Application of a Regression-Based EOS PVT Program to Laboratory Data K.H. Coats, SPE, Scientific Software-Intercomp G.T. Smart," SPE, Scientific Software-Intercomp Summary. An equation-of sate (EOS)-based PVT program was applied to match laboratory PVT data for three published and nine additional reservoir fluid samples. This paper includes laboratory test data forthe nine samples and deseribes PVT program features, especially regression, that we find conducive t0 rapid determination of EOS parameter values needed to match data. With regression, both the Peng-Robinson (PR) and Zudkevitch-loffe Redlich-Kwong (ZIRK) EOS give comparable and generally good agreement with labora- tory data, Without regression or significant adjustment of EOS parameters, neither EOS adequately predicts ob- served reservoir fluid PYT behavior ‘Our EOS tuning uses a small degree of C>.. fraction splitting. The agreement of these EOS results with data ‘compares favorably with that obtained in previously published studies that used extensive C7 splitting, Introduction ‘A recent trend in compositional simulation is the use of an EOS, as opposed to independent correlations, to cal- culate K-values and equlibrium-phase properties. An im- portant prerequisite in meaningful use of the EOS-based compositional mode! is satisfactory agreement between EOS results and laboratory PVT test data relevant to the reservoit fluid and recovery process. ‘A number of studies!” report comparisons of cubic BOS and laboratory PVT results for a wide variety of reservoir fluids and conditions. Most of these studies om phasize the C;.. characterization as the key element in attaining agreement between EOS and laboratory results Some studies use more than 40 components that result from spliting the C7. fraction. Some authors imply a predictive EOS capability provided one EOS parameter $sadjusted to match the reservoir fluid sauration pressure “The work reported here reflects our experience tht the EOS is generally not predictive and extensive spiting of the Cr. faction to match laboratory data is generally un- necessary. We indicate that more of the available labora- tory data than were frequently used (oF reported) in past studies should be used in evaluating and tuning an EOS. ‘The reservoir uid studies presented illustrate the capa bility and efficiency of multivariable, nonlinear regres- sion in seeking agreement between EOS and observed PVT results ‘We do not dismiss “proper Cp characterization as a necessary element in tuning an EOS. Rather, we sup- porta philosophy of minimal spliting followed by adjust- tment, using regression, ofthe heaviest (plus) fraction’s two EOS parameters, generally denoted by 03 and 0° We describe regression-based PVT program features that we feel contribute to time-efficient tuning ofan EOS which is necessary before its use in feld-scale simula tion, Laboratory data given for six oil and three retro- grade gas condensate samples include reservoir Copyh 806 Set of Posnum Engines [SPE Reservoir Enginceing, May 1985 temperature expansions, surface separations, N> reser- Voir fluid behavior, and one set of multiple-contact data, Results are presented for three additional fluids with data reported in the literature, Generalizations regarding the regression procedure and results, based on these 12 fluid systems and a larger number of unreported fluid studies, are stated where possible or warranted. Description of the PVT Program ‘The PVT program isa general-purpose program that uses 2 generalized cubic EOS" to perform phase-equilibrium and property calculations. The generalized EOS reduces to any of the Redlich-Kwong (RK),"! Soave-Redlich- Kwong (SRK), " ZRK, "and PRIS EOS. The pro- ram may be used to calculate fluid behavior solely on the basis ofthe predictive capabilities of any of these equa- tions. More important, however, is the capability to use 4 nonlinear regression calculation that performs an auto- matic adjustment of EOS parameters to match a variety of laboratory PVT measurements. The resulting tuned EOS is then used in a compositional reservoir simulator. ‘The first step in use of the PVT program is to define the components that comprise the uid system. The pro- gram contains an internal table of properties for CO>, No, HoS, CO, Hp, SO», O2, and pure hydrocarbon components from C) through Co. Internally stored bi- nary interaction coeficents closely resemble values given by Yarborough? for the RK EOS and by Katz etal! for the PR EOS, Properties for user components not contained inthis internal table are either entered by the user or deter- ‘mined by interpolation on the basis of molecular weight. The program also provides the option to spit the plus fraction of a sample into a number of extended fractions. The internally stored properties of extended fractions and the method of spiting are those presented by Whitson. In addition to his preservation of molecular weight and role fraction of the origina plus fraction, we added a ific gravity of the plus frac- mtion. The molar distribution of the single-carbon-number sroups in the plus fraction through Cao is first deter- mined. A grouping ofthese single-carbon-number groups into fewer multiple-carbon-number groups then completes the spliting procedure. Inthe predictive mode, the program can perform a num- ber of calculations on the basis of the current fuid-system definition as determined by the EOS parameters. For ex- ample, these calculations may be performed before and after a regression to compare the EOS-predicted perform. ance with the tuned EOS performance. In addition, fol- lowing a regression to match data for one or more samples, a prediction of results for one or more different samples may be performed. The calculations available in the PVT program include: (1) saturation pressure and cquilibrium-phase properties for a given composition and temperature; 2) density and viscosity calculation for spe- cified pressure, temperature, and composition; (3) constant-composition, constant-volume, and differential expansions for specified sets of pressure levels; (4) single- ‘or multistage flash separation tests; (5) phase-envelope calculations for swelling tests; and (6) pseudoization umping) to fewer components. The program uses the Lohrenz et al viscosity correlation © with automatic tun- ing to match experimental viscosity data. “The data to be matched inthe nonlinear regression con- sist of laboratory measurements for one or more fluid sam- ples that may be at the same or different temperatures. Fluid samples from a swelling test that correspond 0 different mixtures of reservoir fluid and injected gas may also be included. For each sample, the following data may bbe entered: (1) saturation pressure; (2) densities of oil (gas) and associated gas (liquid) at saturation pressure; G) K-values at saturation pressure; (4) constant- composition expansion data including relative volume, ‘volume fraction liquid, and gas and liquid gravities; (3) constant-volume expansion data including volume frac tion liquid, cumulative gas removed, gas z factor, and oil and gas gravites; (6) differential expansion data includ- ing oil FVF, solution gas R,, z factor, and oil and gas sgravities; (7) K-values for any or all of the pressures in any of the expansions; (8) multistage separation data in- a8 ‘TABLE 1—FLUID COMPOSITIONS AND PROPERTIES AT RESERVOIR CONDITIONS Gasa" Gas2"* Gas4 GasS O11 O12 —_O9S_—_ONA OS ONT co, 0.0068" 0.0061 0.0350 0.0217 0.0044 G.0090 601 0.0285 0.0103 0.0008 Ne 0.002 C014 0.0084 0.0045 0.0080 0.0088 0.0081 0.0055 O.0T6A H3S 0.0008 0.0004 0.1810, C, 05832 0.5749 05762 0.7084 0.3505 05347 0.0705 0.9521 0.9847 0.2840 G2 0.1355 0.1345 0.0739 0.1076 0.044 0.1748 00157 0.0872 0.0093 O.O716 C5 0.0761 0.0752 0.0802 0.0494 0.0248 0.0873 0.0908 0.0824 0.0885 0.1048 CZ 0.0403 0415 0.0231 0.0802 0.016 | 0.0456 0.0331 0.0507 0.0600 0.0840 GS o0281 0.0233 oora, ora O16 0.0209 0282 O0s23 OUN7B CORRE G{ 0.0190, 0.0179, 0.0554" 0.0000, 0.0546, 0.0151, 0.0258 0.0410, 0.0356 0.0405, GS 0.1145* 0.1220" 0.0588" 0.4828" 0.1682" 0.0216 0.3497" 0.9049" 0.3507" a 0226 c o.o2r0 Cio oi90t Me 4991937532057 2913002 y+ 0.8135 0.8115 0.7748 0.8100 0.9000 0.8364 08570 C8405 0.8366 0.8829 DP, 44804415 3,960 4.842 25204480 «2507547748 1.604 be 2885 28.54 1915 4795 3301 44.17 40.36 SOT a8 “1901902028571 178179250 28k AST ‘cluding GOR, oil and gas densities, and K-values for each stage; and (9) swelling-test saturation pressure and volu- metric data In some cases, as shown by Hoffman et al.,"7 the available laboratory data for an oil sample include expan- sion data for the associated gas phase. The program al- lows these gas-expansion data to be entered in the cil-sample regression data, In addition, the capability to calculate an exact match of the density of a pure compo- nent ata specified pressure and temperature is provided For example, if injection of pure CO or No of methane ‘were anticipated in the reservoir, the density of that in- jected gas could be preserved within the context of a si- ‘multaneous match of all laboratory data for the fluid system. The set of all observed data for the regression calculation is denoted by {d;}, j=1,2.. ny The regression variables are user-specified and may be any subset of the EOS parameters. These parameters are 5; and 2; for cach of the n components and the n(a~1)/2 binary interaction coefficients. In addition, the program allows the definition of a single regression vari- able to represent the average of a range of EOS parame- ters, This feature is useful when matching data fora fluid system that has an extended analysis. Instead of inelud- ing a regression variable for each 0; of the extended fraction components, a variable can be defined that rep- resents the 93; of a group of the heavy components. This results in fewer regression variables bu still allows each heavy component to contribute to the parameter adjust- ‘ment process. ‘The regression isa nonlinear programming calculation that places global upper and lower limits on each regres- sion variable vj. The user may overread the program default limits to ensure thatthe variables are allowed to take on only those values that he considers to be physi- cally reasonable. Subject to these limits, the regression determines values of {v,} that minimize the objective function F defined as Fe & W, | dy-dyend, |. ay SSPE Reservoir Engineering, May 1986‘TABLE 2— EXPANSION AND SEPARATION DATA FOR GAS 2 I CE at 10°F COE at 10°F tor Bewport Semple | 2 i. t, ho wv, 6) oto wy, oH) Esco" O50 ase" 9505 5i00 | os00r S00 8589 eam Deer So Oarer Sow eres seo Oa 500 ane eco Same ‘io oss isco aera ‘S00 ose Sieor+ T9000 090 Gite 9000 10000 Sata" o00e 8 cio too "tae Sag0 1808 e738 fico tuo Gere © Soe0t0ar Soe 2300 noce Gass haan toned Stee “Se ow Goa fe00 Yoosa stae {> tte Sera fie torer sts ‘Ser oss Ses S00 Youre Shae “i tere Gran Sye0 tones Soar Seer tosses Seco Suso joa eras Seer ou Sea9 eee ane 4296 Simo toot Sam bso taree 3675 Sues tooo Sete 2000 ena 2Ba8 Sow tiers Srzr tes 2aare 2020, Bor tae aaa ase Sere 08 zor isons Sase or Geese aoe Separator Tet or Dewpoint Sample Sock Tank Specie | pT separ stcktank “Gauly vay of cof) (er Bon” SSSR eamratcore) Separator Gas 138 70 aS 165 a8 arr? | CConstantVolume Expansion at 190° F for Dewpoint Sample seni Pressure Component Faso" 3800-2700 ~«—«1900~—~1100~—~—«SO on CO, +H~S 0.0073 0007S 0.0073 0.0073 0.0075 0.0081 000m ‘,*” 05892 osex2 0.8875 0.7201 O7S41 0.7190 0.6599 C0955 C1955 0.1948 0.1359 0.1389 0.1502 0.1720 © —«Go7ét 10761 010695 0.0544 010633 0.0708 0.0885 Ci 0.0404 0.0408 10382 0.0202 0a © 9.0285 0.0388, CS 0024s O02 012 014d GOITT 0113 0.0158 Gi 0.0190 0.0190 0.0131 0.0079 0.0062 0.0049 0.0069 Ce Ontas 0.1148 0.0377 00212 00111 0.0076 0.0081 Ms Yeo 193) tae 19 z 1.1889 0.9962 0.8402 0.7066 0.8140 0.8608 0.9108 1.0000, 6 0 0,09589 022551 os8165 0.58225 0.72743 0.87957 0 «00823 0.4940 0.4533 0.4051 0.9682 0.9097 (9519) ‘here dic and dj are calculated and observed values of “observation j, respectively. The terms W; are weight fac tors with internally set default or user-dverread values ‘The default factors are 1.0 with the exceptions of values cof 40 and 20 for saturation pressure and density, respec tively, at reservoir temperature. If several samples are jn a data set, each with saturation pressure and density, then 40 and 20 are used for the first sample and weight factors of 12 and 2 are used for subsequent samples. ‘The theoretical values of .,°, 0° for the PR and RK EOS are roughly 0.4572, 0.0778 and 0.4275, 0.0866, respectively. The default lower and upper regression limits are (0.1, 1.3) for M3; and (0.02, 0.25) for @f,. The default limits on binary interaction coefficients are (1.0, 0.9), These extremely wide limits are rarely approached SPE Reservoir Engineering, May 1986 in applications and the interpretation of any such approach is discussed in the Appendix. The Appendix also discusses the particular EOS parameters we normaly select as regression variables and the justification of thei selection. “The nonlinear programming technique is basically an extension ofthe least squares, linear programming (LSLP) method. '® At each iteration of the regression, a local Subregion of the global parumeter space is defined by {(1£0.03)»;} where v; are last-iterate EOS parameter Valdes. Linearity between {djc} and {vj} values is assumed in this small cubic subregion and the LSLP cal- ulation is performed to calculate new iterate ¥, values in this region. If any ofthe new iterate values le on a toundary of the subregion, then anew subregion (1:20.03); is defined using the new iterate, valves and 29‘TABLE 3—EXPANSION AND SEPARATION DATA FOR GAS 4 GE of Reservoir CCE of 20.40% Ns Somers CCE of aH N, Gas at 2 Mix at 240°F Mix at 240°F “ 2 i Pp i P ‘i {esto VV, 9%) (psig) _—_WV,_ (98) tig) WIV, _(96)_ 5500 0.7600 5800 0.7910 5500 0.8230 5,000 0.7901, 5.000 o.8271 5000 0.8611 4500 0.8328 4500 0.8737 4500 0.9128 4000 0.8897 000 0.9355 4200 0.9518 3900 0.9031 3,900 0.9503, #100 0.9669 3.800 0.9180 3.200 0.9867 #000, 9822 3700 0.9999 3,700, 0.9840, 3.595" 1.0000 3,600 0.9508 3,608" 1.0000 3700 1.0366 0.58 3500 0.9670 3,400 1.0845 0:99 31800 1.0807 1.10 3.360" 1.0000 3000 3.1577 123 3000 1.2014 290 3.200 020 2500 13772 392 2500 1.8725 3.43 3.000 1.0906 0.63 2500 1.2048 461 CE of 30.9% Ns Mix at 240°F of Reservoir Gas. a 7, 2 7 T Pp (osig) Viv, 8) (sia) een, (si) 5500 0.830 5500 73 2.505 000 0.9814 51800, 108 2773 41300 0.9681 5,250" 173 3175 4700 09763, 4928 040193 3220, 4500 0.9098, 4500 om 207 3253, 4528" 1.0000, ‘4.000 118 220 3,388 4200 1.0501 050 31500 188237 3.360 4000 110865 0.78 21705, 232 282 3.323 3500 12021 1.48 3000 1.3700 222 2500 186237 293 Gas Separator Test Specie pT Separator Separator | _Gravily of (P59) (°F) GOR" Liquid Gravity’ Separator Gas ves 148 7.465 0.6442 0812 Ti ge parte ga a 120 po. 1 ‘CGE of 10.47% N, Partial Phase Diagram the LSLP method is applied again. This sequence of iter- ations converges when all of the new iterate values lie ‘within the latest subregion. Several final iterations are then performed using 140.015, 10.0075, etc., to reduce the final subregion. This reduction enhances validity of the above-mentioned linearity assumption. The LSLP method obtained djc as linear functions of {v;} using a least- squares fit of calculated observations from a number of history-match runs. Here we obtain {djc} as linear func tions of {v} by numerical partial differentiation using the EOS, We do not consider the effects of component pseudo- ization on EOS calculations in this work. The optimal ‘number and definition of components should be dictated bby what process will be carried out in the reservoir. '° In addition to single-contact(¢.g., expansion) laboratory ‘ests, multiple-contact tests and/or reservoir condition flow tests may be necessary to confirm validity of the PVT description. Definition of Terms For convenience and brevity in presenting results, sever~ al terms are defined here. An average deviation, ¢, is de- fined as F*/n,, where F* isthe final or converged value 280 of F. This deviation is not equal tothe true average devi- ation because not all weight factors are unity ‘The term “predicted” is applied to EOS results calcu- lated with no alteration of any EOS parameters. The term “adjusted” is applied to EOS results calculated after one binary (e.g.,C; ~C7.. Xj) is adjusted to match exact- ly the sample bubblepoint or dewpoint pressure, The term “regressed” is applied to EOS results calculated after a umber of EOS parameters have been determined by regression upon a set of laboratory PVT data Except where stated otherwise, the regressions de- scribed use the five variables of methane 4°, 9°, plus fraction 9,°, 5°, and the methane-plus fraction binary interaction coefficient. Rationalization of this selection part from experience is discussed in the Appendix. We refer to the methane-plus-fraction binary simply as the binary, denoted by b or b,. Its value, determined by OS adjustment, is referred to as the adjustment binary, denoted by 6. The term “plus fraction” denotes the heav iest component used in the EOS calculations. For exam- ple, if the original plus fraction, C7.., of a fluid is spit into three fractions, F, Fg, and Fg, then Fy becomes the new plus fraction, SPE Reservoir Engineering. May’ 1986TABLE «EXPANSION AND SEPARATION DATA FOR GAS 5 OE ove at2erer were Devaton end Component 4842 9900. _oo00 2100 1200700700" 9h) WY, 5; Goa? “DoeiT “dood “oozes Voses “Dood aoe “Tao Dae —ia10 we; boos Sonar 1908 o.00s8 cess © Sense 0002 S00 oer tta8 G —SSoet Game aes | Ser oases aes S000 asoas 108 © Shore Sore tour O000 crtss rae oosos seer 948 Gowen ows Gates cour Ome comet asso 990 | esse sass aes Baers cease Oost cover sree 1000 S Sotes Goes ore Bors coms! ors ooass sess 1000 Si Bowwo ooo8o Sore Senne | S.aorr Ooom O.aes Oases 3.90 Es Goss ose Oost Sarse Gen00 Ooeie dessa "aoeo 8888 {Bowe me 18mm tm {Bret {ouea 2 0968 091 osm oss amie ges ‘boos G, “0 onasie user ovetio cewor obras aes tf —-§ Sooo Sewer Gowo “Sone Veet tae {Bias seer __ separator Tes a Sick Tank Sposa istee | pt Sopra vy Gry ot 238 | ha ch Bom cabiateorr)__saprao Gas soe a 7 oes sr ‘TABLE 5—VAPORIZATION AND EXPANSION TEST DATA FOR OIL 1 [raion Test at 2520 psig and 80°F SPE Reservoir Enginering, May 1986 Ineremental Incremental Mois ot lative CCE of Reservoir Compostion of injected 925. injection "Mole of Mois of Gas Liquid Phase Liquid) _Ollat 180°F | Component Mol Fraction Number Gas nected Produced’ Aemaining Volume “p, psig WIV, ° ‘0.0000 ~ 00000 ~~ 1.0000 7.0000 1 oesis = ogase ©8864 .8att 0, 0.0086 2 oszs = 08877 © ae1s, 8616 © 5.000 9782 Ny ore 3 08067 sass Ogz18 «= O84s 4.000 08862 cy 0.8898 ‘ O87! = 0872889598245 3.000 0.9951 c 0.0704 5 o7e13 07912 ©8860 «0.9075 2.900 09961 G o.o163 6 09017 Ogaa2=0.865 0.8889 21800 D997: ci 0.0026 7 ‘ove? 00807-08375 0.8688 2.700 Ogee c 0.0008 8 10353 «1.0600 .8128 0.8502 2.600 O.s0e2 c .0002 8 0.4600 0498307865 0.8408 2.520" 1.0000 on CT oT | n 09705 0908307873 .BT10 Hydeocarbon Analyses (na! traction) of Gases Produced ‘During Veporization Test at 2520 psig and 160°F Injection Number ot 2) 8 #8 | Component ‘CO; 0.0082 0.0082 0.0083 0.0083 0.0083 0.0084 0.0084 0.0084 0.0085 0.6085 0.0085 0.0080 Ne Bote O.0122 O9120 OO1e BOTI8 DONT 0016 BOM OO115 O15 DOTA 0.0020 Gi Gras 0.8781 0.8739 0.8738 0.8745 08754 08760 08757 BTS 08771 08776 03M8s G) 010584 0.0850 0.0670 0.0688 0.0683 0.0685 0.0698 0.0701 co7d2 cOTG2 0.0703 0.0681 | G5 dorss core O.0rG: O02 GOVE O.01Es OIE OIE O.O164 Cor O.O165 O.0270 | G1 0.0087 0.0088 0.0062 0.0039 0.0038 0.0033 0.0081 0.0028 0.0028 0.0027 0.0025 0.0068 Gi 0.0037 d.0027 O.0022 0018 0.0018 0.0014 DO0T2 GO010 0.0009 0.0009 0.0009 0.0060 Gi O.oaat O.006e 0.0055 00047 0.0081 0.0038 0.0032 0.0028 00027 0.0025 O.0023 0.0208 El dois ari 0.0108 0.0106 0.0105 0.0104 9.0103 O.O102 0.0101 0.0101 O.0100 O5189 Wi 408° “tas” sor? “soe” “to” 308° 109-109 to 10910 58 281‘TABLE 6—EKPANSION AND SEPARATION DATA FOR OIL. 2 "aut hace pss CCE at 176° > Component (sig) _wy, CO, 6,000 0.9585 Ne 5500 0.9700 cy 5,000 0.9827 or 4,900 0.9856 Ss 4,900 0.9883 & 4700 0.9919 Ss 4600 0.9951 Ce. 4,500 0.9984 er, 480" 1.0000, 4443 1.0009, M+ 4305 1.0097, 31900 1.0412 z 0796 0783 0.788 0.843 0.913 31581 1.0812 6 © 0.07535 0.17882 0.32871 0.49808 0.63067 3182 1.1426 2769 12002 2422 1.3388 2128 1.4738 11980116084 1.660 1.8415, 351 22768 1081 2.9852 _ DE at 176°F Relative Deviation Ol Gas or Oil Solution Factor Viscosity Viscosity Density 4000 234320510825 0.280 ©3830 562 3492 20581814 0.788 «0.88 «0.0827 (0.5883 30031888 4710772 0.380 0.02800 8082 | 2614 1756 «120507730440 02080 8282 | 2004 11845 = 8700780051022) O.BAST 18341555775 BIS BDZ UIT 8500, 1001 1.484 «5730855748 O0dO B52 505 1972 «98382 0.0120 0.6940 209 1208285 0.858 oor 0.7085 0 1087 0 09851547 0010907813 | ‘Separator Tests | ‘SiockTank Specific 2 Separator —_Stock-Tank Gravity Gravity of (sig) GOR GOR FVF_ Separator Gas 300 60 1.597 275 215 O74 50 60 1,983 8 air 0.805 Bubieoont sree. We refer to constant composition, constant-volume, and siferential expansion data as CCE, CVE, and DE data, respectively. We use the symbol G, for cumulative gas removed (mol fraction of original) from a cell during a CVE, The symbol f,. denotes volume faction liguid in cell during expansion, At each expansion pressure, / is liquid volume divided by cel volume at that pressure Fora CCE, the cell volume increases as pressure drops. For a CVE, the cell volume is constant and for a DE, cell volume decreases as pressure decreases ‘Gas gravity, 7g, is simply gas-phase molecular weight divided by the molecular weight of air (28.97). Liquid sravity, 71, oF yoy is defined relative to water= 1.0 (ve, ‘yz is roughly liquid density in pounds per cubic foot divided by 62.4). Standard cubic feet of gas are defined relative to standard conditions of 14.7 psia {101 kPa} and 22 (60°F [16°C]. All pressures are in units of psia unless stat- ed otherwise. ‘Sample Data ‘Tables 1 through 10 list composition, expansion, and separ ration data for Gases 2, 4, and S,and Oils 1 through 4, 6, and 7. In these tables, all temperatures are in degrees Fall pressures are in psig, a, is in pounds per cubie foot and viscosities are in centipoises. Unless otherwise not- ed, separator GOR is standard cubic feet of primary sepa- ator gas per stock-tank barrel. For separation test data, single spacing is used to indicate multistage separation, Entries that are double spaced correspond to different separation tests on the same sample. For example, for Oil 4 in Table 8, three different separation tests are given, each consisting of three stages. Following the last entry SPE Reservoir Engineering, May 1986‘TABLE 7—CONSTANT-COMPOSITION EXPANSIONS FOR OIL 3 CoE ata CCE at 6 P. Ep i (os) VV, (8) _(63ig) VV, 8) 5,000 0.9900, 4500 0.9488 #000 09551 3500 0.9664 3000 09752 2500 0.987 2400 0.9907 2300 0.9837 2200 0.9970, 2,15" 1.0000 100.0 2092 1.0082 1000 068 1.0089 043 1.0148 94.8 ‘990 10279 8a. 927 3.0887 854 ont 834 3.0877 788 854 1,888 1.1862 685 805. 1487 1.9820 546 739 1/908 1.5980 45.0 6731 1163 1.8705 385 603 1083 21616 501 ‘O49 2iaaaa 426 B69 27542 389 817 29498 206, 74 3.1917 180, 72 34819 158 653 3.8585 212 eo2 4.2312 495 559 4.6202 184 172 *satepes esr in any separation test, there is an implied final flash to stock-tank conditions of 0 psig {0 kPa], 60°F [16°C]. The reported gravities are separator gas gravities and stock tank oil gravities unless otherwise noted. Data are given for Gas 1 by Firoozabadi et al.,? Gas 3 by Vogel and Yarborough,? and Oil 5 by Hoffman et al.” ‘Because reservoir fluid samples occasionally vary with location and time, these 12 samples may or may not be representative of their respective source fields. Discussion of Results ‘The average deviation gives the most concise but least informative comparison of observed and calculated re- sults. Table 11 lists these deviations for the 12 samples after EOS adjustment and regression for both EOS. Regression reduces the adjusted deviation by factors rang~ ing from 17 to only 1.28. In general, the adjusted ZIRK EOS compares better with data than does the adjusted PR EOS. However, the agreement with data after regression is, on the average, slightly better with the PR EOS, For the oil samples, except for Oils 1 and 5, the adjusted ZJRK EOS resulis compare reasonably well with the data and ‘are improved only moderately by regression. ‘Table 12 compares experimental and calculated values ‘of a number of PVT quantities pertaining to reservoir tem- perature expansions and surface separations for the 12, samples. The PR EOS results are listed and the ZJRK EOS [SPE Reservoir Engineering, May 1986 CCE at 180% CoE DOO > 7% i (si) vv, 8) viv, (8) 5,000" 0.9967 e386 4,500 0.9481 0.9463 4.000 09567 0.9588 3500 0.9601 09727 3.000 0.984 0.9829 | 2900 0.9878 0.9866 2800 09014 0.9907 2700, 0.9956 0.9943 2597" 1.0000 1000 0.9985 574 1.0032 4.0000 100.0 551 1.0070, 10027 | 2596 1.0003 92.7 {0070 947 | 2521 ome en7 son? 922 2492 10177 89.9 0220 872 Hoag as 0868 805 10624 78.9 $4082 712 11902 69.8 12188 603 32785573 13849 50.0 eos 14988 46 15216 438 1381 1771S 968 1.8684 98:7 | 4203, Zosre 4.8926, 1981 2.3786 2.1598 72 27187 wizt 2aaae 995 28858 208 1019 26951 217 868 30577 187 98920622 193, 881 32510 178 669 S117 18, 788 3.4507 165 B24 3.4298 75s 3607 155 757 3.7608 TIT 38322 145 698 41252, 634.1721 843 4.4922 | 617 4.5967 pin 4es70 results are given in parentheses. For examples with more than one surface separation, the results given are for the lowest-pressure separation. The B, and R, values are at bbubblepoint. The first and second Gas 2 entries are dew~ point and bubblepoint samples, respectively. The first and second Oil 5 entries correspond to use of 7 and 22 com- ponents, respectively, in the calculations. The Oil 3 eal culated results used 12 components, through Cp ‘The results listed show the rather poor predictive abil- ity of either EOS. In general, the predicted bubblepoint ‘oF dewpoint pressures are consistently and significantly low. With only adjustment, the ZIRK EOS yields saturated-oil densities (at bubblepoint pressure and reser~ temperature) and stock-tank oil specific gravities (at (60°F [16°C)) that are consistently higher and significantly more accurate than those from the PR EOS. The table shows that adjustment (changing only the binary) has vir- twally no effect on stock-tank oil gravity and calculated surface separation results for either EOS. Surface separation calculations show that both EOS generally predict erroneously low GOR and oil FVF. As stated previously, adjustment does not alter calculated separation results. An obvious question is whether repres- sion only on reservoir temperature data (e-g., expansions) gives EOS reliability under surface separation conditions. Table 13 compares experimental separation results for ils 5 through 7 with two sets of regressed PR EOS re- aes‘TABLE 8—EXPANSION AND SEPARATION DATA FOR OIL 4 | ~ DE at 2508 tet ore | Fauve Dewaton OF Foaive ‘eviaton Ol pot Soutien “Factor” Deraty e sstuton "Factor (099) Volume “GOR __z__igem") os) GOR” _ Doar yer: 932 0. 1958 701 2360 183 © BBS «aso sat tise tae 6a o706 ora 235 1595 78a) B.609 Sr (BN Sr hoe Oka arte ‘eos ss Yon ee Cas tae 12 Sioa a0 8708 ves 1512 65ers (oer79 iss 12050367570 our vas 473 SB ee Oa oo fer a Gase Game ON Miso 1498 477888 gase fae ita 30a Ose ON ‘bes or Mr Ose one Si ties mo Gag? Osan Oren eer 385 S38 Oar arise Ser hte Boa Ger Geer Ora foo S388 bes Geo a rane tr tie sae Ore er Se i2rs 180 One crams 95s hagr tosses rag ag 7 2a ae ©9802 tse Dts 07004 1506 0 toa D Ofer 2881 CCE at 10°F (689 WV, og 5) VY, na ww, 3500 D985 3000 9897 3000 9890 2547 10000 0222 22H3 1.0000 0287 1998 Sa00 065 2540 10865 0256 2.053 10445 0483 1705 t cae? | 2056 13098 0289 yer Nis Oae2 1598 II | eer 1238 1450 12685 9890 1238 25rd os | Vee 1400 ther {4765 08s 1082 1307 Oee0 50 26005 oro ‘N2 21798 0700 "Mea 270 1190 am assae Sis 27s so 133 1 392 37876 0.820 | Separate Tess | StockTank Separator Oi | PT Separator StockTank Gravy Volume arty | (oS, oh “GOR” “GOR” abit sore) Facer —_(glem’) 20 110 S61 08 0786 250 350620 1320 Ore 0 180 53 387 tom O70 250 180677 iis 765 aja “Bh tin Over 8 180 2 37 oss ree sults. Regressed Values | result from a regression includ- ing both reservoir temperature and surface separation data, Regressed Values 2 result from regression only on reser~ voir temperature data. Table 13 illustrates our general finding that regressed EOS surface separation results are about the same regardless of whether separation data are included in the regression data set. ‘The degrees of C7, splitting used for these 12 sam- ples ranged from none to four fractions. A general, a pri- fri guide to this need is given by the experimentally “observed range of C74. molecular weights during an ex- pansion or multiple-contact test. For example, we found C7 splitting into one (no splitting), three, and two frac- tions advantageous in matching Gases 1, 2, and 3 data, respectively. The experimental ranges of C4. molecu- lar weight were 145 to 110, 199 to 118, and 171 to 123 for Gases 1, 2, and 3, respectively. Also for any given sample, the anticipated recovery process affects the required degree of splitting. Gas in- jection processes with vaporization phenomena require somewhat more splitting than depletion/waterflooding Processes. Table 14 shows the final values of the five regression variables for the 12 fluid samples for both the PR and ZIRK EOS. Inall cases, the regressions converged to the variable values shown, The adjustment binaries show no 2a correlation with plus-fraction properties. This was also noted by Katz etal." in their applications of the PR EOS. ‘The man-hours spent in studying the 12 samples, in- cluding data preparation, ranged from about 6 for Gas 3 to about 20 for Oil 1 and Gas 4. Obviously, required man-hours depend on the engineer's experience and familiarity with the PVT program used, the amount of available data, and difficulties that arise in the matching effort. The rather low man-hours quoted, however, reflect primarily thatthe regression feature allows rapid evalua- tion of EOS parameter sets and values. Gas 1. Gas 1 exhibit a dewpoint pressure of 4,075.4 psia (28 100 kPa at 180-5°F (82.5°C]. CVE data are given by Firoozabadi er al.* Published 148 7.796 (7.613) 8,818 (7.840) 7.485 (7.485) separation, Gass! 96. 8,407 (81508) 8,408 (8.502) 6,898 (8,953) Aye (setiSt@) O12 832.460) «2.185 (2.454) 2.543 (21626) O14 0 604 (693) 603 (692) 1693 (705) 350 649 (750) 649 (748) 748 (762) 180 889 (791) 682 (790) 78 (803) ois 77 752 (857) 752 (888) 855 (896) ols 77 760 (878) 761 (877) O16 78 910 (1.048), 911 (1,088) 1,081 (1,089) O17 2 545 (695) 546 (635) 601 (615) Stocktank oi i260 0.709(0.803) 0.709.803) 0.897 (0.868) gravity from OS 110 O7i1 (0815) 071210815) 0.814 0.820), De 250 0.737 (0.890) 0.717 (0.830) 0.827 (0.834) ols 60 0.718 (0.823) 0715 (0.824) 0.837 (0.868) 0170 0.790 (0.808) ‘SPE Reservoir Engincering, May 1986on > © 70 60 is so 2 4 8 01000-2000 3000 40608000 P,Psio po SIG Fig. 3—Gas 3 CVE dewpoint sample at 190°F. 288 TABLE 12—Continued _ Propety Sano Eominenal __Podiod Adee _—_—Ragasnd Sockiank ot! as 2 o7eo o7e 0529 Samion Si osis.oe) —oseecosen) Sean esis Sa" gag or: Shusioray —Stae(ea) Sor em Stain Sia bas Shsorey Seater Sarton oie $38 Gray Garien) Selene See Grd Gay Saislesta Ser SPSS) Goatees) asian | ons ee EH Sheps — onvtoaen | ole 0.821 0.712 (0.818) 0.712 (0.816) 0,828 (0.854) ai oH Gress © Graces) Sen\ee at ors 2" 35120 saos0 onsen) e9280 (ae) 5 002590) Crs RISES) See Zen eno en 08 LSE Bee a wor | ° | ara dicated by the 4% deviati od to the 5% | > jae by the 4 average deviation compared the PRREGRESSED PP deviation shown in Table 1 ZAK, REGRESSEL ‘The regression data set excluded K-value data from the 5 tS reported CVE gas and residual liquid compositions at 2g 1 514.7 psia [3548.7 kPa]. Table 6 shows that the regressed : PR EOS gives god agree with he expereta ig 4 \ Sid compostions The ZIRKC EOS ge equal good \ agreements The represion daa st sso exclude o- | Sbge seperation da forthe dwpoin sul Table 16 2 | compares PR apd ZIRKC EOS rete with hse dat To Bic shows the poor predixosofdewpont pressures \ Be ows er 5 10 ‘As shown in Table 11, the average deviation with ad- jjustment was reduced more than two-fold by regression; splitting the C7, resulted in better agreement between EOS and experimental results. Table 14 shows that, gener- ally, large methane © values and small plus-fraction @ ‘values were obtained in regression on Gas 2. We feel that slight changes in sample compositions within the realm ‘of experimental error might have a large impact on these regression-variable values. “A mass balance on the dewpoint-sample CVE data gave ‘very reasonable liquid-gravity values. The regression data sot included these 7, values and available 7_, z,,z, and G, values for each of the six CVE pressure steps. 1 also included saturation pressure and density and CCE rela tive volume and fi, data for both samples. ‘The proximity 10 critical of the Gas 2 compositions is indicated by the K-values for the dewpoint and bubblepoint samples at their respective saturation pressures shown in Table 17 that were calculated by the PR EOS after regression. Gas 3. Gas 3 is Vogel and Yarborough's® “Gas Dewpoint pressure is 4,453 psia (30 702 kPa] at 225.8°F [108°C]. They presented plots that compared observed values with their RK EOS-calculated value of liquid dropout for the reservoir fluid and for 10, 30, and 50% No mixes. Their 30 Na mix, for example, is a mixture f 0.7 moles of reservoir gas with 0.3 moles of Nz- Vogel and Yarborough used 42 components in their EOS, splitting the C74. fraction (9.05 mol%) into frac- tions C; through Co. They tuned this extended analy- ‘SPE Reservoir Enginering, May 1986ois Separation Conctons PT, Property ‘cont Ys % 060 © Goa 310 x a 1475 ous 200,74 GOR 722, % ois a 1.611 100,74 GOR ‘10 0 0.8123 a 5.610 80,74 GOA a8 Ye 08137 a 1.627 07 ada {oso "6 0.9960 x 0.8208 B 1722 ou7 160,72 GOR 366 0 0179 & 1.302 8,72 GOR ‘24 Ye oir a 1.299 or ‘a72 10 08173 a 1.308 0.72 GOR 580. % 1.075 x 0.8265 a 1340 ‘Rearesee rss ncag abe sano an nutcn conns Sine nach ay sis with the reservoir gas data and then calculated good agreement with observed liquid dropout data for that gas and the three Np mixes Fig. 4 compares observed results with our regressed, 10-component PR EOS results for Gas 3 and its three Nz mines. The agreement with data is comparable to that ob- tained by Vogel and Yarborough. The C>.. was split into ‘wo fractions and the five regression variables were the Usual methane and plus-fraction s and the inethane/plus- fraction binary. Table 14 shows the reasonable values found by regression. No N2 EOS parameters were al- tered or regressed. The regression data set included Gas, 3 dewpoint and liquid dropout data and the single addi tional data point of 8,006 psia [55 200 kPa] dewpoint pres- sure for the 30% Nz mix. For both EOS, the average deviation fell from over 40% after adjustment to about 1.8% after regression, as shown in Table 11. The regressed ZIRK results agree with the data on Fig. 4 equally as well as the PR results. The agree SPE Reservoir Engineering, May 1986 19--SURFACE SEPARATION DATA AND RESULTS FOR ‘OILS 5 THROUGH 7 “Repent aaa nar ony reser cnaSrce sparen wee ae based 7a? 726 | ose = o.r23 1.606 1.605 808 208 oss = oni | 1.603 1.602 | 76 ‘73 | oats 08136 1612 1817 1061 1060 ogee = o.9789 oars = 0.8270 | 1.730 1720 32 ae o7es = 0.7786 1.242 1.208 | 489 85 07631 or7ee 1.208 1.230 ‘7 ‘es o7se 07790 1.246 1.237 600 ‘596. 1053 052 07908 (0.7883, "1302 1.298 ‘ment shown on Fig. 4 is only slightly poorer when the 30% Nz mix dewpoint is excluded from the regression data set. ‘No relative volume, dewpoint fluid density, or usable surface separation data were given by Vogel and Yar- borough. In cases of missing density data, we closely ex- amine the liquid gravities calculated after regression. If poor data or EOS inadequacy has resulted in unrealistic parameter values, this will frequently appear in the form of obviously erroneous calculated CCE or CVE liquid _gravities. In this case, both EOS calculated very reasona- ble liquid gravities at reservoir temperature, increasing (with decreasing pressure) from about 0.52 t0 0.7 for the reservoir gas expansion. At any given intermediate pres- sure, liquid gravity increased significantly with increas- ing No content. However, the highest calculated gravity, for 50% No at 1,015 psia [6998 kPa], was 0.770. ‘When the methane-plus fraction binary was omitted from the variable set for Gas 3, the methane 3 value 289‘TABLE 14—FINAL VALUES OF REGRESSION VARIABLES PREOS Sample ne _6 Gast 9 0140 2.069 Gas2 9 0096 0108 0254 0.058, Gas? 11 0272 0.098 0391 0054 © — 0.037 10 9021 0.068 0278 0051 0.186 12 0.089" 0.073 0205 008 © O.a54** 90177 Dos 0342 0.065 0.285, 0 046 0.096 034 0.08 © 0.181 9 0135 0.085 0763, 0.088, -0.211 10 0.058 0.082 0488" 0.058" 0.108 a 0x37 0.064 0899 0.080 0.178 8 oe 0078 0828 0.081 © 0.088 | 2 orm 0080 0420 0.069 0.072 | @ 0308 0408 Gai 0.067 0.056 7 v.08 0.080 0.288 0.088 0.284 | 22 0.283 0.067 0347 0.048 © 0.056 9 OnT 0088 0478 0.087 0.157 8 0082 oor osm 0075 0.100 2uAk Eos O11 04a 0077 03820105 —0ase Ose 0629 0110 0257 0069 © oaIe 0200 0486 0.103 0226 0045 © 0.058, 0.07 0.453 0085 9310 0.058 -0.020 0s 052) 0085 0529 0.056 ° 0143 0.509 0.090 0820 0.077 0.380 | 0148 0549 0.103 0498 0075-0017 0073 0445 0088 0798, 0087, -0.280 0.043 0425 0.087 0518" 0.075" | 0.108 0.086 034 0.079 0.500 0.095 0.181 | O48 0.361 0063 0816 0108 0.180 O14 0362 O07 0610 0.103 0138 0082 0475 0.096 0.508 0.087 0.058 0030 0423 00s! 0328 Odes 0.318 0182 0416 0083 0.578 0.082 on O06 0.963 0082 0.468 0.088 0.102 0017 0839 0082 0343 0.071 -0.080 dU Regreon olss uses ty Oy ‘converged to a value near 1.1. Such wide departure of ‘a rogression variable from its theoretical value can result from poor data, EOS inadequacy, too many regression ‘variables, or too few regression variables. In this case the ‘cause was too few regression variables. Addition of the methane-plus fraction binary resulted in converged, reasonable values of all regression variables. Gas 4. Gas 4 exhibits a dewpoint of 3,375 psia [23 270 KPa] at 240°F [116°C] and 134 bbl [21.3 m°] of sepa- ‘ator liquid at 1,215 psia [8377 kPa] at 148°F [64°C] per Experimental co, 0.82 c, 1585 co 362 cS 3.88 ce 463 cS 548 c eas 20 Cn 5848 1108 sof [28 317 std m?) of separator gas. The reser voir fluid composition through Cy. given in Table 7 shows an H§ mol fraction of 0.1819. Available data in- clude reservoir fluid dewpoint vs. temperature from 73 to 262°F [24 to 128°C], dewpoint vs. mol% No for four mixes of reservoir fluid and Ny, and CCE data includ- ing liquid dropout values for the reservoir fluid and the four Nz mixes at 240°F [116°C]. These data are given in Table 3. The 4.89% Nz mix is to be interpreted as ‘a mixture of 4.89 moles of Nz with 95.11 moles of reser- voir gas. TABLE 15—CVE RESIDUAL LIQUID COMPOSITIONS (MOLY) Gas =714.7 psia 5.05 428 4at 362 467 445 351 348 3.85 3.68 05.23 e847 SSPE ReservoicEnginering, May 1986oa ¢ @ EXPERIMENTAL PR, REGRESSED 3500| 2009.5 —T00 150 200 TEMPERATURE, °F 250 500 Fig. 4&—Gae 3 N, mix CCE at 225, Fig. 5—Gae 4 devpoint prossure vs. temperature, All calculated results presented were obtained with a splitting ofthe Cg, into four fractions. Spliting into few- er fractions resulted in a poorer match of data and split- ting into more fractions did not improve the match. The ‘methane 03 value was used as the single regression vari- able for adjustment in place of the methane-plus fraction binary. This was done because the spliting of Cg, gave 1 plus fraction of only 0.088 mol%. The regression data set included only reservoir-fluid CCE and single-stage sur- face separation data. No temperature-dependent dewpoint data or Nz mix CCE or dewpoint data were included in the set. ‘Table 11 shows that both EOS give average deviations of about 50% after adjustment. Regression lowers those deviations markedly to 1.02 and 0.67% for the ZIRK and PR EOS, respectively. Fig. 5 shows that the ZJRK EOS reproduces the observed dewpoint pressure variation with temperature somewhat better after regression, even though no temperature-dependent dewpoint data were in the regression data set. The PR EOS, after regression, gave ‘2 somewhat better match ofthis temperature dependence than did the ZJRK EOS. Fig. 6 shows that the ZRK EOS match of dewpoint pressure vs. mol% Nz is poor without regression and ‘ery good with regression, even though no N> dewpoint data or Nz EOS parameters were used in the regression ‘The regressed PR results are comparable with these ZJRK results, Fig. 7 shows good agreement between observed and calculated CCE relative volume results for the original reservoir fluid and the four Nz mixes. However, Fig. 8 shows rather poor agreement between observed and cal- culated CCE liquid dropout curves. All calculated results shown are for the regressed ZIRK EOS. The regressed PR EOS results are insignificantly different. The regressed BOS matches the reservoir fluid liquid dropout nearly ex- actly but seriously underestimates the amount of liquid ‘dropout near dewpoint pressures as No is added to the SPE Reservoir Engineering, May 1986 [TABLE 16—COMPARISON OF CALCULATED AND. OBSERVED TWO-STAGE SURFACE SEPARATION RESULTS. Gas 2 Primary Separstor: 208.7 pala, 70°F | Primary Stocktank Separator ne GOR" % 7 | expetimental ~~" Bato" 07800777 | PR 9 3609 © ogra (0754 PR 11 30027174 WRK 8 38490818754 ZRK 1 34970828782 Gas 5 Primary Separator: 490.7 psia, 96°F oss" «O71 ——«OT8S 9 seo © 0800 ©0722 10 8804 ©0805 0.722 9 9089 08210721 10 8953 0.805 0.723 {Separate ST a OF ead [ ‘TABLE 17—K-VALUES FOR GAS 2 DEWPOINT AND BUBBLEPOINT ‘SAMPLES ‘Dewpoint Bubbepoint Component Sample __ Sample CO, 7.00188 c, 4.00602 c; 1.00126 c(0.99818 Cc. 0.99533 cS 0.88277 Cc, (0.98984 FE ogeata 0.97390 F, 0.93465 21reservoir fluid. The liquid dropout date match was not improved by including No Mix 4 CCE data in the regres- sion data set and Nz @3, 93 values in the variable set. Because of the small pls fraction mol faction after split- ting, the regression variable set included only the four variables of methane and phus-raction 2,25. Table lists the values of these variables converged on by regression, Both regressed EOS calculated separator GOR and lig: uid gravity (at separator conditions) as 7465 scfbbl [1345 std m?/m3) separator liquid and 0.615, respectively, compared to observed values of 7465 scfbbl [345 sd m?/m?] and 0.644. Many variations of regression data and variable sets and degrees of spliting were tried without improvement in the liquid dropout match, A numberof possible explana- tions for that mismatch are possible; We donot know ‘which isthe most probable Gas 5. Gas 5 exhibits a dewpoint pressure of 4,856.7 psia (33 486 kPa] at 267°F [131°C]. Separation yields 136 bbl [21.6 m?] of condensate at 440 psia [3034 kPa] and {60°F [16°C} per 110° scf [28 317 sid m’) of separa- tor gas. Table I gives the reservoir fuid composition through C7... Available data in Table 14 include CCE, CVE, and surface separation data, |—FLUID COMPOSITIONS AT LAST INJECTION STEP FOR OIL. 1 TABLE tava PR Component Experimental no spit CO, 0.0060 0.0037 Ne 0.0022 oar c 03480 0.3829 a 0.0680 0.0372 cS 0.0279 © ooras c 0.0084 0.0037 c 0.0051 0.0013 ce 0.0208 0.0043 or 0515805516 PA PR PAL split Experimental no split _split ‘Doo 0.0085 «0.0088 0.0086 O00 = O01t4 = OO118 ONE 03825 08776 0.8887 0.8800 0.0523 0.0703. 0.0703 0.0704 0.0217 0.0165 | ooi69 9018s | 0.0062 0.0025 0028 o.0026 0.0081 0.0008 0.0008 0.0008 00147 0.0023 0.0013 0.0024 (05528 0.0100 0.0000 0.0070 ‘A mass balance on the CVE data gave reasonable but slightly erratic liquid gravtis as shown on Fig. 9. The repression data st inchided K-values a the last CVE pres- sure, surface separation data, CCE data, and values of fis Tg 11, and G, for each expansion presse. The er rate 66852 liquid gravity at 3,015 psa [20 788 kPa] was ‘omitted from the regression data se. Fig. 9 compares CVE data with PR and ZJRK results calculated after regression with Cy split into two frac- tions. Where the circular PR points are not shown, they coincide withthe triangular ZIRK points. The agreement with dat is very good for both EOS with the exception Of zz and 7). Table 15 shows reasonably good agree- rent between CVE residual iid compositions using the ZIRK EOS. The PR EOS compositions do not agree as well. Table 16 compares observed and calculated (regressed) EOS rests forthe two-stage separation. The 1c Ey MOL PERCENT N2 35 3505 805 305 2 psc j—Gas 4 dewpoint pressure vs. mol% Nz at 240°F. | Fig. 7—Gas 4 CCE at 240°F. SPE Reservoir Engineering. May 1986DATA, 267 °F 3800 Fig. 8—Gas 4 CCE at 240° GOR is matched exactly, while the calculated stock-tank liquid gravity of 0.805 differs from the observed 0.781. Oil 1. Oil 1, with composition given in Table 1, has @ saturation pressure of 2,535 psia [17 478 kPa} at 180°F [82.2°C]. This fluid was subjected to a multiple-contact vaporization test in which gas with composition given in Table 5 was injected into the oil sample in a visual PVT cell at constant pressure and temperature in a series of steps. At each step, the fluids were allowed to reach an equilibrium. The gas was removed at constant pressure and analyzed. The volume of oil was measured before the ‘next gas injection. This process was continued for 11 in- jection steps. Measured data from this test are given in ‘Table 5 and include the moles of gas injected and pro- duced, the moles of liquid phase remaining in the cell, the composition of the gas at each injection step, and the ‘composition and molecular weight of the residual ol af- ter the last step of the test. Table 18 compares calculated and observed liquid and gas phase compositions atthe last injection step. ‘The PVT program uses mass-balance considerations to calculate additional data at each step, including oil and £888 gravity and liquid-phase molecular weight. The meas- ured molecular weight of the C74. fraction of the gas at the different steps ranged from 105 to 110. The meas: ured oil-phase C74. molecular weights for the reservoir fluid and last-stage fluid were 225 and 258, respectively. This wide range of molecular weight of C>,. presented , mol fraction with a 173 molecu- 2Where 90 mu sabes | oof 20 * tee INseeTiOn Muner Fig. 10—Oll 1 multiple-contact vaporization te lar weight. CCE, CVE, and DE data in Table 6 were used along with all data from the two surface separations in regression. Both EOS predict bubblepoint pressures of about 3,400 psia (23 442 kPal, considerably below the observed 4,475 Psia (30 854 kPa]. With adjustment, both EOS yield B., and R, values significantly lower than observed. Table 11 shows that regression reduces the average deviation from 10.25 to 4.68% for the PR EOS. With splitting of the C7, fraction into fractions F7, Fg, and Fy, and regression on nine variables (93, 9% of C1, F7, Fs, Fo, 24 Fig. 12012 DE. tnd the C/Fy binary), the average deviation falls fur- ther to 2.17%. Figs. 11 and 12 compare DE observed and calculated Bo, Ry, and liquid gravity for the PR EOS for the cases of prediction, adjustment, and regression. Fig. 12 shows that the Cy, splitting with regression results in a virtu- ally exact match of the DE data. ‘None of the regressions with either EOS gave good matches of the surface separation data, as shown in Ta- ble 12. Regression with the surface separation data alone also resulted in a poor match wit either EOS. This, com- bined with the mass-balance error in the data and occur- rence of waxing, lead us to suspect the data Oil 3. Oi 3 contains 60 mol% CO; and exhibits a bub- blepoint of 2,612 psia [18 010 kPa] atthe reservoir tem- perature of 179°F (82°C). CCE data for this sample at temperatures ranging from 140 to 200°F [60 to 93°C] re shown in Table 7. The reported analysis to Cio. for this sample is given in Table 1. Fig, 13 shows the match of saturation pressure with the PR EOS with 12 compo- nents over the range of temperatures. Predicted values are approximately’ 500 psi [3447 kPa] lower than ex- perimental values. Adjustment gives good agreement, ‘while regressed results virtually duplicate the experimental data. CCE results are shown in Fig. 14. The PR EOS- predicted values give large error for both relative volume And liquid volume. Regressed results agree well with the data, ‘As indicated by the average deviations in Table 1, the regressed PR results match the data significantly better than the regressed ZIRK results. However, use of only nine components (through C7), with ether EOS, gives agreement with data almost equal to that obtained with 12 components through C9. ‘The regression data set included CCE data at the four temperatures. The usual five-parameter regression vati- bie set was used except that CO> replaced methane. Oil 4. Oil 4 is slightly volatile with B,=1,671 and R,=932 at 250°F [121°C]. The nine-component analy- SPE Reservoir Engineering, May 1986Fig. 13—Etlect of temperature on saturation pressure. | sas of this sample through C7 is shown in Table 1, with CCE, DE, and separation data given in Table 8, The ef- fect of temperature on saturation pressure is shown for Oil4 in Fig. 13. The PR EOS-predicted values for satu- ration pressure are in error as much as 500 psi [3447 kPa] ‘Adjustment improves the calculation, but regression again virtually duplicates these data. Results from a DE of Oil 4.at 110°F (43°C) are given in Fig. 15. The adjusted PR results are very low for Rj, By, and, with the +, values showing the most error. Regression provides an excellent match of all data for this sample. Spliting the C7. resulted in insignificant improvement in agreement between observed and regressed EOS re- sults. The average deviations of 2.3 to 2.6% after regres- sion shown in Table 11 indicate that the two EOS give comparable agreement with data. The regression data set included all CCE and DE expansions and all surface sepa- ration data lS. All data used for Oil 5 are given by Hofmann et al."” Their data include extended analyses through F22 for a saturated oll and is associnted gus. Data are given for fash ofthe ol at 4.7 psi (101 KPa} and 60°F [16°C] flash of the associated gas at 800 psia [5516 kPa] and 83°F {28°C}, and CCE data a 201°F [94°C] forthe associat ed eas Katz and Firoozabadi' applied the PR EOS to these data. They concluded thatthe EOS accurately predicted the associated gas data and, with adjustment, matched the oil data In pat, teit conclusion rested on close agree- tment between the observed ges composition and tha cal cilated from the oil composition by use of the adjusted Eos. Practical considerations in simulation require that a sn- fle st of BOS parameters be sed to represent both the Oil leg and gas cap ina saturated reservoir. Calculations here therefore use only te oil composition as known in- put data. All calculations ofthe gis CCE and fash use the calculated composition of gs in equilibrium withthe olla calculated bubblepoin pressure and 20°F [94°C] Figs. 16 and 17 compare observed and calulsted lige uid dropout and gravity from a CCE ofthe associated pas Sample for Oi'5. Adjusted vals for a22-component ss tem and regressed values fr a seven-component system [SPE Reservoir Engineering, May 1986 Teo + Fa Eos aevuSres Pr E98 RESRESSED Fig, 14-03 CCE. | — ara sal 4200 +76 ®. . 2 deoo dre Re 1% 7 100 <0 | «| ob is eas as aaa Heh P.PSIG | ssFig. 16—CCE of OW 5 sociated gas. | Fig. 17—Gravity of separator liquid, CCE of associated gas. from OW 8. | using both the PR and ZIRK EOS are presented. The ad- {usted 22-component EOS results compare reasonably well with data; the regressed seven-component results com- pare significantly better and about equally well for both EOS. Table 11 shows that for both EOS, seven-component regressions give average deviation of about 2%, as low 8 Flower than 22-component regressions. With 32 com ponents and the usual five regression variables, noncon- vergence occurred. Removal of the binary from the variable set resulted in convergence. Both EOS's predicted bubblepoint pressures about $00 psia [3447 kPa] too low with either 7 of 22 components ‘The gas flash results in Table 12 show thatthe use of 22 rather than 7 components results in more accurate EOS- predicted and adjusted values ofthe flashed-gas GOR. The seven-component regressed EOS result, however, com- pares well with this GOR. Oil 6. Oi 6 is moderately volatile with B, =1.866 and y= 1,230 at 234°F [112°C]. The nine-component anal- ysis includes minor amounts of CO; and No and a 10,3043 C>.. fraction with a molecular weight of 200. The CCE and BE data, along with data from four two-stage separations at 74°F [23°C], given in Table 9, were used 296 °,P516 re oreon | in regression. Table 11 shows that regression resulted in average deviations of 2.1 and 2.67% for the PR and ZJRK EOS, respectively. Fig. 18 and Table 12 show the close match of data after regression with the PR EOS. Split- ting gave insignificant improvement. Oil 7. Oil 7 is the least volatile of the oil samples with B, =1.324 and R, =557 at 131°F [55°C]. The CCE and [DE data along with data from four two-stage surface sepa- rations at 72°F [22°C] given in Table 10 were used in regression. The nine-component analysis includes minor amounts of CO2, Nz, and a 0.3597 C;, mol fraction with a molecular weight of 252. Table 11 shows that both EOS give average deviations of about 4% after regres- sion, With adjustment only, the ZIRK EOS gives a sig- nificantly better fit of the data. Splitting the Cy, fraction into three fractions gave insignificant improvement in the atch of data. Conclusions Among the PVT program features described, we find the regression capability most important in efficient valida- tion of an EOS before its use in a compositional simulator, Data given for six oil and three retrograde gas conden. sate samples include constant-composition, constant- volume, and differential expansions, surface separations, temperature-dependent saturation ‘pressures, and Nz reservoir fluid behavior. One set of multiple-contact oil vaporization data is reported. ‘The PR and ZJRK EOS are applied to these nine fluids and three published fluid data sets under conditions of pre- diction (no alteration of EOS parameters), adjustment (al- tering one binary coefficient), and regression. Agreement between laboratory data and regressed EOS results is generally good to excellent. Results for these 12 fluids and a larger number of unreported studies indicate that regressed PR and ZIRK EOS give very comparable agree- ‘ment with data, In either predictive or adjusted modes, both EOS give generally poor agreement with any reasonably complete ‘SPE Reservoir Engincering, May 1986set of laboratory PVT data. We find regression necessary for required engineering accuracy in EOS results. ur studies indicate that regression on the methane/plus- fraction 92, 0 EOS parameters and the methane-plus fraction binary is frequently necessary and sufficient for good data matches. Further, we find a minimal need for the extensive spliting of C7, used to match data in several published studies. In this work, generally good agreement with data was obtained with C>, splits rang- ing from none to four fractions. ‘The extent of splitting required depends primarily on the recovery process anticipated. Below-, fraction ‘The results ofthis work illustrate our general observa- tion that an EOS tuned by comparison with only reservoi temperature (c.g... expansion) PVT data frequently gives ‘good agreement with surface separation data. In some cases, such as Oil 2 and Gas 4 of this study, 4 portion of laboratory PVT data may remain poorly matched by regressed EOS results. Such disparity can fre- quently be resolved by more fully exploring regression variable sets and C7, characterization (splitting). Re~ maining disparity leaves an open question regarding ‘causes of EOS inadequacy as opposed to poor data. Data errors and inconsistencies can be detected in some cases by simple mass-balance calculational checks. Nomenclature B,, = oil FVF obtained from a differential expansion, RB/STB [res m3/stock- tank m?] il FVE obtained by surface separation, RBISTB [res m*/stock-tank m3] bby = methane-plus fraction binary interaction coefficient binary interaction coefficient between ‘components i and j salue of b determined in EOS adjustment sbservation (data item) j included in a regression data set dic = calculated value of d ‘fi = volume fraction liquid in expansion cell, wiv f= volume fraction liquid in expansion cell, VIVs F = objective function, defined in Eq. 1 F* = value of F on convergence of regression Bor G, = volume or mole fraction of gas removed from a laboratory constant-volume expansion cell ‘M+ = molecular weight of the plus fraction ng = number of components ny = total number of observations in regression data set p = pressure, psia [kPa] [SPE Reservoir Engineering, May 1986 Subse = critical pressure, psia {kPa} Ry = solution gas obtained from a differential expansion, scf/STB [std m?/stock- tank m?] Rig = solution gas obtained from a surface separation, sef/STB [std m3/stock- tank m3] T = temperature, *F [°C] T. = critical temperature, °R (K] regression variable i laboratory expansion cell total volume folume of liquid in expansion cell folume of expansion cell at saturation pressure ‘weight factor on observation d; in definition of regression objective function, F ‘gas-phase deviation factor gas gravity, air=1.0 hydrocarbon liquid gravity, water =1.0 oil gravity, water=1.0 average deviation, F*/ny density of fluid at saturation pressure and reservoir or test temperature, Ibm/cu ft kg/m?) cubic EOS parameters for Component i 95,05; = critical alculated, component surface separation or flash 8 = gas i, j = component number yydrocarbon liquid o=oil References 1. Katz, DLL, and Ftoozabadi, A: “Predicting Phase Behavior of Condenste/Crude-Oi1 Systems Using Methane Tateraction octets," J. Pet. Tock, (Nov. 1978) 1689-85; Trans. AIME, 2s 2, Firoorabadh, A, Heim, Y., and Kat, D.L.: "Reservoir Depletion Calelations for Gas Condensates Using Extended Analyses in the Peng Robimon Equation of Sate." Can. J.Chem. Eng. (1978) 6, 610-15. 3, Yarborough, Lyman: “Application of « Generalized Eguationof| State o Petroleum Reservar Fluids," Equations of State in Engi: neering, Advances in Chemisty Series, K.C. Chao and RL Robinson (eds), American Chemical Society, Washington, DC (4979 TaD, 385-435. 4, Baker LE and Laks, K.D.: "Chit Pin and Sasration Pressure (Cleiations for Malicomponent Systems,” So. Pet. Eng. J. (Fe 1980) 15-2, 5. Vogel, IL. and Yarborough, L.: “The Effect of Nitrogen on the Phase Behavior and Physical Properties of Reservoir Flu,” paper SSPE 815 preened athe 1980 SPE/DOE Enhanced Oi Recovery Symposium, Tulsa, April 20-23 6. Whitin, CH. "Characteriang Hyrocatbon Ps Fractions,” Se. Pet Eng. J. (Aug. 1983) 683-98 7. Whitson, CH. and Torp, 8.8: “Evaluating Constant-Volume Depletion Data" J. Pet Tech. (Mar. 1983) 610-20. A “Application of the Devel. oped. Redlich-Kwong Equation of ‘State to. Predict the ‘Thermodynamic Properties of Condensate Gass," paper SPE 8287 presented atthe 1979 SPE Annual Technical Conference and Exhibition, Las Vepa, Sep. 23-26 29. Wiliams, C.A., Zana, E.N, and Humphrys, GB: “Use ofthe eng-Robinson Equation of Sate to Predict Hydrocarbon Phase [Behavior and Mise for Fant Dipacement,” paper SPE 8817 presented at the 1980 SPE/DOE Eahanced ‘Oil Recovery Symposium, Tals, Apri 20-23, 10, Coats KH: 'Simlaton of Gas Condensate Reservoir Perform ance." J, Pet. Tech. (Ost. 1988) 1870-86 11, Redlich, 0. aod Kwong, 1.NS.: On the Thermodynamics of Solution V. An Equation of State, Fupactis of Gaseous Salut,” Chem. Review (1949) 44, 233, 12 Soave, G.: "Equlbriam Constants From « Modified Redich- Kwong Equation of Site." Chem Big. Se, (1972) 27, 1197-1208, 13, Zadkevite, D, and Joffe, 3: “Comelation abd Preditions of Vapor Liguid Equilibria withthe Redlch-Kwong Equation of State,” AICHE J.'Gan. 1970) 16, 112-19, 14, Joe, ., Schroeder, G-M., and Zudkeviteh, D: “Vapor Liguié Equilibria wits the Reich Kwong Equation of Sute," AICHE J (May 1970) 16, 496-98, “Y, and Robinson, D-B.: “A New Two-Consant Equation * Ind. Eng. Chem. Pandan ( 16, Lote, J, Bray, B.G., and Clank, CR, of Reservoir Fids From their Compositions. 1968) 1171-76; Trans, AIME, 23 17, Hoffmann, A.E., Crump J.S., and Hocot, C..: “Eqiibriam ‘Constants for Gas Condensate Sytem," J Pet Tech, Gan. 1983) Trans, AIME, 198. 18, Coats, KH., Dempsey, and Henderson, JH.: "A New Tech ‘nique for Determing Reservoir Description from Field Perform: tance Data,” Soc. Pet. Eng J. (March 1970) 66-78 19. Coste, Ki: "Resor Salon: Ste ofthe Ar," J. Pet. Teck (hug, 1982) 1633-42, Appendix—Selection and Range Limits of Regression Variables Our experience with EOS regression includes a wide var- jety of fluid samples and types of laboratory test data. In general, a necessary and sufficient regression variable set has been the five parameters of methane and plus-fraction 9's and methane-plus fraction binary. When a CO, swelling tests part ofthe data, we usul- 1y find it necessary to add CO 9§ tothe variable set and pure COy density (at reservoir temperature and a pert ‘ent pressure) tothe data set. The frther addition ofthe ‘COp-plus fraction binary may or may not prove helpful For Nz/reservir fluid mix data, we have found the basic five-parameter variable set sufficient with no alteration or inclusion of Nz EOS parameters. However, we have more experience with CO reservoir fluid data than with No mix data. One obvious rule in selecting regression variables is ex- clusion of any EOS parameter that, by inspection, can not affect significantly the calculated value of any of the regression data. For example, if compositions of al sam ples in a regression dataset include very small amounts Of some component, then one would not select any ofthat component's 0's or binaries as regression variables. The pragmatic converse ofthis rule is inclusion of one or more EOS parameters for any component that is composition- ally predominant in all or some regression data samples. In many cases, methane satisfies this predominance. ‘TABLE A-1—EFFECT OF PSEUDOIZATION ONC, 23,95, Or EOS 03.0, S PR 0.5205 0.0872 8 ZURK 0.3095 0.0858 3 o.0as8 3 oie 0.3874 0.0848 PR ZuRK 208 The characteristics of a good or optimal regression vari- able set are that the regression converges; the variable values converged upon are realistic; deletion of any mern- ber of the variable set results in either or both of (1) a significantly worse data match and (2) unrealistic varia- ble values; and addition of any other EOS parameter re- sults in either or both of (1) nonconvergence and (2) insignificantly better data match. Nonconvergence can result from redundancy among the variables in the sense thatthe objective function is insen- sitive to values of two or more variables provided they satisfy some relationship to one another. Nonconvergence ccan also result from simple insensitivity of the objective function to one or more of the variables. The symptom of nonconvergence may be either the tailing off toward ‘global limit or the ““bouncing" within a small range of ‘one or more of the variables. In any event, nonconvergence is obviously dependent ‘on the regression data set as well as the variable set. That is, a given variable set yielding nonconvergence may yield quite reasonable convergence with additional regression data. For a fixed regression data set, the remedy for non- ‘convergence is simply removal of one of the regression variables. The response to a convergence with unrealis- tic variable values should be addition ofa regression vari able, as illustrated in the case of Gas 3 of this paper. ‘The occurrence of a poor data match with a regression variable set that obeys the previously mentioned charac- teristics of a good regression variable set, with or without realistic convergence, indicates either erroneous data or inadequacy of the EOS. In some cases, suspect data can bbe detected by simple mass balances on CVE data and/or surface separation results. In one lean retrograde-gas- condensate case, the laboratory data included liquid dropout V/V values for a CCE about three times larger than the CVE V/V values. All “‘good” gas-condensate data we have seen exhibit CVE fy, values larger than CCE fi, values. A mass balance on that particular con- densaie’s CVE gave liquid gravities ranging from 22 near dewpoint to 18 at lower pressure. Omission of the CVE ‘fz Values from the regression data set gave quite reasona- ‘ble EOS parameter values and a good data match, except that calculated (more correct) CVE fi, Values were about five times larger than reported ‘The above discussion gives no rationalization for ac~ cepting or allowing alteration of EOS parameter theoret- ical values. The theoretical @5 and % values in cubic EOS arise from the required satisfaction of the van der Waals conditions of dp/dV=d2 p/dV* =0 at the critical point. The component temperature functions in the SRK and PR EOS and the altered (temperature-dependent) ZIRK component 2° values essentially reflect satisfac- tion of pure-component density and vapor-pressure data below critical temperature. At reservoir conditions, ‘methane in particular is well above its critical point and there is no theory or clear-cut guide to selection or alter- ation of 0s for components well above their critical tem- perature. One might argue pragmatically that the theoretical methane 0 and 03 values satisfying the van der Waals conditions at p and T far removed from our range of interest do not satisfy the requirement of correct ‘methane density at the reservoir p and T conditions that are of interest. Pursuing this observation leads to the sug- ‘gestion that methane 3, 9% be determined at reservoir SPE Reservoir Engineering, May 1986temperature by requiring exact satisfaction of methane densities at that temperature and two pertinent pressures. ‘Annext step is pinning only the relationship between the two 1s by satisfying experimental density at one pres- sure and regressing on one 2. We have done this with the CO, 0s in connection wit a swelling test match'® and found that the resulting CO» 0% and 0 values yield ed excellent agreement with pure CO density over a wide range of pressures ‘Arguments in favor of accepting altered plus-fraction {0° values basically reflect the simple fact that, unlike al other components, that fraction isa mixture of many com ponents. One argument for accepting altered values of 05,.,.03,, can be based on the results of pseudoizing or Jumping an extended analysis to a C7. fraction. A pseu- doization procedure" was applied to the Oil S and Oil 3 extended analyses. The Oil 5 F7 through Foo. frac- tions and Oil 3.Cy through Cio,” fractions were each SPE Reservoir Engincering, May 1986 lumped into single C7. fractions, using both EOS. Ta- ble 18 lists the resulting C7, 95 and Q§, values. SI Metric Conversion Factors “API 1 ° ak " tonen tl mt ste 141.5/(131.5+°APD, E-01 = glem? x 1.589 873 ° x Lor X 2.831 685 X 1,638 706 (CF=32)/1.8 x 1.601 846 X 6.894 757 x 1.781 073 stock-tank m? SPERE ‘sgn mans he Sait of Pau Engen tee A. 27 {S22 ace sop puan Fo 8 08 Rares aero coe ig oa Bape GPe 1 rps tbe 1984 Arun at Teste Co ‘re te Eaton ho Now Ores Sant 22 299