
model is salisfilcloiy agrecl~]cnt bct\vccn
Iaboratoti PVf d~lta and EOS rcs(llts. AIIlio\tglI, san~pling
gas condensitte isa COSIISprocess al)d. soll~ctinlcs. subject to
errors h =ing
ini[ial reservoir pressure and tcnlperature.
tl]e laboratory d;ltil \vitll aII tile errors ii contidns arc needed
for tuning of tI]e EOS1.
Knotvlcdge of tile plI:ISC bcl]i]\ior Of tlIC giis Condellsil[c
systenls is in]port:lnt 10 prcdici res.:t~oir perfornlancc and
future processing needs. Pllasc bcl~:lvior calculations fron]
EOS are desirable bcciltlse expcrinlental dctcrnliniltion [s
laborious and expel]sive, Tl~e degree of confidence in
equation of stille results incrcilscs i]s EOS is tuned to n]iltcl]
exprimcntal]y n~ei~sllrcddiltil or i]c[lli]lticld di~ti].Once tile
EOS is tested against actual data for a given fltlid. it is
generaI1y assunled tl~i]t it ciln be usc:d to describe tl~e pl]ase
bekavior of tl~atfluid at any condition1.
A number of sttldies l~il\e been reported co[]lpi]ring EOS
results and li]bor;~to~ PVT datii for v:irious kinds of reservoir
ftuids2-23. Soulc oftllcse studies cn)pliiisi~~ spli{[ing tile C7+
fictiou as tl~e key factor in il[t}lining significant n]iltcl~
bet\veen l~lboralory data and EOS rcs(llts24. Tile degree of
splitting, tl~eC7+ fractiion depends on tl~canticipated rccovcry
process. : Gas injcc[lon process wi[l~ Viljorizing gas
pllenonlcna
require SO1llC\~llill nlorc splitting
tIlilll
depletedAvilter flooding proccsscs, Otl\crs enlpl~i]size
adjustnlen[ or regression of sonic of tl~e EOS paranletcrs 10
matcl~ tl~e yturatiol] pressure or tl~c PVT datals. Katz itnd
Firoozabadi used ,yuclIlane intcrtict 100 cocmcicnt to study
gas condensate belul\ior. _dc!ypoint pressure and pergcnti]ge
liquid dropout, ~tey Corrcli)(cdbitli];i~s bctwccn illc[l~ilticill]d
petrolcun] frilc[ions and used [Ilcsc bini)ries to predi~ dew
point pressure for ~)s Condct]si][c. PcncloIIx and Ji]in used
tl~e EOS tO predict tl~e properties or gi]s colldcllsiltc usitlg
tl~rec possibilities: I) lllctIli]Ilc in[crilction only. 2) Ilictl]i]ne.
etiurne, and ptiopi]i~c. 3) an tl~e alk:lncs liglltcr tl]an .C6 as
adjustable paranleters of tl]e EOS, They noticed tl~at tI]e
observed difference bct\veen tl~e results of usin~~tlle tIlree
possibilities were not significant. Pcdcrson et al. reported
tkat setting tile binary interaction (Ku) equitl to zero for all
nonllydrocitrbon l}ydroc:~rboncon]poncnts silvc con)ptlter tinle
since usna~~ tIleir concentrations ilr~ less illilll lL}%), COiltS
and Smart studcd tl}e pllasc b~llil\~it)ror gils condcnsaIc
using tI~c EOS. n~ey reported tl]ilt EOS is generally not a
good prcdicti\e tooi ror silntllilting tllc b~llil\ior of gi]s
condensate and e.<tei~sivespIit[ing tl]e C7+ rrilction to maIcll
tile laboratory data is gcllCrillly tll]l)~ccssil~, Tllcir rcs~llts
snowed tile poor predictive i]bility or tf)c EOS witl~o(it
adjustn]ent or regression of tl]e cq[li][ion of s{ilte. paranlcl~rs.
tl)iit EOS ci]pilbility inlproved wl~ei~
Tllcse rcsrdtstilcd
EO~p:lrillilc[cis \vercr6~resscd tI}i~ii~l)cn il\~i]s~]~tIstcdoF
tl~e plus fractio!~.~veresplit. Wllitsoli st(ldied tlie effect of
clk~llgillgtI~ecritic:]l pro~&fl[cs:iltd s})litting ttle plus frac(ioi)
on tllc EOS sin]uliltig]l of gi]s coltd~llsi][c. HC sllo}~cd [Ililt
rcl:lti\cly snli]II differences in criticill properties i]ndilcccntric

Tile objectives of this study can be sumnulrized as follows:


1) Assess [IIC accuracy of tile EOS to deternlinetl~e dew
point pressure of gas condensate. in ,case ~t. i~~~ot
.
nle.iisured. using two ditTerent sci~enle of clulrilcterizing
tl]e plus rraction.
2) Sin~ulate constant voltln]e depletion performance of gas
condensate using tile adjusted or regressed . ...
EOS.

Adjtlstnlent oftllc nlolcculnr weight oftl~e plus fraction


orregrcssion ofll\e EOS parameter is usedto nultcl~tlle
si)tur:)tion pressurc of [Ilc gas condcnsa!e as fi first step
for sinlulaling Ille CVD.
3) Develop a Ile[lri]l ne[work nlodel capable o~pre~clirig
tl~e dcw point pressure and sinlulating tl~e.deplefion
perforn~ilnce ofgas condensate reservoirs using constant
volunle depletion data,
4) Con]pare tIlc perfornlancc of tile proposed network n~odel
and tl~csin]~tliltedbelui\ior of tile equation of state to tile
experinlental[y nleasured data on constant volunle
, depletion.
Present study

Collstill)[ volllnle depietion data of 142 giis condens:lte


si]nlplcs\\erc ilvilili]b]efor this study. Tl]e dat:l -include twoPlliisc giis deviation, factor (Ztp), cunlulati veproducedgas
(Gp). iind ilil:ll~sis of tl~eprod(lced gas condensate for a total
of 1(182stilges of depiction. IJescription of tile data is given in
Table ( 1), Tl~e di]tii represents analysis of a }vorldwide range
or gas condensate systelns collected world wide.
CVD
volunle depletion (CVD) is perfornled on gas
Col)dcl]siltcsili]d voIatiIe oils to simulate reservoir depletion
process iind conlposit ionill variation, Measured data ~an bc
..==:..
.,.
~otlstiil)t

188

SPE SPE 3W=.

EOS SIMULATION

AND GRNN MODELING OF THE CVD BEHAVIOR OF RETROGRADE-GAS

Simulation

used in nlaterial balance calculations and recovery process.


Oewsionally, vshen a gas condensate is first discovered, .a
mse~oir fluid study is not al}vays conducted, ThIIs. the
eo~sitional
data of the CVD In:Iy not bc available for
resefimtr simlation
or Inilterial b:lli]]~ce C:llCtlliltiOIIS.
Therefore. ~OS and correliltions, if lhcy are i]vaili]bie. are
used to predict fiuid properties and conlposition i~t depleted
stages. Geneially, constant volunle depiction test begins at
dmv point pressure TdJ. Vohlnle at saturat ion pressure is
used as a reference vohln]e. In this test, pressure is redticed
in steps. usuaIIy 4 to 10 steps. a[ rescivoir ten]perature. ~er
each pressure red~iction, nlcrcury is ii]jected into the Pm cell
and gas is ~vhhdra\vn at constant prcssllre unt i] the original
volunle is reached. Vohln~e of gas !vithdratvn frotn the celI
(G ), conl-@sitional :]l~idysisof the prochlced gi~s,and liquid
vo funle condensed in the ceil (SL) expressed as a percellta~~
of the reference voIun]e ire nlei]sured at each pressure
Volun}e of the produced gas at ~iicll pressure provides
ifloru~a[ion about gas condensate rect~vc~. Col]dcnsed Iiqllid
vohune represents the Iiqtlid saturation during depletion of
gas condcnsrrte. The test also yields conlpressibility f:lctor of
the tlvo- phase (~p) present at ei]ch pressure depletion.
Although laboratory experinlcnt (CVD) forn~ the cornerstone
for determining fluid properties and depletion prfornulnce
of gas condensate reservoir, such cxperi nlellts CaII Ilot be
peflorn~ed for all conditiolis. Therefore, EOS nlodels and
artificial neural nel\vork nlodels, if tllcy are avi]ili]ble. can be
used to predict pcrforlnance of gas C(li]dcl)si][ereservoirs.

RESERVOIRS

of CVD

The first step in establishing the EOS parameters for studying


depletion performance of gas condensate reservoir is to nlatcb
the de\\ point presstlre predicted by the equation of state to
the experin]entillly nleastlred one. This nu~tch cou[d be
i]chievcd by adjusting the equation of state paratneters,
ChilraCteriZiItiOllof the CT+ fraction (criticaI properties),
adjuslnlent of the n]oIecular \vcight of the plus fraction. or
controlling the bi[li~ry interaction paranleters m~).-- The
binary interilction paranleters are equal to zero for
hydrocarbon-hydrocarbon
interaction and nonzero for
interaction bet\vcen a hydrocarbon and non-hydrocarbon and
Ki ~ CaIl be seen aS
unlike pairs of non-hydrocarbons.
indicator of discrepancy in size and po1arity bet\veen t\vo
n~olecuIes. Table (2) S11OJVS
the binary interaction coefficient
used in this study. TJVOnlelhods are considered in this \vork
for tuning the EOS paranleters. The first nlethod involves
adjustnlent of the nlolecular weight of tl~e plus fraction to
nu]tch lhe n]ei]sured de~vpoint pressure. The second n]ethod
en]philsizes regression of the EOS pilralneters to match
experin~entally nlcasured dc~vpoint pressure. The adjusted or
the regressed EOS at dc\v point is used latter to sinluIate
con~positiona] changes and recovery of gas condensate dllring
pressure depict ion for oun]erous sanlples fronl the data bank.
Sonle of these sanlples are presented and discussed later in
this study,
General Regression Neural Network Models
Several dificrent ncurid net}vork schenles tvere tested to

nlodcl the CVD of gas condensate reservoirs. Anlong the NN


trilillillg schen~csilrC:
1) Bi]ckpropagi]tioll \vith n]on]entunl.
2) Acceleriltcd Backpropagnt ion.
3) Levcrnberg-Marquardt BP net}vork.
4) E1nuIn reclirrent net\vork.
5) General regression net}vork.
It ~vas fo~lnd that the general regression net~vork (GRNN)
s{Icccssfullypredicts the detv point pressure and forec:Ists the
CVD performance of gas condensate reservoirs. Therefore,
the present study introduces t~vo general regression ncllral
net}vork (GRNN) n]odels, The first nlodel, GWI,
is
the de~v point pressure ii~if gas
used to estinli]te
con~pressibility filctor at the detv point pressure. The second
lnodel, GRNNM2. is used to estin}ate the conlposition and the
Clilntlliitive prodllction of the produced gas at any stage of
pressure dcplct ion for gas condensate reservoirs. G*2
can also be llsed to cstinli]te the initial con}position of the gas
Col)deitsiltcin cas5 it isunavi~i]able. Detail Of the G~
is
explain else\\Ilere .

EOS Modet
Soav-Rcdlich Ko\vng equation of state (SRK-EOS) is used in
this study to caIculate the de\v point pressure and sin~ulate the
gas recm;i~ and con]positionai changes dilring pressure
depletion oFgiis colldetlsi]te rescmoirs.
Calculation

CONDENSATE

of dew point pressure

De\v point pressures of 142 sanlples of gi]s condensate ~vcre


calculated using SRK-EOS and conlpi~red with the
ex~rimntally
nleasured dctv poiltt pressure using t~vo
nlethods of charilctcrizat ion of [hc p]IISfri~ctiol~. For all the
samples presented in this study, tllc C7+ frilctiol~ }vasdivided
into C7, C8, c~, a$y C lo+ ~lsillg the spiit til]g schclne
described by Pedersen . TJVOn]cthods i~reconsidered in this
study for chariicterizing the C 10+ fract~~n of the gas
condensate. The first nlcthd tlscs Cavc(t correlation to
eaIcuIate critic~]l prcss~lre and criticalg tcnlperature,
Thoolassen et al correliltion and KcsIer-Lee correlation to
calculate the accentric filCtOrfor thl: pltls fraction. In this
study. this nlethod is rcferrc~ to as standnrd cll:lnicleriz,]~~olt.
The second n~ethod uses correlations of Pcdersen et al to
calcuMc the criticid prope~ies and i]ccentric filctor. This
n~ethod is referred to ils the op{iol]i]lc ltilrilclcri~.:~tioll.

GRNNMI

Ani]lysis of (he iniliili con]position of 117 st]l~lples(tr:lil;illg


~nlples) fronl the dilta bi]]~kdescribed in this study is used[o
develop GRNW 110 cstinulte the detv point press~lrc,and gas
189

.
4

ADEL M. ELSHARKAW

AND SALAH G. FODA

Ditta of the
compressibility fi~ctor at the dew point.
remaining =mples from the data bilid( (testing samples) are
used to tes[ the ~i]Iidit) and :]cc~lriicj of the proposed model.
Input dilt~] to the G~
I inclu(lcs initial conlposilional
analysis of the non-hydrocnrbol)s (H2S, C02, N2).
hydrocarbons (Cl through C7+), Inolecu]ilr weight and
specific gravily of (he heplane plus friiclioil, and reservoir
temperafilm. Outputs from the nlodel i]re dcw point pressure
@d) and gas comprcssibil~ty f:lCIOriit tile dew point (Zd).
The designed GRNNM t. Fig~lre (l). has three slabs. 15
neurons tihe
input slilb. 180 neurons in the middle slab
(about 33% more neurons], and two 2 neuron in the output
slab with initial smoothing factor cs of 0.3. The city block
norm is@
for convergence.

SPE sFE 374S

GRNNM I for the training and testing samples. Tlfis table


reports thilt the maximum deviation in calculated pressure by
EOS is 1525 psi and nlaximum absolute error is 53V) for the
st:ll]di}rdcl~:lfilcteriz~ltiol~.1753 psi and 4 lVO for the optioaat
chari~ctcriz]tion. 93 psi il[~d 1.8% for the training samples,
and 723 psi and 37/0 for the testing samples. These results
and others Sltowfn in Table (3) proves that GW
1 is
successfully developed and trained to predict dew point
pressure for gas condensate reservoir with better accuracy
than using the EOS.
Cross plots of measured dew point pressure versus calculated
dew point pressure by EOS for tile standard and optional
character ifi]tion are shown in Figures (3) and (4),
respectively, and crossplot of measured versus estimi]ted dew
point pressure by GRNNM 1 is shown in Figure (5). Fi$lres
(3) and (4) show some scatter of the plotted points around the
unit slope line. howe\er, Figure (5) shows almost perfect 45
degree line of all plotted points for the training and testing
samples which indicates excellent agreement between
meanlred pressures and estimated dew point pressures by
G~l,
Table (4) reports errors in calculating gas compressibility
factors at dew point pressures (ZdP) by EOS for the two
methods of characterizing tl~eplus fraction as well as those by
GRNNMI for the training and testing samples, The
maximum de~iation in calculating (Zd ) by EOS is ().364 for
the standard characterization. ().6+4 for the optional
chilrilcteriz~~
tioi], whereas the maximum deviation is U.O1 and
t),03 by GRNNM 1 for the training and testing samples,
respectively, Maxinlum deviiltion, Maximum absolute error
percent (EIll:lx), average retative error percent @r), average
absolute error percent (Ea), and correlation coefficient
percent (R) reported in Table (4) show that GRNNM 1 is
properly trained to estimate Zd more accurately than the
EOS for the Iilrge. range of sampYes that was used to develop
the model and successfully tested to estima(e Zdp for the rest
of the samples.
Cross plots of tneasured versus calculated gas compressibility
factor at dew poiby EOS for the standard and optiollill
cll:lr:]cteriz:ltioll ilre shown in Figures (6) and (7),
respectively. and cross plot of measured versus estimi]ted gas
compressibility fiictor iit dew point by GRNNMI is shown in
Figure (8). These plots show little scatter around the unit
slope line for the Zd by EOS methods and no scatter in the
=e of Zdp estllll:lte#by GRNNM1.

GRNN~~
The GW2
is designed to estinl;lte the conlposilion and
cunndittive prodnccd giis at any sti]ge of pressure depletion of
gas mden~?te reservoirs. The n]odcl is developed using
compositional anaIysis OF922 stages of pressure depletion of
gas cotiensate samples i~sa lraining set from the data bi]nk.
Compositional data of the rest of the depletion stages (testing
set) are used for test ing the v;llidity, accuracy, and
performance of tl~e proposed model. The GW2
uses dew
point ptissnti,
compressibility itt dew point. reservoir
tempcmture and rescnoir pressure (or press~!rcs) as input
data to estimate the conlposilion of the produced gas. nonhydrocarbons ( H2S. C02, N2), hydroci]rbons (Cl through
C7+). moleculi]r weight slid specific gravily of tile l~epti~ne
plus (both needed for cl~i]rilcteriziltioi)
of the C7+),
cumulative produced gi]s, and [tvc) pliase compressibi lily
factor for the remaining gas (Zt ). NIus, tI~e model C:In be
used to forecast consti~nt \IOIUIIIC t cplution performilnce of ~ils
condensate reservoirs in case expensive and laborious
The proposed GRNNM2.
experimental data is unavililable,
Figure (2), has 3 slabs. 4 neurons in the input slab, 1500
neurons inlhe nliddIc slab (i]bout 60Y~Jn~ore neurons), and 16
neurons in ii]e output slilb with initi;ll snloo[hing factor ~ of
0.3. The cily block nornl is i]lso used for con~crgcnce. Laler,
in thissludy, pcrfornli]nce of [he GRN2
ill predicting the
change in composition and cun~uIati\ e produced g:ls for sonic
of these siunplcs is co[npiired to EOS predictions.
Results and discussion
In this section calcldiltiolls of dcw poini pressures i~nd gas

cornprexsibility factors at dcw point by EOS as well as by


GRNNMI are discussed, AISOdiscussed in this section is the
simulation of CVD using the E05 and modeling CVD
behil\tior ofgiis condensate resenoir by GRNNM2.

GRNNM2 modeling and EOS Simulation of CVD


Performance of the GRNNM2 to predict the changes in
composition of the produced gas, cumulative produced gas,
and two phase gi]s compressibility factor of the remaining gas
at Vil rious stilges of pressure depletion ik discussea ill this
section. Also. son]e selected samples fron~ the d~ltilbank are
presented to show simulation results of CVD by EOS and
GRNNM2 modeling of CVD tests.

Dew point calculations

Table (3) sho\ts the ilcc[ir:)c~ of tllc de\\ point pressures


~lcllliitcd b} EOS for the two methods of chilrilcterizillg the
plus ~raction and the dew point pressures csti ll~iited by
190

SPE SPE-397k

EoS SIMULATION

AND GRNN MODELING OF THE CVD BEHAVIOR OF RETROGRADE-GAS

CONDENSATE

,.=..
RESERVOIRS

simulation

results of CVD obtained


by adjusting the
weight of the hcptane plus fraction at dew point
sho\ved better agreement with experimental data than that
when the EOS parameter were regressed to match the dew
point,

GRNNM2 modeling of CV6 tests


Table (S) sl~ows t~cctlii]cy of GRNNM2 in predicting
compositional analysfi of the produced gas, cumulative
produced gits, and two phase compressibility filctor of the
remain;rig gbs at various stages of depletion for the tri]illing
samples as well as testing samples. The model succkssfidly
predicts composition of the produced gas. H2S through C7+,
of the training slmp?es wi(ll a corrcliltioa coefficient of at
least 99U! and Miix. Abs. error of less tliiin 0.04V0 for aII
components except the heptane plus fraction. Thus, the
model iS properly trained and capable of capturing the
physical tri)id-of chaliging ille composition i]l~dproperties of
the plus fraction as \vell ils cun~ulative produced gas during
Cm. ~le GRNNM2 is also able to foreci)st composition of
produced gas. cumulati\e produced gas. and compressibility
of the remaining two philse for ail the depletioli stages of the
testing wampleswith a correliltion cocfficicl]t of at least 95V0
except Cotiposition of ~~ And specific gravity of C7+ (94/0
and 756A,respectively) and a mi~xii])littlabsolute error of less
than 0.3 Y. for all sampIes except the molecular weight of the
C7+ fraction. The prediction accuracy of the Sg7+ could be
improved by corr~atlng the relationship of MW7+ and Sg7+.
Summation of tile mole ffiictions of alI component of
produced gas predicted bj GW2
is checked for training
and testing samples at i~lldepletion stages aad found to equid
to one. Thus the proposed model is :lccurilte. vaIid and
reliable. The model can be used to forecast CVD datil needed
for many resen~oir caIcuI:itions in case this dilt~] is
unavailable, G_2_wou]d
sa\e computer time slid cost
invoIved in predicting gils condensate behavior using EOS.
Once properly trained., the model can be used to cut the
&wnses of frequent the samp] illg and the Ii]borious
exTerin~entalCW-t~st for gas condei~sate reservoirs.

molecular

Gas # 1
Gas #1 is a Iei]il gas condensate containing 94.20/. methane at

dew point. Consti]l]t volume depletion data for this gas is


shown ill Table (6). Figures (10a) through ( 10d) show that
GRNNM2 closely matched the methane content, heptane
plus, two-phase compressibility factor. and cumulative
produced gils during pressure depletion of this gas. These
figures show good agreement between GRNNM2 and the
espcri i~]en[i]Idiitii.
Gas #57

This gas has 28. 16% H2S at dew point pressure of 4190 psi
and 2S0 F. CVD data for this gas is shown in Table (6).
Simuli]tion results of CVD for this gas by EOS as weli as by
GRNNM2 are shown in Figures (1 la) through (1 Id),
Figures (11a) and (1 Ib) indicate that GRNNM2 closely
niatched the experimental changes in H2S and C7+
,respec[ively. better tIu\n the EOS-adj.
Nevertheless.
GR.NNM2 and EOS-adj equally predict two phase
compressibility faclor of the reservoir gas (Ztp) and the
cumulative produced gi~s(Gp).

Gas #58

Compositional analysis of gas # 58 during CVD is shown in


fable (6). This gas contains 12.51 YO nitrogen at dew point.
Simulation results of CVD of this gas by EOS and GRNNM2
are shown in Figure (12a) through ( 12d). Figure (12a) shows
that neither the EOS nor GRNNM2 acceptably matched the
experimentill di]til of measured nitrogen content of the
produced gas. Figure ( 12b) shows that GRNNM2 closely
matched the alcasured C7+ content of the produced gas better
than the EOS. However, Figures ( 12c) and ( 12d) indicate
that the EOS matched the experimental data better than
GRNNM2 for the two philse compressibility and the
cumulative produced gas. This is probably beciluse the
GRNNM2 \viis trained using gas condensate samples \vitll
low l~itrogen colltc]]t. see Table (1). However, gas # 58 has
high nitrogen content. It is important to note that the
frainiilg and testing samples were selected at random and Gas
#58 were not chosen in the training set.

Examples showing the perfori]]i]ncc of GRNNM2 and EOS in


predicting Cm results are presented below for some selcctcd
gas condensate reservoirs, These exalllplcs were choseil to
represent sarr~pTeifronl re-wrvoirs that has the maximum
percentage Of C7+ , Cl. H2S. C02. i~ildN2.
Gas # 39
This gi]s is a rich gas co]~densate that co!lti)ins 12.68% C7+
Experimental results of constant volume
at dew point,

dep[etion for this gas are show11in Table (6). Figures (9i])
ti~rougIl (9d) show compilrisoll betwcetl EOS simulations by
adjusting tile molecular weight of tile C7+ (EOS-adj). EOS
simulatioiiiby regresdon of the EOS pi]ri]met ers (EOS-reg).
and general regression net\vork modeling (GRW2)
in
matching experimel][al]y measured C7+ content of the
. .....- ...-.
produced gas. methane con[ellt of the produced g:ls,
cumuli]tive prodllccd gas (~p). and t\vo phase compressl%ility
factor (Ztp) of the rcscrvolr gas. T}\Io conclusions can be
driiwn from lllcse figures. First. GRNNM2 performed bc[ter
tluu~ the EOS in miltcl]ing the cxpcrinlelltal d~~til. Second,

Gas # 124

This gas contains 63.52 % C02 at dew point, Constant


volulue depletion diiiii for this gas is shown in Table (6).
Figure (13a) through ( 13d) shows comparison between
experiitlcntally measured dala for C02, heptane plus, twophase compressibility factor and cumulative produced gas by
191

6-

ADEL M. ELSHARKAW,

AND SALAH G. FODA

sP~sPg
.-.

References

Gand EOS-reg and EOS-adj. Figure (13a) indicate


tit
tl~e EOS-reg and EOS-adj c[oscly l]liltcl~cd tl~e
eWerimen[al data better tl~i]n GRNNM2. However tile
GRNNM2 SIIQWSbelter agreement wit]] tile experimentally
mensured C7+ content of tl]e prodliccd gas tl~an tl~eEOS-ildj
and EOS-reg, Figtlre 13(b), Simulation results of tile two
pl}ase compressibility factor of resenl~irgas (Figure 13c) and
-dative
produced gas @gure 13d) sl~ow tl]at EOS a!ld
GRNNM2 equally mi]tcl]ed tile expri menti]i dilttl.
,,

3974s

.,.,.=~-

1- Conrard, P. G. and Gri~vier J. F. 1~80. Peng-Robi!~son


Eqliiltion of Stafe Cl]ecks tile V;~lidity of PVT
Experiments Oil & Gas Journal Aprif 2 1,,72T8fi.._
2- Ragl\avan, R and Jones JR, 1996 Depletion Performance
of Gi~sCondensilte Resewoirs, JFT August, 725-731.
. . ...
3- Redlicll, O. and Kwong, J. N, S. 1949. ,OU ftle
thermodynamics of solution V. An equatio~] of state.
Fugacities of gilseotls solution. Cllem. Revie\y , 44, 233244.
_.. ,..
4,- Soilv, G. 1972. Equilibrium consti~nts fro~l~ modi~ed
Rcdlicl~-Kwfongequotion of st:~te. Cl]em. ,Eng. Sci., Vol.
27.1197-1203.
5- Pcng, D. Y. and Robenson, D. B. 1976. ,A new two-r
col~sti]lltequiltion of sti~te. Ind. Eng. cl}em. Fund., VOI.
15, No. 1, 59-64.
,6- Zudkevitcll, D. and, Joeffe, J. 1970. Corr;l~Qon. and
prediction of vilpor liquid equilibria wit]] tl]e RedlicllKwong equiltion of state., AICHE J., May. 16, 496-498.
7- Mi~rtine, 1979, Clibic Equation of state ~~icll? Ind. and
Eng. Cl~en],Fund., May, 81-80.
8- Penelox, A. Freeze P. 1982. A consistent ,correct~on to
Kwong-Soave Volumes. Fluid Pl~ase Equilib., 8.7-23.
9- Saker. R. Danesll, A. S. and Todd A. C. 1991. Pl~ase,
Bellilvior Modeling of Gas Condensi]te Ffuids Using an
Eqtli]tion of Sti\tc. Paper SPE 22714 presented at tl]e
Ann. Tecl]. Conf, & Exl~ibit. Dallas, TX Oct. 6-9,
lo- Droluu, J. K. and Goldtllrope W. H, 1988. 01} tl~e
Quality of Data from Standard Gas doi~densa!e. Pm.
Experiments. SPE 17768 Gas. Tecl~. Symp, Dallas, TX,
June 13-15.
11- A1-Mid~ros.F. M. and Tijoa G. H. 1987. Analysiqa~~d
Pl]ilse Bcl~avior of K]]uff Gas Co~ldensate Systel~lin
Bi]l]r:~in field SPE 15766. Presented at tl~e ~OS,
Bi]llri]in.Milrcll 7-10.
12- Firoozbi]di. A.. Hekim Y. and Katz ~. L,, 1978.
Reservoir Depiction Ci~lculations for Gas Condensate
Usiag Extended Ani]lysis in tl]e Peng Robinson Equatio!~
of State, J. Canal. Cl~em. Eng, Vol. 56, Oct. , pp. 610615.
13- Besserer. G, J. and Serril, W., 1979. An Eff]ci,e!~tPl~ase
Bel~vior P;~cki]gefor Use in Con}positioniil- Reservoir
Simu]i]tion Sfudies, SPE 8288. Ann. Tecl~. COnf. &
Exl~ib.Los Vcg:ls, Nevada, Sept. 23-26.
14- Willii]nls C. A., and Zana, E, N. 1980. Use ok tl~ Pel~gRobinson equi]tion of state to predict Ilydroci]rbon pl]ase
bcl~i]vior and nliscibility for fluid displacement: Paper
SPE 8817. Elllli]nce Oil Recovery Sym. T(dsi~,OK, April,
20-23.
,.
,,
15- Vogel, J. L., a;~d Yarborougll. L. 1980,.Effect of ~ljtrogen
On tl~epflitse bellilvior and properties of reservotr fluids.
SPE pilpcr 8815. Enl]i]nced oil Recovery Synlp.,TTillsil,
Ok, April 20-23.
.,
16- Wl~itson. C. H. and Torp S. B. 1983. E\;:dtI;t ing

8ummaq

TWOscl~emes of characterizing tl~e plus fri]ction lli~~cbeen


used in tItis study to caIculirIe tl~c dc\v poini pressure by
equation of-stn~e for gi~s coi~det~satc rcsenoirs, Tl~e study
indicates tlmt tl~e optional cl~aracterization has a little
improvement over tl~e standard cltari]cterizat ion. Several
examples of simulating CVD for g:is condensate reservoirs
using tile molecular weiglltor IIle plus frilctioll ilSa llliltCllillg
paramete~for tl~edew point pressure snowed better agreement
befween EOS rmd experimenti]l data tl~an when tl}e EOS
parametem w~m re~es~d to mittcl~tllc dew point pressure.
.Tl~e study aho presents two gcl]cri~l network models. Tile
tirst model. G~
1 uses iltitiai conlposition of ~i]s
condensate
to predict dc\v point pressure and gils
compressibility filctor ilt dew point. Tile secol~d model.
G~2
uses dew point pressure. gas compressibility at
dew point, reservoir temperature, ill]d reservoir pressure (s) to
predict composition of produced gas, cumulative production
of produced gas, and t\io pl]i]se c,)mpressibiiit y factor of
ren~iining ~ during pressure depiction of gas condensate
reservok. Tl~e models were trail~ed and tested using CVD
dafa of wide range of gi]s condensate samples collected
worId\\tide, Accuri~cy of dle proposed models were cl~ecked
by comparing tl~eirpredicted perform;lnce for Yi]riousgi]scs to
e.l~rinlenfaIly measured CVD and simuli]tion results by
equatiol]-of state. Tfle study sflo\vs tlli]t ~l~csemodels ilre
properly tri]ined, ilild did succcssfutly capture tile physical
trend of cl~angiug gas condcnsn[e properties during pressure
depfetion. Se\~erillexi~mplcs arc pres:ntcd to demonstrate tl]e
accuracy and reIiilbility OF tl]e proposed GRNN models.
~ese GRNN models call be used to forecast CVD t]eedcd for
reservoir and production engil]eeri]ig calculations SUCI1as
matetial balance calculiltions, reservoir simulations, separi]tor
design, and verfical ptiorn~ilnce calculiltions, Tl}e models
catI aIso & used to cllcck accuracy of CVD tests and
redntime and money involved ill simula(iolts by EOS and
laborious and expensive sampling.
Acknowledgments

Tile autl~ors \\ouId like to d~illk tl~e Ku\}ilit Folmdi~{iollfor


tile Ad~ilnceme\lt of Sciences (KFAS ) for providing fin;] nciil]
SUpporf for tl~isstudy. research gri]nt N(),96-09-04.

192

~..
<
SPE SP~ ~~~=

EOS SIMULA~lON

AN[) GRNN MODELING OF THE CVD BEHAVIOR OF RETROGRADE-GAS

CONDENSATE

RESERVO~S

.
...
3-6.
33- AreI~arr. R, A. . 1989. Drill bit diag~~osis usi]~g aeural
rlet\vorks. 1989 CoIti. ol~ Artificial lii/elligeace ia
Explori]tiol~illld Production. Texas A&M.
34- Eppit~g. W. J. aad Nittcrs, G. 199[~Neuri~l l~etworkfor
it]]alysis aud ialprove]]]ellt of gas wef~ productio]l.
Coalputcr Sialulatioa Coaf.. Calgary, July 16-18, 791796.
35- Brio[les. et iii.. 1994. Applicatio[l of Ileuiirl aefw;ork ia
tl~c prcdictioa of reservoir llydrocarbol] afixture
coa}posilioa frol~l production data. Paper SPE 28598,
69tl~AI~i].Tech. Mcel., Netv Orleaas. Sept. 25-28.
36- Hi]bibi]llill).et i]l. 1996, Use of aeurill aetworks for tl]e
predictiol~ of vapor/liquid equi]ibriual K Values for Iigllt
l~ydrocarboa a~istures. SPE Res. Eag., May, 12I- IZ5.
37- Gl~i\rbi. R. B. :Illd Elsllilrka\W, A. M. 1996. Netiral
uetwork l~lodcl for eslilli:ltiag tile PVT properties of
Middle E;~st cnlde oil systeals. lU Situ. Vol. 20. No. 4,
.
367-394.
38- Gllarbi. R. B, attd Elsllarkaw, A. M. 1997. Uaiversal
l~euri]l act\vork bilsed {t~odel for est iIlultiag tile PVT
properties of crude oil systeuls. pi~per SPE 38099.- Asia
Pacific COI~f.,Klltllil Lull~pur. Mali~~sii].April i4- [6.
39- A1-kili]bi.A. W, al~d Lee, J. W. 1990. All artiliciid ueural
act\\ork approi~cl~to idcati~ tl]e well test iaterpretatiou
a)odel: i]pplicatiol~. paper SPE 2055, 65tIl Aaa. tecl].
sleet,. Nc\v Orleaas. LA. Sept. 23-26.
40- Jui\iilrdi. 1. R. i]jld Erslulgl~i. 1, 1993. Cou~plcxities of
usiag aeural l)c(work ia well test ~l~i~lysisof filulted
reservoir. Paper SPE 26106. West Reg. Meet.. AIaska.
May 26-28.
41- Kutllol~iyi et at, 1994. Ideatificatioa of well test l]~odels
usil]g lligl~order aeural I}et.paper SPE 275S8, European
Colllp. Coafl. Abcrdeea. 15-17 Marcl~.
42- Suag et al. 1996 developaleat of HT-BP acurai aetwork
Systeal for lllc ideatificatioa of well test iaterpretatioa
alodel SPE Coalp. Appl., August, 102-105.
43- Accariiia. P. aad Desbraades, R. 1993, Nuro-coalpu[iag
l~clppore pressure deteraliaatioa. Pet. Eag. Iat. Feb. 3942.
44- Nietlltscl)ik, G. E.. Poetttuaaa, F, H,, aad TlloIIlpsoiI R.
S. 1994. New Correlation Deteril)iues Retrograde GasCoadel~sate Coalpositiotl. Oil aud Gas J. Oct. 31.33-38.
4s- PedcrsoIl K. S., Tl)oi]litsoll P.. aad Fredeasluad A. 19%2.
Pl)ilse equilibriii aud separation process. Report SEP
8207. Iast. Kel)litckllik. Del)a~ark Teraiske Hojskole.
46- Cavett. R. H. 1964. P1~ysical di~ta for distillation
ealculatioas. Vi]por-liquid equilibria]. 27tl~ Midyear
Meetiag, API Divisioll of refiui]lg. Si\l~ Fraacisco. ea.
Mi~y1S. pp. 351-366,
47- Tlloa]assca . P.. Pcdcrsea, K. S. aad Fredeasl[lad. A. A.
1986. Adjustlllcllt of C7+-a]olccular weigl~ls ia tllc
cll:lrilcteriziltion of petroleuill alixtures coataiait~g IleilYy
I]ydrocarbous.
SEP Report 8617. Iustittfie for

Col~dcl]si]{cVolultle Depletiol\ Dati]. JPT Marcl~, 61t)620.


Sill)t]li](io]]
of G:)s COIIdCIIS:ItC
17- Coals K. H. 1985.
Rescrvoir Pcrfonl}iilicc. JPT Ott. 187[)-1886. (l 982
SPE 10512),
18- Coats K. H., iilid S[iiiirt G. T. 1986. Applici]tioit of iI
Regression-Bi{sed EOS PVT Prograll~ to Lilboratory
Dati]. (1982 SPE 11197j SPE Rcscnoir Eag, May. 277299.
19- Douglt]s. E. K., itild BoI1lc. G. A, 19%7, Tl\ird SPE
Colupi~rz][iveSolutio]] Project: Gits Cycling of Retrograde
C!oadellsate Reservoirs. JPT Augtlst 98 I-997.
20- Drolull J. K.. Goldtllropc. W H,, al]d Trcl~go\/e R. 1988.
El]l]a]~ci]]g tile Ev~du:]tioa of PVT Di]tii, OSEA 88174,
7tl) OflsI~ore SOUIII Ei]st Asiil. Feb., 1988.626-635.
21- Be[lc, S. and Hcil]cl\~i\I~.R. F, 1989. Coll]positiollill
Modeliag of Higll Teatpcri]lure Gas Col~deasi]te
Reservoir !vitl~ Water Vaporizat ioa. SPE 18422. SPE
SyIl]p. 00 Res. Sill]ulittioll.. HOUSIOI1, TX Feb. 6-8.
22- Cll:~bak, J. J. al]d Willit~ll}s.M, L.. 19(J2. P-X Bcl\i]~ior
of Ricl~Gils Coi]dcl]silte ill ildl]lislurc will~ C02 aild (N2
+ C02). SPE 24132 prcsclttcd ill tl~eEOR Tlllsi]. pp. 2224.
23- FurltiuoI. S. R. 1993.
Successful Prcdictioa of
ColldcIlsi~ieWCll bore Bcllil~ior [ Isillg EOS Gctlcri}l frolll
Black Oil Data. SPE 26683. Prcscl}icd at OtTsl\ore
Europeaa Coltf,, Abcrdccl]. ScpI, 7-10,
H}droci]rbol] PIUS
24- Wllilsoll, C. H, 1983. CI)i}ractcrizil]g
Fractioa JPT AugusI, 683-694,
25- Katz. D. L., Firoozi]badi A, 1978, Prcdictiag pl]:]se
beiltivior of co]~dellsa[efcnlde oil-systeltls usiug [tletl]itlte
intemctioa coe~ciellts, 3PT. No\ cll~bcr. 1649-1655,
26- Peaelox C.. Ji~il\E.. 1979. Appl ication of Ille de\rcloped
RedIicll-K\vol]g Eqlli]tioll of State to Predict tf~e
Tlleralodyaaalic Properties of (~olldcllsi][e Gases. SPE
8287. SPE AIu~.Fall. Meet, Los Vegas,
27- Pedersoa K. S.. TlloII~asou P., al}d Frcdcllsluad A, 1985.
Ttlcralodyualllics or PctroIctlll~ Mixtures Coati~iail]g
Havy Hydrocarbons, Iud. EI~g. Cllea~. Des. Dev. Vol.
24, No. 4, 948-954.
28- Wl~itsou, C. H. 1984. Effect of C7+ Properties oa
Equatiou of State Prediction SPEJ, Decefttber, 685-696.
29- Ali, J. K. 1994, Neural uctworks: A aew tool for tl]e
pctroleual illdust v?. pi]per SPE 27561. E~lropcilli Pcr
Ccraf,. Aberdcc]l, 15-18 Milrcl~,
30- Osborae. O. A. 1992. Ncurid i]ctworks provide Il]ore
accurate rescnoir pcrll]ci)bili[y, Oil & gils J.. Scpt, 28.
80-83.
1996. Pctrolculll
31- Mol~gIlegl~, ct itl.
reservoir
cllariicterifiltioll wiill tllc aid of artificial lleUrill
actworks. J. Pet. Sci. Eag,. Vol. 16. 26-27.+,
32- Zl~ou et al. 1993 dctcrulil~il~g rcscmoir properties ia
mscrvoir studies usil]g a fuzzy ]~c~iri]lact~~ork. paper
SPE 26430, 68111Aaa, Tecl~. Meet,, HOUSIOI1,
TX, Oct.
193

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE SPE 3974S

Kelllitekl)ik, tl~eTccl~llic:lIU1li\ersity of Delllll:~rk.


48- Kesler. M. G. :l!~dLee, B. 1. 1976. llllpro~ed predic[ioll
of fllll}i]lpy of friictiolls., Hydrt~c:lrbol] Processing ,SS,
153.
49- Pdrscll. K. S., Tllol;l;lsscll, P. ;Illd Fredel~sl~llld,A. A.
1989. Ad\tll~cesill tl]erllldyl~[lil~ics. 1, 137.
50- ElsI]:~rki~}}~.
A, M. al~d Fodil, S. G. EOS siillld:]tiol~;ltld
G~
[~]odclillgof tl]e col)stilll( VOILIIIIC bclli]~ior of g:IS
colldelwite rescnoirs, [o :lppc,lr ill E1ler~ & FI;cI,
Mlrcll issl]e. 1098.

-.77

.-.
,

,..

..,.

.:

,.
..
. . .. . .

,1-,

4...

.:
..

.
..

.;.
.:
,-

-,

., .
.,.

. .,

..

..
.,

... ,,

,,

.
,,

,,.

.,.:..,

J,

194

,,,

.,. -

SPE SFE 3976

EOSSIMUWTION

AND GRNNMODELING

OF THE CVDBEHAVIOR

OF RETROGRADE-GAS

CONDENSATE

RESERVtilRS

Table 1- Description of tl]e constant vol~lnleda[:l lised in tl~isst~ldy.


142
No. of san~ples 1082
No. of depletion stags

w
(3

(2

[1 (II(Q

Mw

1(4

I
II(KI17
(1I 2W
II lJh38
(),(13%! [1(KIX2

(1,(m)??

{) ( UK,

(12(?)2
()(1775

T
(.!7
+

0.1719
(! 0279

98
253
127

SC1
C7+

Ztp

()58

{J85
77

Table 2- Binary interaclio[l coefficient


H?<
().()()

N2

C07

N7

0.()()

().()()

co?

(),()()

().()(1

,012

H7S
c1

0,()()
().()2

().12
().10

c?

0,06

(). 10

().()()
().()8
().()7
0.07
().()6
(),()6
().()6
().()6
(),()5

c;-

cd

().()8

i C~
n c~
c~

0.08
0,08
(1,08

10
0.10
(), 10
().10
().1()
().10

C7+

().()8

(). 12

().()8

i CA

().()8

().

().[)5

Titblc (3) Error itl cstilllat ing tl~edc\v point prcssllre by EOS and GRNNM 1

EOS
St:]t]dilrd
cl}tlrtlcteriz~ltioll
Optional
Cllaracierization

1525

(),()65

53.079

1753

().228

41.336

-6.427

1.1%2

10.409

94.783

7,858

95,067
4

Am

Trai]~ingd~ti]
Test ing data

93
723

I ().()()()
().121

1.852
37.71

(),()24 I (),27 1
/ -3.964
6.932

99999
98.512

Table (4)error inestin~ating Zattl~ede\v point by EOSand GRNNMl


Er Yo

Ea y.

R y.

] 5.374
5.353

91.43
85.04

Mas Devia[ion

EMill Y.

EMZIX7.

Standard cll~lr:]c[criziltioll 1

(),364

1 ().1()3I

.2792

Optional Clulrilcteriultion

().674

().()98

44.722
94.134

().0I
().()3

I ().()()() I
().()64

(),()~(j
6.782

-().()()5 I ().133 I 99.99


-3.()()() 1.968 I 98.88

EOS

-().335

ANN

Trainiilg data
Testing data

i 95

10

ADEL M. ELSHARKAWY

T
I

c?

~ (),()()1 I
.. ...

.-

11-c~

[),()()()

().()16

99,9

().()15

~ 99.5

I 0.()()2 !

[).(102

I ().()()()

().()()5 I 99.5
()()7
(1.()(12

i-C<

().()()()

I (),()()()
I
. . .. .. ..

(1.()()2

99.8

[).()())

(),6()7

().()()1
(),()()5
().()13

().()19 1 98,()
1

().()( )-1

().()16I
(),()()5
I
I

I ().(1()1 I

1().()()()

90+5

11.147

Z,n
G:

().

11-c$

WC7+
SG,.7+

Tcstil]g si]l~lPlcs

I ().()(1()

(),()()()

~)!).1

I ().()()() I
(1.()() 1

().()()9

079

93.5

().()27

63 ~)

().()73

894

98,8

(),()()4

ggeg

~,glg

53.625

90.9

([).022
[[),077
----

99.8

().0()5

0.138

754

99,8

/ ().()3() !

().276

92.5

1 99.4

I (),()4!) I

(),2%()

97.4

95.5 -1
95.9
1

7.474

(1,422

spE SPE 3974s


.=

..

,:

Triiitlitlg siill]Ples

co?

l-c

AND SALAH G. FODA

1
I

.:

,-

..
..

.,,.- -j

.-

$
.

,.~

_:.

.-,.. ...

,.

....

:.

.
.!

SPE SPE 3974S

EOS SIMU~TION

AND GRNN MODELING

OF THE CVD BEHAVIOR OF RETROGRADE-GAS

CONDENSATE

RESERVOIRS

_.
_

0000000

0000000

11

..,

:----
SPE ~P~3974S
-.

ADEL M. ELSHARKAWY AND SALAH G, FODA

12

,,

kput layer
(15 Neurons)
~
.

2s

1
2
2
1
2
3

C4
C4
C5
C5
6
7+
7+
\{c

Hidden Layer
(180 Neurons)

Output Layer
(2 Neurons)

Pattern
And
Sununation
Layer

C7+

Figure I- ,hchitecture

of GRNNM 1

,.

. ..

~
,-

>.

-,

.-

:,

,=

Output Layer
(2 Neurons)

Hldti

Layer

H2S

Hdden Layer
(1500 Neurons)

C02

N2
cl
C2
C3
IC4

(2 Neurons)
Pd
Zd

P
T

Pattern
And
Sununation
Layer

NC4
IC5
NC5
C6
C7+
MWC7+

C7+

M
P

Figure 2- Architecture of G~M2


198

..

f
EOSSIMU~TION

SPE SPE 3974S

AND GRNNMODELING

OF THE CVDBEHAVIOR

OF RETROGRADE-GAS

CONDENSATE

Reservoirs

13

,
4000
6000
8000
1~
Measured dewpoint pressure, psi

Om

0,8
1.0
Experimental Z@ ;ewPoi;t$ressurl-6

0.6

12000

Fig6. Crossplot of Z @dewpoint


Characterka!ion)

Fig
. .3. CrosspIot of dew pint pressure ( Standard
Charactetiation)

1.8-

(Standard

1.8

12000

9
!

2000
t

m4000m

8000

10000

120Q0

0.8
1.0
E#rimental

1.6
1.4
1.2
Z @ Dew Point Pressure

-_

1..
1.8

Measured dewpoint pressure, psi


Fig, 7 crossplot of Z @ dew pint (Optional

Fig. 4 Crossplot of dew point pressure (Optional


Characterization)

Characterization)

= test data

test data

. Tm data

0200040008000
Measured

o Tm data

8000

dew point pressure,

10000

12000

psi

0.6

Fig. 5 Cross plot of dew point pressure, GRNN model

199

0.8

1.0
1.2
i.4
Measured Z @ dew point

Fig.8 Cross plot of Z

1.6

@ dew point, GRNN mdel

1.8

..

--14

ADELM. ELSHARKAWYAND SALAHG. FODA

0.14

0.78,

0.12

0.?6

0.10

0.74

SPE SPE 39745

0.72
0.06

0.70

~p

0.04

0.68

-O-

Eos-reg

0.02

--0.66 1%

Eos-sdj

-sk-GRNNm2

0.64-1

Figure9b
1

ZcrJo

Pressur~ psi

400D

ti

Pressure, psi
1.4,

0.80

0.70
0.60

\a

0s0

,.:~

0.30
0.20
0,10

0.00
o

,1
4000

2000

a
~ 0.6

0.40

+s
Figure 9C

-4-*
+

0,4

Eos-reg
Eos-edj
+GRNNM2

Figure 9d

0.2 ,
0.0

4000

2000

6000

8CU0
-

Pressure, psi

Pressure, psi
Figure 9- Simulation of CVD for gas #39

+
m--==- .

0.s2

Exp
+GRNNM2
~--::y
-Reg

0.80

EXP
+GRNNM2
EOS-adj
EOS-Reg
~:+

0.74

Figure 10b

Figure 10a
o

1000

3000

sooo

Pressure, psi

Pressure,psi

1.0

1.0
0s
0.6

0.9

a 0.7
s
0.6
1
E

0.5
0.4
03
0.7

0.2
0.1
0.0
0

Im

2GIM

soOO

0.6
o

1000

Z.(nm

3rxro

Pressurq psi

Pressure,psi

Figure

10- Simulation of CVD

200

for gas #l

8PESPE39T4S

EOSSIMUMTION ANOGRNN MOOELINGOF THE CWBEHAWOR OF RETROGRAOE-GASCONDENSATE Reservoirs

15

-.. .

Figure Ila

Pressurepsi

Pressure.

psi
-..

-:---

0.2

1XSTNM2

Figure lld

0.04
Pressure,

..-:

psi

Pressure,

EOS-Reg

-. --

psi

.4,

:.,
Figure 11- Simulation of CVD for gas 57

._fi_..

0.04,

.102,1
n

.. .

I
001

am
Pressure,psi
1.1 -

0.90
0.s0

a
m0.70

!
I

1.1

0,60

N
$

0.s

2~

= -.=-.

EOS-sdl
+GRNNM2
-0EOS-ad

1.0

0.40
0.30

Figure 12c

1.0

o.2f!
O.lc

0.9

O.m
Pressure.

Pressure,

psi

Fqure

12- Simulation of CVDfor gas 58

201

psi

SPE SPE W74S

ADEL M. ELSHARKAWY AND SALAH G. FODA

16

,.

.
0.06.
<
.

z+

0.56

~-~

Figure

0s2
0.s0

31
+Exp
-O-

EOS-Rq
--EOS+dj
+GRNNM2

13a
-~

.-:3

0.0s -0.04 --

Exp

Eos-Reg
Eosadj

-.:

*GRNNM2

g 0.03 -m
G
I
0.02 -z
0.01 --

figure

13b

O.oi)4

Pressure, psi
.o-~

Pressure,

--

0.60

0.50

0.40

0.30

0.20

0.10

. .
-- .

0.9U -

0.60

:
m
%
g

psi

Figure 13 c
0.70

+
Exp
~
EOS-Reg
~
EOS-adj
+GRNNM2
0.80 :-- nk
0.65

a
:
G
~

O.m L~.

+
Exp
*
EOS-Reg
~
EOS-adj
+GRNNM2
a

1000

0.75 -0,70
0.65- -

2000

3000

/
Figure 13d

0,60 -I

5W

1OQo

2000

3oci)

Pressure, psi
Pressure,

Figure

psi

13- Simulation of CVD for gas # 124

..
/

202