Professional Documents
Culture Documents
Abdolhossein Hemmati-Sarapardeh
a
, Amin Shokrollahi
a
, Afshin Tatar
b
, Farhad Gharagheizi
c,d
,
Amir H. Mohammadi
c,e,
, Ali Naseri
f
a
Department of Chemical and Petroleum Engineering, Sharif University of Technology, Tehran, Iran
b
Department of Chemical Engineering, Sahand University of Technology, Tabriz, Iran
c
Thermodynamics Research Unit, School of Chemical Engineering, University of KwaZulu-Natal, Howard College Campus, King George V Avenue, Durban 4041, South Africa
d
Department of Chemical Engineering, Buinzahra Branch, Islamic Azad University, Buinzahra, Iran
e
Institut de Recherche en Gnie Chimique et Ptrolier (IRGCP), Paris Cedex, France
f
EOR Department, Research Institute of Petroleum Industry (RIPI), Tehran, Iran
h i g h l i g h t s
Reliable viscosity models for oil systems have been developed.
A large database consisting of over 1000 data have been used to develop them.
Their reliability is successfully examined against independent data.
a r t i c l e i n f o
Article history:
Received 26 November 2012
Received in revised form 14 April 2013
Accepted 18 July 2013
Available online 2 August 2013
Keywords:
Crude oil viscosity
Model
Reservoir uid
PVT
Support vector machine
a b s t r a c t
Viscosity of crude oil is a fundamental factor in simulating reservoirs, forecasting production as well as
planning thermal enhanced oil recovery methods which make its accurate determination necessary.
Experimental determination of reservoir oil viscosity is costly and time consuming. Hence, searching
for quick and accurate determination of reservoir oil viscosity is inevitable. The objective of this study
is to present a reliable, and predictive model namely, Least-Squares Support Vector Machine (LSSVM)
to predict reservoir oil viscosity. To this end, three LSSVM models have been developed for prediction
of reservoir oil viscosity in the three regions including, under-saturated, saturated and dead oil. These
models have been developed and tested using more than 1000 series of experimental PVT data of Iranian
oil reservoirs. These data include oil API gravity, reservoir temperature, solution gas oil ratio, and satura-
tion pressure. The ranges of data used to develop these new models cover almost all Iranian oil reservoirs
PVT data and consequently the developed models could be reliable for prediction of other Iranian oil res-
ervoirs viscosities. In-depth comparative studies have been carried out between these new models and
the most frequently used oil viscosity correlations for prediction of reservoir oil viscosity. The results
show that the developed LSSVM models signicantly outperform the existing correlations and provide
predictions in acceptable agreement with experimental data. Furthermore, it is shown that the proposed
models are capable of simulating the actual physical trend of the oil viscosity with variation of oil API
gravity, temperature, and pressure.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Viscosity of crude oil is an important parameter for evaluating
performance of the reservoir, reservoir simulation, designing pro-
duction facilities and utilizing the best scenario for production
[15]. Hence, accurate determination of this parameter is crucial
for the petroleum industry. The common approach to determine
the viscosity is laboratory analysis on the bottomhole samples or
recombined liquids and gases collected from the separators at
the surface. When PVT data are not available, in order to save time
and cost, uid properties are determined by empirical correlations
and equations of state (EOS).
Depending on the input variables, it is possible to divide corre-
lations used for determining oil viscosity in two classes [4]: The
rst class uses common oil eld data such as reservoir tempera-
ture, saturation pressure, oil API gravity, and solution gas oil ratio.
The second one is empirical or semi-empirical models that use
some characteristics not used in the rst approach such as reser-
voir uid composition, critical temperature, acentric factor, pour
point temperature, molar mass, and normal boiling point [68].
0016-2361/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.fuel.2013.07.072
k
2
Table 6
Statistical parameters of the presented model for determination of dead oil viscosity.
Statistical parameter
Training set
R
2a
0.959
Average absolute relative deviation
b
21.3
Standard deviation error
c
1.79
Root mean square error
d
1.78
N
e
93
Validation set
R
2
0.986
Average absolute relative deviation 21.9
Standard deviation error 2.30
Root mean square error 2.29
N 11
Test set
R
2
0.914
Average absolute relative deviation 19.7
Standard deviation error 1.68
Root mean square error 1.65
N
e
11
Total
R
2
0.955
Average absolute relative deviation 21.2
Standard deviation error 1.83
Root mean square error 1.82
N
e
115
a
R
2
1
P
N
i
Calc:i=Est:iexp:i
2
P
N
i
Calc:i=Est:iaverageexp:i
2
.
b
%AARD
100
N
P
N
i
jCalc:i=Est:iexp:ij
exp:i
.
c
STD
1
N
P
N
i
P
N
i
Calc:i=Est:i averageCalc:i=Est:i
2
q
.
d
RMSE
P
N
i1
Calc:i=Est:iexp:i
2
N
1=2
.
e
Number of data points.
Fig. 2. Comparison between the represented/predicted results of the developed
model and the experimental data of dead oil viscosity.
Fig. 3. Relative deviations of the represented/predicted dead oil viscosity values by
the proposed model.
Table 5
The optimum values of the LSSVM [36] models parameters at three different regions.
LSSVM Model Input parameters of
model
Model parameters
c r
2
Dead oil T API 436.90 2.34
Saturated reservoir oil l
od
P 51753139815.22 11.64
Under-saturated
reservoir oil
l
ob
P
b
P 364181.17 21.45
42 A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948
Subjected to the following constraint:
y
k
w
T
ux
k
b 6 e n
k
; k 1; 2; . . . ; N
w
T
ux
k
b y
k
6 e n
k
; k 1; 2; . . . ; N
n
k
; n
k
6 0; k 1; 2; . . . ; N
8
>
<
>
:
3
where x
k
and y
k
are kth data point input and kth data point
output, respectively. The e is the xed precision of the function
Table 7
Comparison between the performances of LSSVM [36] model and common correla-
tions for prediction of dead oil viscosity.
Author AARD, % R
2
RMSE
Beal [9] 891 0.1088 83.14
Beggs and Robinson [10] 217 0.0376 245.05
Glaso [11] 33.41 0.9270 3.84
Kaye [12] 52.05 0.2268 10.19
Al-Khafaji et al. [13] 29.97 0.7283 6.04
Petrosky [14] 41.62 0.8695 4.18
Egbogah and Ng [15] 55.60 0.9208 3.26
Labedi [16] 178 0.3910 14.95
Kartoatmodjo and Schmidt [17] 36.89 0.9065 4.50
Bennison [18] 70.97 0.6689 12.25
Elsharkawy and Alikhan [5] 72.92 0.9065 13.25
Hossain et al. [19] 68.92 0.6000 16.24
Naseri et al. [4] 27.57 0.8233 3.88
Alomair et al. [20] 72.42 0.8275 6.36
LSSVM Model (this study) 21.2 0.955 1.82
Fig. 4. Trend of dead oil viscosity changes versus oil API gravity at different
isotherms (this study).
Fig. 5. Comparison between the represented/predicted results of the developed
model and the experimental data of saturated oil viscosity.
Fig. 6. Relative deviations of the represented/predicted saturated oil viscosity
values by proposed model.
Table 8
Statistical parameters of the presented model for determination of saturated oil
viscosity.
Statistical parameter
Training set
R
2
0.988
Average absolute relative deviation 13.5
Standard deviation error 0.31
Root mean square error 0.31
N 323
Validation set
R
2
0.976
Average absolute relative deviation 13.6
Standard deviation error 0.24
Root mean square error 0.24
N 40
Test set
R
2
0.838
Average absolute relative deviation 13.2
Standard deviation error 0.77
Root mean square error 0.77
N 40
Total
R
2
0.979
Average absolute relative deviation 14.6
Standard deviation error 0.38
Root mean square error 0.38
N 403
Table 9
Comparison between the performances of LSSVM [36] model and common correla-
tions for prediction of saturated oil viscosity.
Author AARD (%) R
2
RMSE
Chew and Connally 1 [28] 29.64 0.823818 1.18
Chew and Connally 2 [28] 39.64 0.799963 1.46
Chew and Connally 3 [28] Out of range 0 Out of range
Beggs and Robinson [10] 31.90 0.532063 1.35
Al-Khafaji et al. [13] 20.79 0.799255 1.32
Khan et al. [27] Out of range 0 Out of range
Petrosky [14] 27.33 0.739977 1.14
Labedi [16] 249 0.428383 3.70
Kartoatmodjo and Schmidt [17] 25.71 0.789565 1.11
Elsharkawy and Alikhan [5] 24.45 0.657688 1.20
Hossain et al. [19] Out of range 0 Out of range
Naseri et al. [4] 52.32 0.567231 1.38
Bergman and Sutton [26] 26.70 0.733941 1.15
LSSVM Model (this study) 14.6 0.979 0.38
A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948 43
approximation. The n
k
and n
k
are slack variables. It is obvious that
when we choose a small e to develop a very accurate model, some
data points may be outside of the e precision. This may result in
infeasible solution. Therefore, we should use slack parameters to
dene the allowed margin of error. The c > 0 is the tuning parameter
of the SVM which determines the amount of the deviation from the
desired e. In other words, c is one of the tuning parameters of the
SVM. In order to minimize the cost function depicted in Eq. (2)
along with its constraints presented in Eq. (3), we should use the
Lagrangian for this problem as follows [45]:
La; a
1
2
X
N
k;l1
a
k
a
k
a
l
a
l
Kx
k
; x
l
e
X
N
k1
a
k
a
X
N
k1
y
k
a
k
a
k
4
X
N
k1
a
k
a
k
0; a
k
; a
k
2 0; c 4a
Kx
k
; x
l
ux
k
T
ux
l
; k 1; 2; . . . ; N 4b
where a
k
and a
k
are Lagrangin multipliers. Finally, the nal form of
the SVM is obtained as follows:
Fig. 7. Comparison between experimental data and predicted values provided by
new proposed LSSVM [36] model for reservoir oil viscosity of four saturated
samples .
Fig. 8. Percent relative error distribution of the most accurate saturated oil viscosity correlations and proposed LSSVM [36] model.
Table 10
Statistical parameters of the presented model for determination of under-saturated
oil viscosity.
Statistical parameter
Training set
R
2
0.999
Average absolute relative deviation 1.5
Standard deviation error 0.03
Root mean square error 0.03
N 418
Validation set
R
2
0.999
Average absolute relative deviation 1.3
Standard deviation error 0.05
Root mean square error 0.05
N 52
Test set
R
2
0.999
Average absolute relative deviation 1.4
Standard deviation error 0.04
Root mean square error 0.04
N 52
Total
R
2
0.999
Average absolute relative deviation 1.4
Standard deviation error 0.04
Root mean square error 0.04
N 522
44 A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948
f x
X
N
k;l1
a
k
a
k
Kx; x
k
b 5
In order to solve the problem and nd a
k
; a
k
; and b, we should solve
a quadratic programming problem which is very difcult. In order
to facilitate the original SVM algorithm, Suykens and Vandewalle
[36,44] proposed the least square modication of the SVM (LSSVM).
In their LSSVM approach, Suykens and Vandewalle [36,44] reformu-
lated the SVM as follows [45]:
Cost function
1
2
w
T
w
1
2
c
X
N
k1
e
2
k
6
Subjected to the following constraint:
Y
k
w
T
ux
k
b e
k
7
where c and e
k
are tuning parameter in LSSVM method and the er-
ror variable, respectively. The Lagrangian for this problem is as
follows:
Lw; b; e; a
1
2
w
T
w
1
2
c
X
N
k1
e
2
k
X
N
k1
a
k
w
T
ux
k
b e
k
y
k
8
where a
k
are Lagrangian multipliers. In order to solve the problem
we should equate the derivatives of Eq. (8) to zero. Therefore, the
following equations are obtained:
@L
@w
0 ) w
X
N
k1
a
k
ux
k
@L
@b
0 )
X
N
k1
a
k
0
@L
@e
k
0 ) a
k
ce
k
; k 1; 2; . . . ; N
@L
@a
k
0 ) w
T
ux
k
b e
k
y
k
0 k 1; 2; . . . ; N
8
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
:
9
As can be found from Eq. (9), there are 2 N+2 equations and 2 N+2
unknown parameters (a
k
, e
k
, w and b). Therefore, the parameters of
LSSVM can be obtained by solving the system of equations depicted
in Eq. (9) [45].
As mentioned earlier, the LSSVM has a tuning parameter c.
Since, LSSVM and SVM are both kernel-based methods, we should
consider the parameters of the kernel functions as other tuning
parameters. In case of RBF kernel function:
Kx; x
k
expjjx
k
xjj
2
=r
2
10
We will have another tuning parameter r
2
. As a result, in LSSVM
algorithm with RBF kernel function, we will have 2 tuning parame-
ters which should be obtained by minimization of the deviation of
the LSSVM model from experimental values [45].
It should be mentioned that the major advantage of the LSSVM
over the original SVM is the idea of modifying the inequality con-
straints of Eq. (3) to the equality constraints of Eq. (7). The param-
eters of the LSSVM are simply obtained by solving the system of
equations as depicted in Eq. (9) instead of solving a nonlinear qua-
dratic programming [45].
Fig. 9. Comparison between the represented/predicted results of the developed
model and the experimental data of under-saturated oil viscosity.
Fig. 10. Relative deviations of the represented/predicted under-saturated oil
viscosity values by the proposed model.
Fig. 11. Comparison between experimental data and predicted values provided by
the proposed LSSVM [36] model for four under-saturated oil viscosity samples.
Table 11
Comparison between the performances of LSSVM [36] model and common correla-
tions for prediction of under-saturated oil viscosity.
Author AARD (%) R
2
RMSE
Beal [9] 1.85 0.99783 0.133
Vazquez and Beggs [35] 5.23 0.97138 0.575
Khan et al. [27] 3.24 0.98812 0.298
Petrosky [14] 6.71 0.83011 0.940
Labedi [16] 1078 0.09697 43.557
Orbey and Sandler [32] 1.95 0.96242 0.495
Kartoatmodjo and Schmidt [17] 3.65 0.99864 0.107
Elsharkawy and Alikhan [5] 1.85 0.93720 0.618
Hossain et al. [19] 4.70 0.99481 0.225
LSSVM Model (this study) 1.4 0.99999 0.04
A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948 45
2.3. Computational procedure
In the next step, the database is divided into three sub data sets
including the Training set, the Optimization set, and the Test
set. Generally, the Training set is used to generate the model
structure, the Optimization set is applied for optimization of
the model and the Test (prediction) set is used to investigate
its prediction capability and validity [3739,46]. The database
was split randomly into three sub data sets: For this purpose, about
80%, 10%, and 10% of the main data set were randomly selected for
the Training set, the Optimization set, and the Test set,
respectively. The effect of the percent allocation of the three sub
data sets from the database on the accuracy of the nal model
has been studied elsewhere [37,39]. In distribution of the data
through the three sub data sets, we generally perform many distri-
butions to avoid the local accumulations of the data in the feasible
region of the problem. As a result, the acceptable distribution is the
one with homogeneous accumulations of the data on the domain
of the three sub data sets [39].
3. Results and discussion
In this study, three LSSVM [36] models were developed to pre-
dict the viscosity of dead oil, saturated reservoir oil and under-sat-
urated reservoir oil, as mentioned earlier. Furthermore, prediction
performance of the newly proposed models was compared against
all well-known previously published correlations by means of sta-
tistical and graphical error analyses.
The values of the probable global optima of all three models
including r
2
and c are reported in Table 5. It should be stated that
determination of these values are not generally simple and the
numbers of the reported digits are normally obtained by sensitivity
analysis of the overall errors of the optimization procedure to the
corresponding values [37,47]. Furthermore, Simulating Annealing
(SA) technique was applied in this study [48,49].
Table 6 indicates the statistical parameters of the proposed
model for prediction of dead oil viscosity, including Average Abso-
lute Relative Deviations (AARD), Root Mean Square Errors (RMSE),
squared correlation coefcients (R
2
), and Standard Deviation Errors
(STD). A comparison between the predicted dead oil viscosity val-
ues and the experimental values and relative error distribution for
experimental data points are demonstrated in Figs. 2 and 3,
respectively.
To investigate the validity of the proposed LSSVM [36] model
for dead oil viscosity, all previously published correlations have
been selected for comparison. As mentioned earlier, dead oil vis-
cosity is generally a difcult property to be correlated. Table 7
shows the results of all the correlations compared to the LSSVM
[36] model. According to this table, the developed model gives
an AARD of 21.2%, R
2
of 0.955, and RMSE of 1.82.
Fig. 4 shows the trend of changes in dead oil viscosity with
increasing API gravity of oil at various temperatures. As a general
rule, dead oil viscosity decreases with an increase in the oil API
gravity [50,51]. Furthermore, a general reduction in dead oil vis-
cosity is expected with increasing temperature, and this effect is
more sensible at lower temperatures [50,51]. As can be seen from
Fig. 4, this trend is captured by the proposed LSSVM [36] model for
dead oil viscosity.
Next model was developed for estimation of saturated reservoir
oil viscosity. Comparison between the results of predictions made
by proposed LSSVM [36] model, and the corresponding experimen-
tal viscosity values for saturated reservoir oil is illustrated in Fig. 5.
Furthermore, Fig. 6 shows relative error distribution of proposed
model. Its statistical parameters for prediction of saturated reser-
voir oil viscosity are reported in Table 8.
The efciency of the model for saturated oil viscosity was also
tested against all well-known correlations and results are dis-
played in Table 9. According to this table the values of AARD, R
2
,
and RMSE are 14.6%, 0.979, and 0.38 respectively. Statistical and
graphical error analysis demonstrate that the proposed LSSVM
[36] model outperforms previously published correlations .
To make sure that the proposed model is reliable, the validity of
model should be checked. For this purpose, the resulted trend from
the proposed model [36] was checked by four experimental data
samples. Fig. 7 demonstrates experimental and predicted viscosi-
ties made by the new LSSVM [36] model for four samples. It is
Fig. 12. Percent relative error distribution of the most accurate under-saturated oil viscosity correlations and the proposed LSSVM [36] model.
46 A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948
evident by increasing pressure ultimately to the bubble point pres-
sure, viscosity of saturated reservoir oil decreases. This developed
model successfully captured this trend and made predictions mar-
velously in close agreement with experimental data (see Fig. 7).
Moreover, the accuracy of the proposed model corresponding to
the ratio of pressure to the bubble point pressure, and also three
of the most precise correlations namely, Elsharkawy and Alikhan
[5], Al-Khafaji et al. [13], and Kartoatmodjo and Schmidt [17] were
visually checked. Fig. 8, shows this comparison. It is clear that the
proposed model has the smaller error range and least scatter
around zero error line.
The last model is constructed for prediction of under-saturated
oil viscosity. Statistical parameters of the developed model for esti-
mation of under-saturated reservoir oil viscosity are presented in
Table 10. A comparison between estimated values of under-satu-
rated reservoir oil viscosity by the recommended model and exper-
imental values is illustrated in Fig. 9. A tight cloud of points about
45 line for training, validation and testing data sets indicates the
robustness of this model. The percent relative error distribution
for all experimental data points is shown in Fig. 10. In the case of
trend analysis, the behavior of model [36] prediction is in good
agreement with the experimental data (see Fig. 11). In order to
investigate the efciency of the proposed model against all previ-
ously published ones, statistical parameters are displayed in Ta-
ble 11. This table reveals that the proposed models efciency is
the best. Furthermore, the corresponding diagram for relative error
distribution versus the ratio of pressure to bubble point pressure is
illustrated in Fig. 12. In this case, the relative error distribution of
the new model is closer to the zero line in comparison with the
other ones.
4. Conclusions
Several empirical viscosity correlations were evaluated using a
large data bank of the Iranian oil reservoirs. Initially, the previously
published correlations were analyzed for Iranian oil reservoir
in three regions including, under-saturated, saturated and dead
oil. Average Absolute Relative Deviation has been considered as
the main screening criteria. It was found that for the dead oil con-
dition, all the published correlations have high error, however Na-
seri et al. [4], Glaso [11], and Al-Khafaji et al. [13] exhibit more
accuracy for prediction the dead oil viscosity of Iranian oil reser-
voirs. For Iranian oil reservoirs, saturated oil viscosity correlations
show more accuracy than the dead oils ones and Elsharkawy and
Alikhan [5], Al-Khafaji et al. [13], and Kartoatmodjo and Schmidt
[17] correlations give the best results for this region. At the next
step, all the correlations were tested for under-saturated oil region.
Except the Labedi [16] correlation, all the correlations show
acceptable accuracy, however, Elsharkawy and Alikhan [5], Beal
[9], and Orbey and Sandler [32] correlations had the highest accu-
racies among the others. Finally, LSSVMmodels were developed for
predicting oil viscosity at three conditions of dead oil, under-satu-
rated oil, saturated oil, using a large data bank of Iranian oil reser-
voirs. The range of used data [1] for developing the new models
cover all Iranian oil reservoirs and consequently, these developed
models can be reliable for predicting of other Iranian oil reservoirs
viscosities. However, for application of these developed LSSVM
[36] models, the limitations of the data ranges upon with the mod-
els have been developed, should be considered. In addition, a com-
parison among the LSSVM [36] models and other well-known
correlations was presented and it was found that the accuracy of
the proposed models was better than the previously published
correlations; however the present model uses just two parameters
(r
2
and c). Moreover, it was demonstrated that the proposed
models are capable of simulating the actual physical trend of the
oil viscosity with variation of oil API gravity, temperature, and
pressure. The LSSVM [36] models can be implemented in any
reservoir simulator software and it is expected that they provide
better accuracy and performance for crude oil viscosity than the
previously published ones.
References
[1] Hemmati-Sarapardeh A, Khishvand M, Naseri A, Mohammadi AH. Toward
reservoir oil viscosity correlation. Chem Eng Sci 2013;90:5368.
[2] Ikiensikimama SS, Ogboja O. Evaluation of empirically derived oil viscosity
correlations for the Niger Delta crude. J Petrol Sci Eng 2009;69:2148.
[3] Al-Marhoun MA. Evaluation of empirically derived PVT properties for Middle
East crude oils. J Petrol Sci Eng 2004;42:20921.
[4] Naseri A, Nikazar M, Mousavi Dehghani SA. A correlation approach for
prediction of crude oil viscosities. J Petrol Sci Eng 2005;47:16374.
[5] Elsharkawy A, Alikhan A. Models for predicting the viscosity of Middle East
crude oils. Fuel 1999;78:891903.
[6] Xu D-H, Khurana AK. A simple and efcient approach for improving the
prediction of reservoir uid viscosity. In: SPE Asia pacic oil and gas
conference 1996 Copyright 1996. Adelaide, Australia: Society of Petroleum
Engineers, Inc.; 1996.
[7] Ahrabi F, Ashcroft SJ, Shearn RB. High pressure volumetric phase composition
and viscosity data for a North Sea crude oil and NGL. Chem Eng Res Des
1987;65:6373.
[8] Little JE, Kennedy HT. Calculating the viscosity of hydrocarbon systems with
pressure temperature and composition. Soc Pet Eng J 1968;6:15762.
[9] Beal C. The viscosity of air, water, natural gas, crude oil and Its associated gases
at oil eld temperatures and pressures. Trans. AIME 1946.
[10] Beggs H, Robinson J. Estimating the viscosity of crude oil systems. J Petrol
Technol 1975;27:11401.
[11] Glaso O. Generalized pressurevolume-temperature correlations. SPE J Pet
Technol 1980;32:78595.
[12] Kaye SE. Offshore California viscosity correlations. In: COFRC, TS85000940;
August 1985.
[13] Al-Khafaji A, Abdul-Majeed G, Hassoon S. Viscosity correlation for dead, live
and undersaturated crude oils. J Pet Res 1987;6:116.
[14] Petrosky GEJ. PVT correlations for gulf of mexico crude oils, MSC thesis,
University of Southwestern Louisiana, Lufayette, Louisiana, USA; 1990.
[15] Egbogah EO, Ng JT. An improved temperatureviscosity correlation for crude
oil systems. J Petrol Sci Eng 1990;4:197200.
[16] Labedi R. Improved correlations for predicting the viscosity of light crudes. J
Petrol Sci Eng 1992;8:22134.
[17] Kartoatmodjo T, Schmidt Z. Large data bank improves crude physical property
correlations. Oil Gas J 1994;92:515.
[18] Bennison T. Prediction of heavy oil viscosity. In: IBC heavy oil eld
development conference, London; 1998.
[19] Hossain MS, Sarica C, Zhang HQ, Rhyne L, Greenhill K. Assessment and
development of heavy oil viscosity correlations. In: SPE international thermal
operations and heavy oil symposium, Calgary, Canada; 2005.
[20] Alomair OA, Elsharkawy AM, Alkandari HA. Viscosity prediction of kuwaiti
heavy crudes at elevated temperatures. In: SPE heavy oil conference and
exhibition. Kuwait City, Kuwait: Society of Petroleum Engineers; 2011.
[21] Svrcek W, Mehrotra A. One parameter correlation for bitumen viscosity. Chem
Eng Res Des 1988;66:3237.
[22] Mehrotra AK. Generalized one-parameter viscosity equation for light and
medium liquid hydrocarbons. Ind Eng Chem Res 1991;30:136772.
[23] Teja A, Rice P. Generalized corresponding states method for the viscosities of
liquid mixtures. Ind Eng Chem Fundam 1981;20:7781.
[24] Johnson SE, Svrcek WY, Mehrotra AK. Viscosity prediction of Athabasca
bitumen using the extended principle of corresponding states. Ind Eng Chem
Res 1987;26:22908.
[25] Johnson SE, Svrcek WY. J Can Pet Technol 1991;26(5):60.
[26] Bergman DF, Sutton RP. An update to viscosity correlations for gas-saturated
crude oils. In: SPE annual technical conference and exhibition. Anaheim,
California, USA: Society of Petroleum Engineers; 2007.
[27] Khan S, Al-Marhoun M, Duffuaa S, Abu-Khamsin S. Viscosity correlations for
Saudi Arabian crude oils. In: SPE middle east oil show, Manama, Bahrain; 1987.
[28] Chew J-N, Connally CA. A viscosity correlation for gas-saturated crude oils.
Trans. AIME 1959.
[29] Abdul-Majeed GH, Clark KK, Salman NH. New correlation for estimating the
viscosity of undersaturated crude oils. J Can Pet Technol 1990;29.
[30] Almehaideb RA. Improved PVT correlations for UAE crude oils. In: Middle east
oil show and conference, 1997 Copyright 1997. Bahrain: Society of Petroleum
Engineers, Inc.; 1997.
[31] Dindoruk B, Christman PG. PVT properties and viscosity correlations for gulf of
Mexico oils. In: SPE annual technical conference and exhibition, 2001. New
Orleans, Louisiana: Society of Petroleum Engineers Inc.; 2001.
[32] Orbey H, Sandler SI. Prediction of the viscosity of liquid hydrocarbons and their
mixtures as a function of temperature and pressure. Can J Chem Eng
1993;71:43746.
[33] Sutton RP, Bergman DF. Undersaturated oil viscosity correlation for adverse
conditions. In: SPE annual technical conference and exhibition. San Antonio,
Texas, USA: Society of Petroleum Engineers; 2006.
A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948 47
[34] Sutton RP, Farshad F. Evaluation of empirically derived PVT properties for Gulf
of Mexico crude oils. SPE Reservoir Eng (Society of Petroleum Engineers)
1990;5:7986. 13172.
[35] Vazquez M, Beggs HD. Correlations for uid physical property prediction. SPE J
Pet Technol 1980;32:96870.
[36] Suykens J, Vandewalle J. Least squares support vector machine classiers.
Neural Process Lett 1999;9:293300.
[37] Gharagheizi F, Eslamimanesh A, Farjood F, Mohammadi AH, Richon D.
Solubility parameters of nonelectrolyte organic compounds: determination
using quantitative structureproperty relationship strategy. Ind Eng Chem Res
2011;50:1138295.
[38] Eslamimanesh A, Gharagheizi F, Illbeigi M, Mohammadi AH, Fazlali A, Richon
D. Phase equilibrium modeling of clathrate hydrates of methane, carbon
dioxide, nitrogen, and hydrogen+water soluble organic promoters using
Support Vector Machine algorithm. Fluid Phase Equilib 2012;316:3445.
[39] Eslamimanesh A, Gharagheizi F, Mohammadi AH, Richon D. Phase equilibrium
modeling of structure H clathrate hydrates of methane + water Insoluble
hydrocarbon promoter using QSPR molecular approach. J Chem Eng Data
2011;56:377593.
[40] Raee-Taghanaki S, Arabloo M, Chamkalani A, Amani M, Zargari MH,
Adelzadeh MR. Implementation of SVM framework to estimate PVT
properties of reservoir oil. Fluid Phase Equilib 2013;346:2532.
[41] Ahmadi MA, Ebadi M, Shokrollahi A, Majidi SMJ. Evolving articial neural
network and imperialist competitive algorithm for prediction oil ow rate of
the reservoir. Appl Soft Comput 2013;13:108598.
[42] Chamkalani A, Narehei MA, Chamkalani R, Zargari MH, Dehestani-Ardakani
MR, Farzam M. Soft computing method for prediction of CO
2
corrosion in ow
lines based on neural network approach. Chem Eng Commun
2013;200:73147.
[43] Omole O, Falode O, Deng DA. Prediction of Nigerian crude oil viscosity using
articial neural network. Pet Coal 2009;51:1818.
[44] Pelckmans K, Suykens JAK, Van Gestel T, De Brabanter J, Lukas L, Hamers B,
et al., LS-SVMlab: a matlab/c toolbox for least squares support vector
machines. Tutorial, KULeuven-ESAT. Leuven, Belgium; 2002.
[45] Suykens JAK, Van Gestel T, De Brabanter J, De Moor B, Vandewalle J. Least
squares support vector machines. World Scientic Publishing Company; 2002.
Incorporated.
[46] Chamkalani A, Mohammadi AH, Eslamimanesh A, Gharagheizi F, Richon D.
Diagnosis of asphaltene stability in crude oil through two parameters SVM
model. Chem Eng Sci 2012;81:2028.
[47] Gharagheizi F. QSPR analysis for intrinsic viscosity of polymer solutions by
means of GA-MLR and RBFNN. Comput Mater Sci 2007;40:15967.
[48] Xavier-de-Souza S, Suykens JAK, Vandewalle J, Boll D. Coupled simulated
annealing. IEEE Trans Syst, Man, Cyber, Part B: Cyber 2010;40:32035.
[49] Corana A, Marchesi M, Martini C, Ridella S. Minimizing multimodal functions
of continuous variables with the simulated annealing algorithm Corrigenda
for this article is available here. ACM Transactions on Mathematical Software
(TOMS) 1987;13:26280.
[50] Ahmed T. Reservoir engineering handbook. Gulf Professional Publishing; 2006.
[51] Reid RC, Prausnitz JM, Poling BE. The properties of gases and liquids; 1987.
48 A. Hemmati-Sarapardeh et al. / Fuel 116 (2014) 3948