Chem. Eng. Comm. 1990, Vol. 96, pp. 259-278 Reprints available directly from the publisher. Photocopying permitted by license only. © 1990 Gordon and Breach Science Publishers S.A. Printed in the United States of America APPLICATION OF THE GAMMA DISTRIBUTION MODEL TO MOLECULAR WEIGHT AND BOILING POINT DATA FOR PETROLEUM FRACTIONS CURTIS H. WHITSON Department of Petroleum Engineering Norwegian Institute of Technology 7034 Trondheim-NTH Norway THOMAS F. ANDERSON Department of Chemical Engineering University of Connecticut, U-139 Storrs, CT 06268 INGOLF S@REIDE Department of Petroleum Engineering Norwegian Institute of Technology 7034 Trondheim-NTH Norway (Received August 4, 1988; in final form August 3, 1988) ‘This paper applies the gamma distribution model for describing both the molar and boiling point distributions of heptanes-plus (C;,) fractions. The three-parameter distribution model has been fit to ‘TBP (true boiling point) data from forty-four samples of stabilized petroleum liquid (stock tank oil and condensate), obtained from separation of reservoir fluids. An excellent fit was achieved for both ‘molar and boiling point distributions, though molar distribution seems to be more accurately described by the distribution model. ‘Two of the parameters, a and 9, were correlated and fit to empirical equations for both molar and boiling point distributions. Since the third parameter, B, is defined in terms of a, n, and either average molecular weight or boiling point, it appears that a generalized correlation for molar and boiling point distributions may exist. We have not developed such correlations in this work, but our results provide the necessary groundwork for further research. KEYWORDS — Gamma distribution Molecular weight Boiling point Petroleum fractions. Presented at the 1986 AIChE Annual Meeting, Miami, Florida, November 2-7. 259260 C.H. WHITSON et al. INTRODUCTION Petroleum reservoir fluids are complex mixtures made up of light compounds such a Nz, CO;, HS, Cy, and C2, intermediate hydrocarbons C; to C,, and heavier components with varying content of paraffinic, naphthenic, and aromatic ‘compounds. Proper characterization of the heavier or “heptanes-plus” (C;.) components is important when cubic equations of state (EOSs) are used to describe the complex phase behavior of reservoir fluids. The basis for most characterization methods is TBP data that includes mass, mole, and volume fractions of distillation cuts with measured molecular weight, specific gravity, and boiling point. Each distillation cut may be considered as a pseudocomponent with a critical pressure, critical temperature, and acentric factor. Correlations for estimating pseudocomponent critical properties are usually based on specific gravity and boiling point. ‘An alternative to characterizing the heptanes-plus fraction as a series of pseudocomponents is to use a probability model that expresses mole fraction as a continuous function of molecular weight. There are several advantages to using the continuous approach. First, the residue or heaviest C,, cut can be readily divided into discrete fractions by extending the distribution to molecular weights greater than can be measured directly (Whitson, 1983; Whitson and Torp, 1983). Another advantage is the definition of C;, into a series of pseudocomponents based on statistical reasoning and the model parameters used to describe the molar distribution (Behrens and Sandler, 1986). A continuous description can also be used directly in the EOS (Briano and Glandt, 1984; Cotterman and Prausnitz, 1985; Cotterman et al., 1985). To describe C;, a8 a continuous distribution requires that the model parameters be fit to experimental TBP data. This problem has received little attention in the literature and it is our purpose to introduce a procedure for fitting molar distribution by minimizing the sum of squares between experimental and model molecular weights (Whitson, 1983b). We also consider boiling point as a continuous distribution in terms of weight fraction and a similar procedure for fitting the distribution is proposed. More than forty samples have been described by the gamma distribution model using molecular weight and boiling point data from TBP analyses. All three parameters in the gamma distribution model are determined simultaneously to give a best fit of experimental data. The fit of model parameters is to some extent dependent on the number and boiling-point range of distillation cuts. We checked the stability of best-fit parameters for each sample by varying the number of fractions from six to the actual number of distillation cuts. The behavior of best-fit model parameters was monitored and the stability was determined for each sample. Finally, we found that two parameters of the model, a and 7, are highly correlated for both molar and boiling point distributions. Empirical relations for @ as a function of 1 are given for molar and boiling point distributions. This essentially reduces the number of model parameters to one. It also indicates that there may exist a general correlation for all three model parameters, a, n, and B in terms of C;, properties.PETROLEUM FRACTIONS 261 DISTRIBUTION MODEL The three-parameter gamma function (Type 3 of the Pearson system; Pearson, 1895) is used to describe the molar and boiling point distributions, =n)" expl = («= n)/B] 2) te @) where p(x) is the probability density function. Parameter a defines the form of the distribution, 7 is the minimum value of x with a non-zero probability of occurrence, and f is a composite parameter defined in terms of a, 1, and the average (or mean) value Of x, Xagy ave = OB +7) @ Molecular weight, M, is defined as the variable for molar distribution (x = M) described by parameters ay, Nw, and By. Boiling point, 7,, is defined as the variable for boiling point distribution (x = T,) described by parameters a7, nr, and By. Figure 1 shows the probability density function for molar distribution of two oil samples (Nos. 44 and 2 in Table I). Figure 2 shows the probability density function for boiling point distribution of an oil sample (No. 44) and a condensate sample (No. 28). Model parameters for the different samples are noted in the figures. 0.014 ape = 93 nse we (FE fie 0.012 o.o10 0.008 + 0.006 F 0.004 PROBABILITY DENSITY FUNCTION, p(M) 0.002 9.000 100 200 300 400 500 600 700 MOLECULAR WEIGHT, M FIGURE 1 Probability density function vs. molecular weight for oil samples No. 2 (dashed line) and No. 44 (solid line). °262 C.H. WHITSON et al. Cumulative probability, P(x), represents the area under the p(x) curve from 7 to an upper boundary value, x, Pa)= [pe ae °% which can be shown to equal Pay" Spirle +j+D) ® with y= (¢—9)/B. As x approaches « the area sums to unity The average property, x,y, in the interval x,-; to x, is expressed x, “FTP, Pee ~ [90a] © which can be shown to equal, ©) Fieve = The function P(x) is evaluated by starting the summation in Eq. (4) at j= instead of j= 0. MOLAR DISTRIBUTION For molar distribution, we define P(x) as cumulative normalized mole fraction, a where M; is the upper bound of molecular weight for distillation cut i, and 2, is the mole fraction of cut j, where j=1,...,N and N is the total number of C;, fractions. For the last fraction, My=°. Normalized mole fraction, z,, for an individual cut is given by, 2 = P(M;) ~ P(M,-1) (8) Average molecular weight, M,, of cut iis given by Eq. (6), PM) — Al 1) P(M) — P(M.-1) Normalized weight fraction can be expressed as iM, I(nw + amBa) (10) M= 1+ auBu 9) my,PETROLEUM FRACTIONS, 263 0.006 r 7 T T 7 ¥ 0.005 0.004 0.003 0.002 PROBABILITY DENSITY FUNCTION, p(Tb) 0.001 ° 00 500 700 900 1100 1300 1500 1700 BOILING POINT, Tb,°R FIGURE 2 Probability density function vs. boiling point for condensate sample No. 28 (soli line) and oil sample No. 44 (dashed line). Experimental TBP data is fit to the molar distribution mode! using measured weight fractions and molecular weights. The procedure is outlined below: 1. Calculate experimental normalized weight fraction for each distillation cut (iM exp x ay ECM eo Wiexp This calculation is only performed once since results can be stored. 2. Assume distribution parameters ay, Ny, and By. For the first guess, assume ny =88, a =1.0, and By estimated from Eq. (2) [By = (My — Nw)/ Gy) using experimental M,... 3. Assume an upper molecular weight boundary, M,, for the fraction. Calculate P(M;) and 2; from the distribution model, Eqs. (4) and (8). Calculate average molecular weight, M,, and normalized weight fraction, w,, from Eqs. (9) and (10). 4. If model weight fraction does not equal experimental weight fraction within an acceptable tolerance (e.g., 10’), adjust the upper molecular weight bound, M,, and return to step 3, A Newton or chord method can be used to solve the trial and error problem. 5. Repeat the trial and error procedure in steps 3 and 4 for all fractions except264 C.H. WHITSON et al. the last one. Calculate the sum of squares function, Exo Barta) = 2 (a2) Mio where Myexp and Mjmoui a€ experimental and model molecular weights, respectively. 6. Minimize Ey, by adjustment of model parameters ay, mar, and By using a non-linear regression model. 7. Calculate model average molecular weight using Eq. (2) and compare with the experimental value. We generally find the model value is within a few percent of the measured value. 8. Compare model molecular weight and mole fraction of each fraction with experimental values to establish the accuracy of the model fit. BOILING POINT DISTRIBUTION For the boiling point distribution model we define cumulative probability as cumulative normalized weight fraction, P(Tu) (13) z exe ‘The normalized weight fraction of a cut determined from the distribution model is then, w= P(Tos) — P(Toi-1) (a4) Weight-average 7,, is calculated from Eq. (6), AC Ty = Or + arBr (15) PCTs The procedure for matching model parameters a, 7, and By to experimental TBP data is outlined below: 1. Calculate experimental normalized weight fraction, W,exp. for each fraction from Eq. (11). This calculation is only performed once since results can be stored. 2. Assume distribution parameters ay, nr, and By. For the first guess, assume Nr =640(°R), a = 1.0, and By estimated from Eq. (2) [Br = (Tox — nr)/ar]. An estimate of J;,. (in °R) can be made using experimental M;, and the approximate relation Tyr = 20.853 M325 (16) This equation was developed using model values of Tj», and M;, from the three-parameter fits of the forty-four samples. 3. Assume an upper boiling point boundary, T,,, for the fraction. Calculate (Ty) and w, ftom the distribution model, Eqs. (4) and (14). Calculate average boiling point, 7,,, from Eq. (15).PETROLEUM FRACTIONS 265 4. If model weight fraction does not equal experimental weight fraction within an acceptable tolerance (e.g., 10°’), adjust the upper boiling point bound, Ty, and return to step 3. 5. Repeat the trial and error procedure in steps 3 and 4 for all fractions except the last one. Calculate the sum of squares function, Ex(arBrine)= 5, | st Hts] 7) Te where Tye and Ty,mose: ate experimental and model boiling points, respectively. 6. Minimize Ey by adjustment of model parameters ay, 7, and B; using a non-linear regression model. 7. Compare the model boiling point of each fraction with the experimental value to establish the accuracy of the model fit. APPLICATION OF THE DISTRIBUTION MODELS The distribution models for molecular weight and boiling point have been applied to 44 petroleum samples having complete TBP analyses. The samples represent oil types ranging from light condensate to heavy oil. The number of distillation cuts ranges from six to 26, and most samples are distilled according to an ASTM standard with boiling point intervals of normal paraffins (Katz and Firoozabadi, 1978), Table I summarizes the average properties of each sample. Model parameters determined by the three-parameter regressions are also listed. In general the distributions give an excellent fit of the experimental data. Figures 3 and 4 plot model distributions of an oil and gas condensate (sample Nos. 44 and 28), respectively. The absolute average residual, AAR, is used to quantify the goodness of the model fit, Wap 3 | Mint 1 oN Wap & Tene where the last C;, fraction is not included in AAR values given in Table I. For oil sample No. 44, AARy=1 kg/mol and the difference between experimental and model average molecular weight is about 2.5%: Mr4.eqq = 233 and My, moiei= 227. Boiling point distribution is also represented reasonably well, as indicated by an AAR, = 5°R. The gas condensate sample No, 28 also has an AARy = 1kg/mol for the molecular weight distribution, with experimental and model average molecular weights essentially equal. The boiling point model fit has an AAR; =2°R. The boiling point distribution fit for Hoffman et al.’s (1953) oil sample did not converge. It appears that mr tends to zero and a tends to infinity, indicating a normal distribution. This is seen in Figure 5, where p(T,) is plotted for a7 = 23.5 AARy= Miexp | (a8) AARr= icexp | (a9)C.H. WHITSON et al. 266 ss 16:¥6 a sero coe Om Soa wsiepeg 81 st ace a iso Tut Od Od wosiapeg ait 1076 a aso ofS? Od £08 Uosi9p2 st 9°16 a sso os Od 7-08 wosi9p24 sz 1696 a sro G0IZ Od 1-08 uosiopad, a8 s¥'66 s sero 19st Od Taio, sot 86°78 s ssos'0 BUST COW 2 aidues puepeeyy sot ove a zi ss02 Od sot ae s wu Cust OF V aidwes pueweert ot Sz a6 s woe’) SULT 399 me peRNY DRL 6 s o€ 876 a srrs0 SLIZ OF ages pur nue nt Iu a oat THFE OF ember L at ra s soso 790 Of Lp 9 nou on s 190650 ye Tmirant § as 116 a 20600 Lot 3D. om kaart a ore a “wo sw 3D £ Sot rst ore’) LR6L Od z a+ we a aero Ss OF avy Supues LEAVY +oMW oa eudiy [Ld AVY Heal wa eudy my Aye yom adh) “ataiduws “oN aysi9M IO|MDOIOW uiod sores aysods“JOW aiduies —— ot soreuresed- on, ey wos ep “sq snyd-soutedoy Pounseoyy 14 sorouered-2axqp 403 saqdwes yo pure axesuapu0d 40} latavie jouwsed uonnguisip pur267 jerop oe Ze-6 sayduies vag yUON “2amez0 ‘2y1 woxy waxes are sojdures 19410 “ssHoar9so1 ayesuapU0D se8 pul ynoojou Laqnecr-izery uo paseq ,,uosyaA £q UORENba wos) 10 stuodas 10 soouas9}24 paysgad. PETROLEUM FRACTIONS nuDnnddDunnnnnesnnenazaaa 1'€6 918 90°98 096 88't6 116 #826 Ize I-08 1186, 966, St% 81°96 £9°96 8€76 8C98 SL96 95"16 us £896 07°96, wees, 1ee6 Iss apDDauawvav2DD5DDDDDMDDDMDD 668s 61g £a9 cz6s sug 706s F109 ros: * 8265 168s, ess 8669 Lse9 08s: yor 6969 8665 Le ost. sas o'sis 8's19 e219 8'P65 826s ous z38'1 a9 06°T sit eve 608" 66"1 sez ert 680 sro 280 968° tree sept 80 stoz were ze coat RELL 008" oe e580 szo8'0 1098'0 #5580 Lise e080 e080 180 ‘8008-0 soos" 6580 onc 0080 1ozs'0 oso zis0 t09'0 z96L'0 rIe0 ‘crse'0 shs'0 ‘srre'0 ros'0 ‘3598°0 9998'0 $698"0 og od o4 08 og 99 D9 D9 DD 59 oa 29 55 39 59. D9. 59 D9 Do oa oa og oa oa oa oa TON 380% YON *SUPIEL EON SUPIAL TON *SUPIE TON SUPE 9:99 wasiapad OD wossapod #39 wosapeg. OM U2si9pad Od uasiapag 6-08 wasiapad $081 wasiopad COM u>s19p9q, 9-08 uasiapeg268 C.H. WHITSON et al. D0 ey 600) ‘200 > & 400 = Sstoo ot 2 $1000 soo F 2 900 3 2 ao 8 800 2 — 700 Pont 800020040 060 080 100 CUMULATIVE WEIGHT FRACTION FIGURE 3 Fit of model distribution parameters to experimental boing point and molecular weight data for oil sample No. #4 ee 300 rro0 Het a 250 Sieger" | 1000 200 = 3 900 150 5 3 800 100 g 700 50 609 o G00 (020° 640-060 080 1.00 CUMULATIVE WEIGHT FRACTION FIGURE 4 Fit of mode! distributions to experimental boiling point and molecular weight data for condensate sample No. 28PETROLEUM FRACTIONS 269 EEE] me Uera a tare 00600: 800-«1000 1200 +1400 1600 BOILING POINT, Tb, °R FIGURE 5_ Probability density function vs. boiling point for oil samples No. 2 (solid line) and No. 44 (dashed line). The fit to boiling point data for sample No. 2 did not converge. and nr = 57°R which is the largest a value obtained before numerical problems in the calculation algorithm occurred; a plot of p(T,) for the best fit of oil sample No. 44 is shown for comparison. Analysis of model parameters for the 44 samples indicates that a and 1 are strongly correlated, as shown in Figures 6 and 7 for boiling point and molar distributions, respectively. Empirical relations describing the approximate rela- tion of a and 7 are proposed, 1 658.0] 1-45 (20) 14+ ae ar 1 4.043 1+ ops tw = 110.0] 1 (21) ‘These equations were then used and a two-parameter fit of the experimental data was performed. Parameters a and 6 were fit, where 7 was correlated to a by Eqs. (20) and (21). Table I summarizes results of the two-parameter model fits. It is seen that some loss in accuracy results from using the empirical relations, but the model fits are still very good.600 500 BOILING POINT PARAMETER ETA 4og La ad 0.0 2.0 4.0 BOILING POINT PARAMETER ALPHA FIGURE 6 Boiling point distribution parameter correlation and best-fit results for 43 petroleum samples (No. 2 not included). 110 MOLECULAR WEIGHT PARAMETER ETA es © » § 3¢§ 2 8 "30 0.5 1.0 15 2.0 25 MOLECULAR WEIGHT PARAMETER ALPHA FIGURE 7 Molecular weight cstibution parameter correlation and best rests for 4 petroleum samples,am PETROLEUM FRACTIONS. eo gH | oF tL es9s'0 oe oa vt 0M wosiapad z sue poco | 6 91 os OzET oa vt LOM wosi9pad 8 rescore | oe 96 $6980 L'B9 oa a SOM uasiapad sre 9960 | 6 s sev Zee oa a S-O8 wasi9pad 1 zz 190 | oF 88 uso PLT o8 a 08 wosi9pad zoe toe | 8 a w80 OFT og a £08 wasiapad Z sie 690 | 6 98 9ESeO — O9ee og vt TOM wasiapad z ruc evo | a sre0 «GOT od vr 1-0@ vosrapaq 5 sor tos |S a sie 987 og 3 Laing. zoe eo | 9 a ssoe'o BUST 08 st 2 aides puepeeyy Zz wsoz ovo | s +8 zo 8'S07 08 st @ aidwes purer Zz vac sor |S a weueo UST 0 3 ‘Vv aidwes purjeepy € rer caso | 9 00" eso SLUT 99. st po 2 “ueRny “RRL e cue so | oe 06 gereO | SLIT oa st ‘eg pue ney z Tose ose | ow o 980 TRE og 9 £2277 Z Voor ex, ¥ sosro 7907 of 9 cp R371 Leese 060 a [ooo 68z od 9 Tp 2991 s ost ero. ve ‘206L'0 Lost 99, L Agog 6 os vie’ 0" Lo STE 39, t Aqooer Z yer osez | — 8 ors L861 oa ¥ 7 2 WeWHOH Sou 990 | ce 1 zero Ree oa L ney Bupyeis yyy UMN tudiy | avy my Aaei8 yom ad suonanay, ‘ay aidues ON, —_ee yore) aynads TOW pany tty aides toe 2E]N22]0 HN ary ON ‘uo!e|21409 e19-eyd wosIP wep 1 soroureed-o%) 1s, smd-sourerday poinseayy ty s9;2uesed-om1 104 sofdutes Yo pur ayesuapuod 10} s1o}ouiesed uonqunsTp pue senusdoid snyd-soueidop mW aTtavs.C.H. WHITSON et al, 272 1 roe arw0 we reso oa wy 1 $90¢ aH 36 9798'0 oa & 1 tue oO +8 10980 oa aw € ssc pew 6 rss8'0 oa w 7 cee (oLe0 e LIss'0 oa oF € 9890 wo eat 29 ot € 9650 2 ‘h08 29 te € uso 8 180 29 a € uso 8 0080 D9. 9 y 290 7 ssons'0 59 se ’ 090T o s0s8'0 oa ve z oso 897 soac'o 99 TON w3pN Ee z oro t o08'0 39 Wosipag —E £ £190 fs 1078"0 D9 990 wasioped—1E z 680 8 iso 29 SOD uasiapad 0 z 8zr'0 us zis0 39 & z 1 a 10900 GBI 39 8 l Isc & 29600 Toe 30 a z our'o ts $8180 Fort 59 19 wesiaped 9 £ ose oreo. a uss Lue oa Zi-OM wasiopad $e b rez 90 se are0 | OEIZ oa 10d wasiapeg $2 Lou ev 8 Sbre0 | SOZ oa 01-08 wosi9pad — €z 7 wz aw vose0 sae oa GOH Ws uy +uMN tudiy avy 7 * Gwe Wylam od a1 ardures ON SO sorry sybeds OW pty ajdures 14am sH]n99)09y tulod Sumo ae Suon}a1109 e12-e4dTe WA, ost, wep - sprouesed-ony, 19g, snyd-sauriday pounseajy, (ponunuos)—1) S1aVI.PETROLEUM FRACTIONS 273 98 MOLECULAR, WEIGHT PARAMETER, ETA 8 g & s 88 0.77 0.79 0.81 0.83 0.85) 0.87 MOLECULAR WEIGHT PARAMETER ALPHA FIGURE 8 Contour plot of the error function for molecular weight distribution for oil sample No. ‘44, Dashed line indicates the general direction of correlation for parameters alpha and eta The strong correlation between a and 7 is also seen from the shape of the error functions, Ey and Ey. Figure 8 shows several contours of Ey plotted on an (a, n) grid for sample No. 44. The model average molecular weight was held constant to eliminate one degree of freedom from the minimization problem, however this did not affect the general shape of the contours. We see a long valley along which several combinations of a and 7 give a reasonable match of the data. This general behavior should be considered when choosing a non-linear regression algorithm to determine model parameters. ‘A good model fit depends to some extent on the number of C;, fractions reported for a sample. We studied this effect for each sample by fitting the three model parameters using fewer and fewer fractions, where the heaviest fraction was modified to maintain correct C;,. properties. Figure 9 shows parameters ay, tw, and model N;,. as a function of the number of distillation cuts included in the match. Unstable behavior is observed for matches with less than 16 fractions for this sample. However, the variation of model parameters is not large, and the values determined using only six fractions do not deviate substantially from the stable parameter values. Table I indicates when parameter estimates using all experimental data for a given sample are stable (S) or unstable (U). Figure 10 shows a plot of ay versus my for the varying number of C,, fractions. The general correlation (Eq. 21) is also drawn on the figure as a solid line. The dashed274 C.H. WHITSON et al. ons ons os 08s ona ALPHA oat 0 a5 ma E zat 228 WOOEL C74 MOLECULAR WeGHT ee 0 eee a0 MOMOER OF HEPTANES PLUS: FRACTIONS. FIGURE 9 Molar distribution parameters resulting from three-parameter fit using various numbers of distillation cuts for oil sample No. 44, line represents the minimum along the valley shown in the error function plot in Figure 8. DISCUSSION Results from this study indicate that the three-parameter gamma distribution model can, with reasonable accuracy, be used to describe molar and boiling point distributions of heptanes-plus fractions. The choice of molar distribution in terms of molecular weight and mole fraction, and boiling point distribution in terms of boiling point and weight fraction gives a direct link between the two distributions. ‘That is, when distribution models for molecular weight and boiling point have been determined, the C;, fraction can be split into discrete pseudocomponentsPETROLEUM FRACTIONS 215 98 + + + : suuoers, navea, 10, oF o7+ AUDES, ROME 804g 7 é Pe BS ee 96 | ] & G se 2 . z . = 4 Soap = & 3 sa 4 3 Poe 3 dq cara! 3 90 28 ' : 1 . 6r7 078 081083 ~08S 0.87 MOLECULAR WEIGHT PARAMETER ALPHA FIGURE 10 Cross correlation plot of molecular weight distribution parameters. using various tuumbers of distillation cuts for sample No. 44. Dashed line refers tothe valley shown in Fig. 8, while the solid line represents the general correlation (Eq. (20). for which the weight and mole fractions, molecular weight, and boiling point are readily calculated. A natural extension of this work would be to accurately estimate specific gravity of a fraction from molecular weight and boiling point distributions. This was tried using both the Riazi-Daubert (1980) and Twu (1984) correlations, but results were inconclusive and they are therefore not presented. We should mention, however, that even using experimental molecular weights and boiling points did not result in accurate specific gravity estimates with the Riazi-Daubert and Twu correlations. Consistent correlation of molecular weight, boiling point, and specific gravity using the proposed distribution models is a natural extension of our present research. CONCLUSIONS ‘We have shown that experimental TBP data can be characterized by molar and boiling point distributions using the three-parameter gamma function. Forty-four C;, samples ranging from light condensate to heavy oil are characterized by the proposed distribution model. General correlations relating two of the distribution216 C.H. WHITSON et al. parameters (a and m) are suggested for molar and boiling point distributions. Ongoing research is aimed at linking specific gravity to the molecular weight and boiling point distributions. ACKNOWLEDGEMENTS This work was supported by research grants from NAVF (Norwegian Research Council for Science and the Humanities) and Saga Petroleum a.s. NONEMCLATURE AAR absolute average residual E error function M molecular weight, kg/mol M upper bound on molecular weight, kg/mol M, average molecular weight of fraction i, ke/mol My average molecular weight of heptanes-plus, kg/mol N number of distillation cuts P(x) cumulative probability function P(x) cumulative probability function pix) probability density function t normal boiling point, °R te upper bound on normal boiling point, °R Tw average normal boiling point of fraction i, °R W normalized weight fraction Sune calculated average property of heptanes-plus x probability function variable (M or T;) x upper bound on variable x y probability function variable ze normalized mole fraction of fraction i Subscripts model model property exp experimental property i distillation cut index M molar distribution 7+ heptanes-plusPETROLEUM FRACTIONS a T boiling point distribution Greek Letters @,B,n parameters in gamma distribution function r gamma function y specific gravity (water = 1) REFERENCES ‘Austad, T,, Hvidsten, J., Norvik, H., and Whitson, C.H., Practical Aspects of Characterizing Petroleum Fluids., presented at the conference on North Sea Condensate Reservoirs and Their Development, London, May 24-25, (1983). Berge, 0., Damp/Vaeske-Likevekier i Réoljer: Karakterisering av Hydrokarbonfraksjoer. MSc Thesis, Norwegian Institute of Technology, Department of Chemical Engineering (1981) Behrens, R.A., and Sandler, S.1., The Use of Semicontinuous Description to Model the C, Fraction in Equation of State Calculations. Paper SPE/DOE 14925, presented at the Fifth Symposium ‘on Enhanced Oil Recovery, Tulsa, OK, April 20-23 (1986) Briano, J.G., and Glandt, E,D., Statistical Thermodynamics of Polydisperse Fluids. J. Chem. Phys., 0, 3336 (1984). Cotterman, R.L... and Prausnitz, J.M., Flash Calculations for Continuous or Semicontinuous Mixtures Using an Equation of State. Ind. and Eng. Chem. Process Design and Dev, 24, 434 (1985). Cotterman, R.L., Bender, R., and Prausnitz, 1.M., Phase Equilibria for Mixtures Including Very Many Components. Development and Application of Continuous Thermodynamics for Chemical Process Design. Ind. and Eng. Chem. Process Design and Dev., 24, 194 (1985). Hiaaland, S., Characterization of North Sea Crude Oils and Petroleum Fractions. MSc Thesis, Norwegian Institute of Technology (NTH), Department of Petroleum Engineering (1981), Hariu, O.H., and Sage, R.C., Crude Split Figured by Computer. Hydro, Proc., 48, 143-148 (1969). Hofiman, A.E., Crump, JS., and Hocott, C.R., Equilibrium Constants for a Gas-Condensate ‘System. Trans., AIME, 198, 1-10 (1953). Jacoby, R.H., and Berry, VJ., A Method for Predicting Pressure Maintenance Performance for Reservoirs Producing Volatile Crude Oil. Trars., AIME, 213, 59-64 (1958) Jacoby, RH., Koeller, R.C., and Berry, V.J.. Effect of Composition and Temperature on Phase Behavior and Depletion Performance of Rich Gas Condensate Systems. Trans., AIME, 216, 406-411 (1959) Katz, D.L., and Firoozabadi, A.. Predicting Phase Behavior of Condensate/Crude-Oil Systems Using Methane Interaction Coefficients. J. Pet. Tech., Nov. 1978; Trans., AIME, 265, 1649-55 (1978). Lee, $.T., Jacoby, R.H., Chen, W.H., and Culham, W-E., Experimental and Theoretical Studies on ‘the Fluid Properties Required for Simulation of Thermal Processes. Paper SPE 8398 presented at the Soc. Pet. Eng. 54th Annual Fall Technical Conference and Exhibition, Las Vegas, Sept 23-26, 1979 Pearson, K.. Contributions to the Mathematical Theory of Evolution, Il. Skew Varia Homogeneous Material. Philosophical Trans. Royal Society of London, Series A, 186, (1895), Pedersen, K'S., Thomassen, P., and Fredenslund, A., Thermodynamics of Petroleum Mixtures Containing Heavy Hydrocarbons. 1. Phase Envelope Calculations by Use of the Soave- Redlich-Kwong Equation of State. Ind. Eng. Chem. Process Des. Dev., 23, 163-170 (1984a) Pedersen, K.S., Thomassen, P., and Fredenslund, ., Thermodynamics of Petroleum Mixtures Containing Heavy Hydrocarbons. 2. Flash and PVT Calculations with the SRK Equation of State, Ind. Eng. Chem. Process Des. Dev., 23, 566~573 (1984b). Pedersen, K.S., Fredenslund, A., Christensen, P.L., and Thomassen, P., Viscosity of Crude Oils. ‘Chem. Eng. Science, 38, 1011-1016 (1984c), Riazi, M.R., and Daubert. T-E., Simplify Property Predictions. Hydro. Proc., §8, 115=116 (1980). Standing, M:B., and Katz, D.L., Density of Crude Oils Saturated with Natural Gas. Trans., AIME, 146, 159165 (1942).278 C.H. WHITSON et al. ‘Twu, C.H., An Internally Consistent Correlation for Predicting the Critical Properties and Molecular ‘Weights of Petroleum and Coal-Tar Liquids. Fluid Phase Equilibria, 16, 137-150 (1984), Whitson, C.H., Characterizing Hydrocarbon Plus Fractions. Soc. Pet. Eng. J.; Trans., AIME, 278, (683-694 (1983a). Whitson, C.H., Topics on Phase Behavior and Flow of Petroleum Reservoir Fluids. Dr. Techn. thesis, ‘Norwegian Institute of Technology, Department of Petroleum Engineering (1983b).. Whitson, CH. and Torp, S.B., Evaluating Constant Volume Depletion Data. J. Pet. Tech., March 1983; Trans., AIME, 218, 610-620 (1983).