Energy Sources, Part A, 29:12331242, 2007

Copyright Taylor & Francis Group, LLC

ISSN: 1556-7036 print/1556-7230 online
DOI: 10.1080/00908310500434481
Prediction of CO
2
Solubility in Oil and the Effects
on the Oil Physical Properties
M. K. EMERA
Petroleum and Natural Gas Department
King Saud University
Riyadh, KSA
H. K. SARMA
Australian School of Petroleum (ASP)
The University of Adelaide
Adelaide, SA, Australia
Abstract CO
2
solubility in oil is a key parameter in CO
2
ooding process. It results
in oil swelling, increased oil density, and decreased oil viscosity. Laboratory studies
needed to cover a wider range of data, and are time consuming, costly, and may be not
available or possible in many situations. On the other hand, although various models
and correlations are useful in certain situations, they may are not be applicable in
many situations.
In this study, a new genetic algorithm- (GA)-based technique has been used to
develop more reliable correlations to predict CO
2
solubility, oil swelling factor (SF),
CO
2
-oil density, and viscosity of CO
2
-oil mixtures. Based on the Darwinian theory,
the GA technique mimics some of the natural process mechanisms. Furthermore, GA-
based model correlations recognize all the major parameters that affect each physical
property and also well address the effects of CO
2
liquefaction pressure.
Genetic algorithm-based correlations have been successfully validated with pub-
lished experimental data. In addition, a comparison of these correlations has been
made against widely used correlations in the literature. It has been noted that the
GA-based correlations yield more accurate predictions with lower errors than all
other correlations tested. Furthermore, unlike other correlations that are applicable
to limited data ranges and conditions, GA-based correlations have been validated
over a wider range of data.
Keywords CO
2
ooding, CO
2
solubility, GA, oil physical properties
1. Introduction
A study of changes in physical properties of the oil during CO
2
injection is critical to
develop an understanding of the process performance and the effects of CO
2
on the
process recovery efciency.
Both laboratory and simulation studies are conducted to study the CO
2
ooding
performance. To study the CO
2
ooding performance over a wider range of data and
Address correspondence to Dr. M. K. Emera, Petroleum and Natural Gas Department,
King Saud University, P.O. Box 800, Riyadh 11421, Kingdom of Saudi Arabia. E-mail:
mohammedemera@hotmail.com or memera@ksu.edu.sa
1233
1234 M. K. Emera and H. K. Sarma
conditions, often models and correlations are used to correlate the physical properties that
are not readily available over a wide range of conditions and with reasonable accuracy.
The objective of this research is to obtain a more reliable prediction of CO
2
solubility,
oil swelling factor (SF), and CO
2
-oil density and viscosity. A new genetic algorithm-
(GA)-based technique is used to develop these correlations based on a GA software
developed in our earlier work (Emera and Sarma, 2005). This software has been modied
to suit our objective for this study, and Figure 1 provides a owchart of the algorithm.
Figure 1. Flowchart of the genetic algorithm software design.
Prediction of CO
2
Solubility 1235
2. Factors Affecting CO
2
-oil Physical Properties
The commonly used correlations of CO
2
-oil physical properties that we tested are sum-
marized in Table 1. Also, a brief overview of CO
2
-oil properties follows.
2.1. CO
2
Solubility
The main factors that affect CO
2
solubility are: saturation pressure, temperature, and oil
gravity (

API) (Simon and Graue, 1965). Solubility increases with pressure and oil API
gravity, while its value decreases with temperature. The other factors that affect CO
2
solubility are oil composition and liquefaction pressure at temperatures less than CO
2
critical temperature (T
c;CO
2
D 88

F); that is, when CO

2
is subcritical which is soluble
in the crude oil as a gas rather than as a liquid (Dyer and Farouq Ali, 1989).
2.2. Oil SF
Oil SF is dened as the ratio of CO
2
-saturated oil volume at the system temperature and
pressure to the oil volume at the system temperature and 1 atm (bubble point pressure
(1
b
)). The SF is mainly a function of CO
2
solubility (Simon and Graue, 1965). Besides
CO
2
solubility, the SF depends also on the molecular size of oil molecules (Simon and
Graue, 1965).
2.3. CO
2
-oil Density
The effect of CO
2
solubility on oil density is generally small, and is more pronounced
in lighter oils than in heavier oils (Grigg, 1995). DeRuiter et al. (1994) also found that
the oil exhibits an increased density due to CO
2
solubility increase; however, they stated
that this density increase might be the result of an increase in the system pressure.
2.4. CO
2
-oil Viscosity
Oil viscosity decreases dramatically with increasing CO
2
solubility, resulting in increasing
oil mobility causing an increase in oil recovery. At temperatures less than T
c;CO
2
, CO
2
-oil
viscosity decreases with the saturation pressure up to liquefaction pressure, then levels
off and decreases slightly at pressures higher than the liquefaction pressure. At higher
pressures, the viscosity starts to increase again because of the effect of the pressure
and oil compressibility (DeRuiter et al., 1994). It is also noted that CO
2
-oil viscosity
reduction is higher for more viscous oil than for lower viscous oil (Rojas and Farouq
Ali, 1985).
3. Genetic Algorithm-based Correlations to Predict CO
2
-oil
Physical Properties
1
The following section discusses the GA-based correlations developed in this study to
predict CO
2
-oil physical properties, notably CO
2
solubility, oil SF, and CO
2
-oil density
and viscosity.
1
The database used in this study is available on request from the corresponding author.
1236 M. K. Emera and H. K. Sarma
Table 1
Tested correlations for CO
2
solubility, oil SF, and CO
2
-oil density and viscosity
Authors Correlation and remarks
1. CO
2
-oil solubility
Welker and
Dunlop
(1963)
Graphical correlation. Function of: P
s
and oil gravity (

API) at T D80

F.
Limited to: 20

API oil gravity 40

API.
Simon and
Graue
(1965)
Graphical correlation. Function of CO
2
fugacity and T or P
s
and T at characterization
factor D 11.7, with a correction factor function of the characterization factor and T (for
characterization factors other than 11.7).
Limited to: 110

F T 250

F, P
s
2,300 psia, and 12

API oil gravity 33

API.
Mehrotra and
Svrcek
(1982)
Sol (scf/bbl) D C
1
CC
2
P
s
CC
3
.
Ps
b
1
TCb
2
/ CC
4
.
Ps
b
1
TCb
2
/
2
, where C
1
D 0:04127,
C
2
D0:57292, C
3
D248:9543, C
4
D 1:327846, b
1
D 0:55556, b
2
D 255:3722.
Solubility at 1 atm .P
b
/ D0.
Limited to: P
s
925 psia, 75

F T 207

F.
Chung et al.
(1986)
For P 3,000 psia: Sol (scf/bbl) D
1
C
1

C
2 T
C
7CC
3
T
C
4 expC
5
Ps
C
6
Ps

where
C
1
D0:4934 10
2
, C
2
D4:0928, C
3
D0:571 10
6
, C
4
D 1:6428,
C
5
D0:6763 10
3
, C
6
D781:334, C
7
D 0:2499.
2. Oil SF
Welker and
Dunlop
(1963)
SF D 1:0 C
0.35 (Sol (scf/bbl))
1,000
.
Developed for oils at T D 80

F and 20

API oil gravity 40

API.
Simon and
Graue
(1965)
Graphical correlation. Function of CO
2
solubility, oil MW, and oil density at 60

F. Presents
lower accuracy at high pressures.
Limited to: 110

F T 250

F, P
s
2,300 psia, and 12

API oil gravity 33

API.
3. CO
2
-oil density
Quail et al.
(1988)
D .C
1
c
2
T CC
3
P
s
/
exp.C
4
Sol (mole fraction))
.1CC
5
.C
4
;mole fraction//
, where (for Senlac region oil):
C
1
D0:99, C
2
D0:00036, C
3
D 5:51E-06, C
4
D 35:8E04, C
5
D50:86E-03.
Limited to: P
s
2,466 psia.
4. CO
2
-oil viscosity
Welker and
Dunlop
(1963)
Graphical correlation. Function of
i
and P
s
.
Limited to: T D80

F, P
s
800 psia, and 4 cP viscosity 5,000 cP.
Simon and
Graue
(1965)
Graphical correlation. CO
2
-oil viscosity is a function of P
s
,
i
@ 120

F, T , and CO
2
solubility.
Limited to: 110

F T 250

F, P
s
2,000 psia, and viscosity 1,300 cP.
Beggs and
Robinson
(1975)

d
D 10
X
1, where X D 10
3:03240:02023o
T
1:163
.
For dissolved gas (e.g., CO
2
): DA
B
d
, where A D 10:715 (Sol (scf/bbl) C 100)
0:515
,
B D5:44 (Sol (scf/bbl) C 150)
0:338
.
Neglects oil viscosity dependence on composition and pressure.
Mehrotra and
Svrcek
(1982)
log log. C0:8/ Da
1
Ca
2
T Ca
3
P
s
C
a
4
.b
1
TCb
2
/
P
s
, where a
1
D0:32029,
a
2
D0:002472, a
3
D0:529E-03, a
4
D0:23804, b
1
D 0:55556, b
2
D255:3722.
Developed for saturated bitumen (for: 75

F T 207

F and P
s
925 psia).
Chung et al.
(1986)
For P 3,000 psia: They modied Lederer (1933) equation as follows:
ln.
m
/ Dx
o
ln.
i
/ Cx
s
ln.
s
/, where x
s
D
Vs
VoCVs
, x
o
D1 x
s
,
D0:255
4:16
T
1:85
r

e
7:36
e
7:36.1Pr/
e
7:36
1
, T
r
D .T C460/=547:57, P
r
DP
s
=1,071.
Relationship among x
s
, CO
2
solubility (Sol (scf/bbl)), and SF is: x
s
D
1
.

B.Sol (scf/bbl)
/C1
,
x
s
D
B.SF1
CB.SF1
.
Prediction of CO
2
Solubility 1237
Table 2
A comparison between GA-based CO
2
solubility and other CO
2
solubility literature correlations
Correlation No. of data Average error, % STDEV, % 1
2
GA-based 106 4.0 5.6 0.985
Simon and Graue (1965) 49 5.7 10.8 0.97
Mehrotra and Svrcek (1982) 106 32.6 36.6 0.756
Figure 2. A comparison results among GA-based CO
2
solubility correlation, Simon and Graue
(1965) and Mehrotra and Svrcek (1982), correlations.
3.1. CO
2
Solubility
The GA-based CO
2
-oil solubility correlations are presented as follows:
1. For temperatures greater than T
c;CO
2
(for all pressures) and temperatures less than
T
c;CO
2
(for pressures less than CO
2
liquefaction pressure):
Sol (mole fraction) D 2.238 0.33, C3.235,
0:6474
4.8,
0:25656
(1)
where
, D ;

T
0:8
1
s

exp.
1
MW
/
For this correlation, it is assumed that the CO
2
solubility at 1
b
D 1 atm is equal
to 0.
2. For temperatures less than T
c;CO
2
and pressures greater than CO
2
liquefaction
pressure:
Sol (mole fraction) D 0.033 C1.14, 0.7716,
2
C0.2176,
3
0.02183,
4
(2)
where
, D ;

1
s
1
liq

exp.
T
MW
/
As shown in Table 2 and Figure 2, the GA-based CO
2
solubility correlations offered
a better accuracy compared to other commonly used literature correlations.
1238 M. K. Emera and H. K. Sarma
Table 3
A comparison between GA-based and Simon and Graue (1965) oil swelling correlations
Correlation No. of data Average error, % STDEV, % 1
2
GA-based 85 0.61 0.94 0.994
Simon and Graue (1965) 83 1.0 1.7 0.970
Figure 3. A comparison results between GA-based and Simon and Graue (1965) oil SF correla-
tions.
3.2. Oil SF
For GA-based oil SF correlation, oils are divided based on their MW to oils with MW
300 (heavier oil) and oils with MW < 300 (lighter oil).
1. For oil MW 300:
SF D 1 C0.3302Y 0.8417Y
2
C1.5804Y
3
1.074Y
4
0.0318Y
5
C0.21755Y
6
(3)
where
Y D 1,000.0

;
MW

Sol(mole fraction)
2

exp.

MW
/

2. For oil MW < 300:

SF D 1 C0.48411Y 0.9928Y
2
C1.6019Y
3
1.2773Y
4
C0.48267Y
5
0.06671Y
6
(4)
Table 3 and Figure 3 present a comparison between the GA-based oil SF correlations
and the correlation by Simon and Graue (1965).
3.3. CO
2
-oil Density
The following is the proposed GA-based CO
2
-oil density correlation:
j D j
i
C0.0008, 0.0008,
0:157
(5)
where , D ;j
i
(1
s
1
b
),T and 1
b
D 14.7 psia.
Prediction of CO
2
Solubility 1239
Table 4
A comparison between GA-based and Quail et al. (1988) CO
2
-oil density correlations
Correlation No. of data Average error, % STDEV, % 1
2
GA-based 136 0.29 0.4 0.995
Quail et al. (1988) 129 3.0 4.8 0.293
Figure 4. A comparison results between GA-based and Quail et al. (1988) CO
2
-oil density cor-
relations.
As is evident from Table 4 and Figure 4, this correlation yields lower error than that
in Quail et al. (1988) correlations.
3.4. CO
2
-oil Viscosity
Based on foregoing discussions, the proposed GA-based CO
2
-oil viscosity correlation is
presented as follows:
j D ,j
i
10.8

Sol (mole fraction)

j
i

(6)
where
, D .
0:74
and . D

j
i

1
s
T

0:2
!
(/Sol (mole fraction))
Again, compared to other correlations, the GA-based CO
2
-oil viscosity correlation
yields more accurate results (see Table 5 and Figure 5).
4. Discussion
It is obvious that although the various correlations available in the literature are appli-
cable to certain conditions, they cannot be applied universally. Simon and Graue (1965)
correlations, for example, are limited to 2,300 psia and oil viscosity up to 1,300 cP and
they predict mixture viscosity incorrectly, especially for lighter oils. Chung et al. (1986)
correlations, also, are limited to 3,000 psia. Welker and Dunlop (1963) correlations are
1240 M. K. Emera and H. K. Sarma
Table 5
A comparison between GA-based and other CO
2
-oil viscosity correlations
Correlation No. of data Average error, % STDEV, % 1
2
GA-based 130 6.0 8.8 0.9998
Beggs and Robinson (1975) 130 56.8 62.7 0.8743
Mehrotra and Svrcek (1982) 130 94.3 95.2 0.18
Figure 5. A comparison results between GA-based and Beggs and Robinson (1975) CO
2
-oil vis-
cosity correlations.
limited to 20

API40

API oil for temperature limited to 80

F. Furthermore, the pre-

ceding correlations do not address the effects of the CO
2
liquefaction pressure (at the
specied temperature) well on the solubility, in particular.
Unlike other correlations, GA-based correlations proposed in this study predict CO
2
-
oil physical properties (CO
2
solubility, oil SF, and CO
2
-oil density and viscosity) with
more accuracy and can be applied over a wider range of oil gravities, pressures up to
5,000 psia, oil MW > 490, oil viscosities up to 12,000 cP, and temperatures up to 284

F.
Furthermore, GA-based correlations take into account all major parameter that affect
each physical property.
Genetic algorithm-based correlations have been validated with published experimen-
tal data and available correlations in the literature. Each GA-based correlation presented
in this study has demonstrated superior performance in terms of its accuracy and covers
wider ranges of data. Furthermore, these correlations can be integrated into a reservoir
simulator for CO
2
ooding design and simulation.
5. Conclusions
Based on the preceding discussions, the following conclusions are drawn:
1. Genetic algorithm-based package of correlations developed in this study presents
a more reliable prediction of CO
2
-oil mixture physical properties with higher
accuracy than all other literature correlations tested.
2. Unlike most other correlations, GA-based correlations account for all major pa-
rameters that affect CO
2
-oil physical properties.
3. Besides the better prediction of results with outstanding accuracy, GA-based cor-
relations are capable of covering a wider range of oil properties, with respect
Prediction of CO
2
Solubility 1241
to oil gravities, pressures up to 5,000 psia, oil MW > 490, oil viscosities up to
12,000 cP, and temperatures up to 284

F.
4. Genetic algorithm-based correlations can be easily integrated into a reservoir
simulator for CO
2
ooding design and simulation.
Acknowledgment
The authors thank Santos Limited for its support to research on CO
2
EOR process
within the Center for Improved Petroleum Recovery at the Australian School of Petro-
leum, University of Adelaide. The rst author is a recipient of the Santos Post-Graduate-
Scholarship.
References
Beggs, H. D., and Robinson, J. R. 1975. Estimating the viscosity of crude oil systems. J. Petrol.
Tech. 27:11401141.
Chung, F. T. H., Jones, R. A., and Nguyen, H. T. 1986. Measurements and correlations of the
physical properties of CO
2
-heavy crude oil mixtures. SPE 15080, 56th California Regional
Meeting of the SPE, Oakland, CA, April 24.
DeRuiter, R., Nash, L., and Singletary, M. S. 1994. Solubility and displacement behaviour of a
viscous crude with CO
2
and hydrocarbon gases. SPE Reservoir Eng. May:101106.
Dyer, S. B., and Farouq Ali, S. M. 1989. The potential of the immiscible carbon dioxide ooding
process for the recovery of heavy oil. Third Technical Meeting of the South Saskatchewan
Section, The Petroleum Society of CIM, Regina, Saskatchewan, Canada, September 2527.
Emera, M. K., and Sarma, H. K. 2005. Use of genetic algorithm to estimate CO
2
-oil minimum
miscibility pressure-a key parameter in design of CO
2
miscible ood. J. Petrol. Sci. Eng.
46:3752.
Grigg, R. B. 1995. Dynamic phase composition, density, and viscosity measurements during CO
2
displacement of reservoir oil. SPE 28974, SPE International Symposium on Oil Field Chem-
istry, San Antonio, TX, February 1417.
Lederer, E. L. 1933. Viscosity of mixtures with and without diluents. Proceedings of World
Congress, London, 2:526528.
Mehrotra, A. K., and Svrcek, W. Y. 1982. Correlations for properties of bitumen saturated with
CO
2
, CH
4
and N
2
, and experiments with combustion gas mixtures. J. Canadian Petrol. Tech.
21:95104.
Quail, B., Hill, G. A., and Jha, K. N. 1988. Correlations of viscosity, density, and gas solubility
for Saskatchewan heavy oils. Ind. Eng. Chem. Res. 27:519523.
Rojas, G., and Farouq Ali, S. M. 1985. Dynamics of sub-critical CO
2
/brine oods for heavy oil
recovery. SPE Reservoir Engineering Journal, 3:3544, Feb. 1988.
Simon, R., and Graue, D. 1965. Generalized correlations for predicting solubility, swelling, and
viscosity behavior of CO
2
-crude oil systems. J. Petrol. Tech. 17:102106.
Welker, J. R., and Dunlop, D. D. 1963. Physical properties of carbonated oils. J. Petrol. Tech.
15:873875.
Nomenclature
z mutation factor D 0.95 for CO
2
-oil density and oil swelling factor (SF), D
0.94 for CO
2
-oil viscosity, and D 0.9 for CO
2
solubility correlations
mutation factor equal to: 0.5 for CO
2
-oil density and oil SF, 0.6 for CO
2
-oil
viscosity, and 0.3 for CO
2
solubility correlations
; oil specic gravity (oil density at 60

F)
;
O
oil gravity,

API
1242 M. K. Emera and H. K. Sarma
j CO
2
-oil density, gm/cc
j
i
initial oil density at the specied temperature, gm/cc
empirical parameter presented as a function of density and temperature and
can be obtained by tting the experimental data
0 ratio between CO
2
gas volume at standard conditions to the volume at system
T and 1
j CO
2
-oil viscosity, cP
j
d
viscosity of dead oil at the specied temperature, cP
j
i
initial dead oil viscosity at the specied temperature, cP
j
s
solvent viscosity, cP
T ratio between oil volume at system T, and 1 atm to the volume at system T
and 1
C
g
genetic algorithm evaluation constant D 1.0 for CO
2
solubility, oil SF, and
CO
2
-oil density and equal to 100 for CO
2
-oil viscosity correlation
Fit(i ) average tness of chromosome i , where the chromosome has many tness
values based on the number of data available ( )
GA genetic algorithm
MW oil molecular weight
n genetic algorithm population size and number of components in the mixture
nn number of available data points
1
s
saturation pressure, psia
Pt(i. ) tness function of data number of chromosome i , fraction
1
liq
CO
2
liquefaction pressure at the specied temperature, psia
1
b
bubble point pressure, 14.7 psia
1(c) crossover probability
1(m) mutation probability
Sol CO
2
solubility, mole fraction or scf/bbl or sm
3
/m
3
STDEV standard deviation, %
T temperature,

F
T
c;CO
2
CO
2
critical temperature (equal to 88

F)
Value
cal
solubility, swelling, density, and viscosity predicted value
Value
exp
solubility, swelling, density, and viscosity experimental value
V
o
, V
s
oil volume fraction and solvent volume fraction
.
j
mole fraction of component