Optimal Distribution Function Determination for Plus Fraction

Splitting

Sajjad Foroughi,1* Davood Khoozan1* and Saeid Jamshidi2*

1. MAPSA Technology Center, Abdal Industrial Projects Management Company (MAPSA), Tehran, Iran
2. Chemical and Petroleum Engineering Department, Sharif University of Technology, Tehran, Iran

Reservoir fluid modelling is one of the most important steps in reservoir simulation and modelling of flow lines as well as surface facilities. One of
the most uncertain parameters of the reservoir fluids is the plus fraction. An accurate and consistent splitting scheme can reduce this uncertainty
and as a result, enhance the modelling of reservoir fluids. The existing schemes for splitting plus fractions are all based on assuming a specific
mole fraction‐molecular weight distribution with predefined constant values that may yield inaccurate and inconsistent results. In this study, an
optimization‐based algorithm was developed to determine the aforementioned controlling parameters of the plus fraction distribution function,
enforcing the relationship between specific gravity and molecular weight of the single carbon numbers (SCNs). The introduced optimization‐
based splitting technique was applied to different samples, covering a wide range of reservoir fluids, including gas condensates, volatile oils, black
oils, and heavy oils. The results showed that the proposed technique yielded a more consistent molecular weight‐mole fraction distribution
concerning the experimental extended analysis of plus fractions, yielding an average relative error of 25.8 % compared to 76, 33.6, and 45.9 %
for the Katz, Ahmed, and Whitson methods, respectively. It was also shown that the proposed method results in more accurate and more
consistent phase behaviour predictions than the existing methods concerning the experimental data. Furthermore, the results showed that the
introduced optimization‐based method yields monotonic split samples regarding specific gravity and molecular weight, while the conventional
techniques do not guarantee to preserve the monotonicity.

Keywords: splitting, hydrocarbon plus fraction, molar distribution functions, characterization

INTRODUCTION tion of the plus fraction has a significant effect on the reservoir
fluid phase behaviour and calculation of its physical properties.[11]

O
il and gas reservoir fluids may contain a wide range of
components, including hydrocarbons and non‐hydrocarbons
(e.g., CO2, N2, and H2S). Generally, these components can be
classified into the following categories: (1) pure components with a
specified acentric factor and critical properties; (2) components with
a measured or estimated molecular weight, specific gravity, and
critical properties (i.e., single carbon number (SCN) components);
and (3) heavy ends or plus fractions containing a wide range of
components with carbon numbers higher than a specific quantity.
Usually, mole fraction, molecular weight, and specific gravity are the
only available measured data for plus fractions.[1,2] It should be
mentioned that the mole fraction of the plus fraction is calculated
based on its measured mass fraction in the laboratory. The focus of
this study is on plus fraction characterization and splitting.
In the petroleum industry, laboratory tests are carried out to
determine the composition of reservoir fluids. Mole fractions of
non‐hydrocarbons and hydrocarbons in the range of methane to
the plus fraction are measured during these experiments, while
molecular weight and specific gravity of the plus fraction are also
measured. As mentioned above, these are the only available data
on the plus fraction.[3]
Poor description of plus fractions can lead to inaccuracy in PVT
predictions.[4–7] Furthermore, the unknown composition of the
plus fraction may result in modelling problems in petroleum
characterization.[6,8,9] As a result, one may claim that the
characterization of the plus fraction plays a major role in PVT
analysis of reservoir fluids. Various techniques have been proposed
to characterize reservoir fluid plus fractions.[10] It should be
emphasized that employing the proper approach for characteriza-
Plus fractions consist of single carbon number (SCN) pseudo‐
components. Reliable determination of the composition of SCNs
in the plus fraction is a fundamental step in the process of
reservoir fluid characterization.[12,13] Furthermore, accurate
characterization of SCN properties is very important for phase
behaviour calculations of fluid samples employing equations of
state.[14] This is a known fact for a wide range of reservoir fluids
from gas condensates to heavy oils.[5,15,16] Detailed experimental
composition analysis of plus fractions is rarely available. There-
fore, mathematical methods, known as splitting schemes, are
developed for their characterization.[16]
Several methods have been proposed in the literature for splitting of
plus fractions into SCN groups. One of the earliest attempts was
introduced by Katz,[17] employing an exponential model as the
distribution function. This method was based on the extended
analysis of the behaviour of different condensate systems.[18] Another
splitting scheme, proposed by Pederson et al.,[19] was based on
assuming a logarithmic relationship between the carbon number of
SCN groups and their corresponding mole fractions. Ahmed et al.[20]
presented a method for plus fraction extension based on a detailed
* Author to whom correspondence may be addressed. E‐mail address:
jamshidi@sharif.edu; foroughi@mapsaeng.com; khouzan@mapsaeng.com
Can. J. Chem. Eng. 1–13, 2019
© 2019 Canadian Society for Chemical Engineering
DOI 10.1002/cjce.23516
Published online in Wiley Online Library
(wileyonlinelibrary.com).

MONTH 2019 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 1

composition analysis of 34 hydrocarbon systems. They proposed a
marching technique to calculate the molar distribution of SCN groups
employing two correlations for molecular weight and specific gravity
of Cn+ in terms of n. In another study conducted by Huang and
Radosz,[21] a Gaussian distribution function was proposed for splitting
plus fractions into SCN groups. Whitson et al.[22] used a gamma
distribution function to describe the relationship between the mole
fraction and molecular weight of pseudo‐components of a plus
fraction. Similar to Whitson’s distribution, Riazi[5] employed a
generalized distribution function to relate the mole fraction of SCN
groups to their physical properties during the splitting process.
More recently, Rodriguez and Hamouda[10] proposed a character-
ization scheme that fits the three‐parameter gamma function to the
experimental molar distribution. In their method, the limiting
molecular weights for integrating the fitted distribution function are
determined simultaneously during the integration process by fitting
the characterized and experimental mole fractions. Petitfrere
et al.[23] proposed a distribution function for the liquid to feed the
mole ratio in the context of the semi‐continuous thermodynamics
developed by Cotterman and Prausnitz.[24]
Accurate characterization of the plus fraction plays a
fundamental role in the modelling of reservoir fluids. All the
methods mentioned above have a common weakness in
neglecting the effect of specific gravity through the process of
splitting. In other words, none of the existing methods take into
account the specific gravity distribution of SCN pseudo
fractions. This assumption can lead to a non‐monotonic
specific gravity distribution in which the specific gravity of
the final plus fraction is smaller than the specific gravity of the
last SCN pseudo‐component. It should be emphasized that the
reported specific gravity and molecular weight of SCN fractions
in the literature are monotonic.[12,25,26] Also, the proposed
correlations for prediction of these parameters follow a
monotonic behaviour.[12,25,26] As a result, developing a new
scheme that employs the relationship between specific gravity
and molecular weight of pseudo‐components to ensure their
monotonicity during the splitting process is necessary.
Among the existing splitting schemes, which are based on
distribution functions, the most extensively used approaches are
the three‐parameter gamma distribution and the generalized
distribution models, proposed by Whitson[4] and Riazi,[27] respec-
tively. Also, Pedersen[19] proposed a characterization method in
which a two‐parameter logarithmic relation between carbon
number and mole fraction is considered.
In this study, the focus is on the two distribution functions
proposed by Whitson[4] and Riazi.[27] It should be emphasized
that the unknown parameters of these distribution functions
highly depend on the reservoir fluid plus fraction properties.
Using default values for these parameters without considering the
reservoir fluid properties may lead to non‐monotonic and
inaccurate results. In this study, a splitting algorithm based on
optimization techniques is proposed to determine the values of
distribution function parameters. The obtained values for
distribution function parameters yield more accurate results in
the phase behaviour of the reservoir fluid, while the monotonicity
of molecular weight and specific gravity is preserved.
METHODOLOGY
Reliable characterization of reservoir fluids requires a complete
analysis of their mixture. For mixtures in which the composition
of heavy hydrocarbons is represented by a single hydrocarbon
plus fraction, it is important to know the distribution of carbon
2 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING
numbers to describe the mixture accurately. Ideally, a full true‐
boiling‐point (TBP) analysis should be performed to obtain the
molar distribution of carbon numbers in the plus fraction.
Although gas chromatography (GC) can also be considered, it
provides less information than TBP distillation.
Unfortunately, experimental data on hydrocarbon plus fractions
are not often available. To solve this problem, distribution functions
can be employed to determine the molar distribution of SCN
fractions in heavier plus fraction components of reservoir fluids.
Two types of distribution functions that have extensively been used
in the literature are gamma and generalized distribution models.
Gamma Distribution Model
Whitson[4] proposed one of the most widely used distribution
models for characterizing plus fractions (Cn+) known as the three
parameter gamma (TPG) probability distribution function (PDF).
The TPG PDF in terms of molecular weight can be employed to
describe the relation between the mole fraction (zi) and molecular
weight (MWi) of SCN components within the plus fraction
(Cn+).[28] The TPG model has been used to express the
distribution of the plus fraction properties in reservoir fluids.[28]
If a probability distribution function is employed to express the
molecular weight distribution of a given plus fraction, the mole
fraction of each SCN group can be calculated by:[26]
+
Pi
z i = zCn+ ∫P − F (P ) dP (1)
i
where z i is the mole fraction of each SCN group and zCn+is the mole
fraction of the plus fraction (i.e., Cn+). It is assumed that F(P) is a
continuous distribution function of P and the total area under its
curve from P0 to ∞ is equal to unity. Moreover, P0 is the lowest value
of the considered property in the plus fraction. Also, i represents the
single carbon number and P is a specified property such as molecular
weight (MW) or specific gravity (SG). (Pi−, Pi+) are the lower and
upper property bounds of the fraction i , which are defined as:
Pi + Pi + 1
Pi+ = (2)
2
Pi + Pi − 1
Pi− = (3)
2
Based on the properties of probability distribution functions,
we can calculate the average value of parameter P for the mixture,
Pav, as:
∞
Pav = ∫P 0
PF (P ) dP (4)
where P0 is the lowest value of the considered property in the plus
fraction. It should be mentioned that in the case of specific gravity
(SG), zi represents the volume fraction of each SCN group.
The probability density function of the gamma distribution
model, F(P), can be expressed as:[4]
(P − η)α − 1 exp [−(P − η)/ β ]
F (P ) = (5)
β α Γ(α )
where α, β, and η are the characteristic parameters of the
distribution function that should be determined for each mixture,
MONTH 2019
and Γ is the gamma function. Parameter η is the minimum Generalized Distribution Model
molecular weight (specific gravity) that can be present in the Cn+ Extensive analysis has been carried out on primary characteriza-
fraction distribution.[4] Substitution of F(P) into Equation (4) gives tion parameters for the plus fraction of a wide range of over 100
the average property of the plus fraction as: reservoir fluids by Riazi.[5] Based on this analysis, the following
versatile distribution model was proposed to yield the best fit for
Pav = η + αβ (6) various properties:
1
The mole fraction (molecular weight case) or volume fraction ⁎ ⎡ A ⎛ 1 ⎞⎤B
(specific gravity case) of each pseudo‐component can be P = ⎢ ln ⎜ ⁎ ⎟ ⎥ (12)
⎣ B ⎝ x ⎠⎦
calculated by substituting Equation (5) into Equation (1):
where
⎡ ⎤
xi =
(
⎢⎣Γ α ,
Pi− − η
β ) − Γ (α , Pi+ − η
β ) ⎥⎦
(7) ⁎ P − Po
P = , and x⁎ = 1 − xc (13)
Γ(α ) Po

and the considered property (i.e., molecular weight and specific P is a property such as molecular weight (M) or specific gravity
gravity) of each pseudo‐component, Pav, i , can be obtained by:[4] (SG). xc is the cumulative weight, mole, or volume fraction. Po is a
specific parameter for each property (Mo or SGo) in each sample,
⎡
Pav, i = η +
(
αβ ⎢Γ α + 1,
⎣
Pi− − η
β ) − Γ (α + 1, Pi+ − η
β ) ⎤⎥⎦ (8)
and A and B are the controlling parameters of the distribution
function. This distribution model can be transformed into a
Γ(α + 1) probability density function, F (P ⁎) , as:[5]

⁎ ⎛ B2 ⎞ ⁎ ⎛ B ⁎ ⎞
It should be mentioned that the parameters of the distribution F (P ) = ⎜ ⎟ × (P ) B − 1 exp ⎜ − (P ) B⎟ (14)
⎝A⎠ ⎝ A ⎠
functions are not independent. Analyzing Equation (6), we can
conclude that if α is determined, using a reliable value for η, the
last parameter, β, can be evaluated. The average properties of the mixture can be determined using
According to Whitson,[4] there are two methods for the the following relation:
determination of α: minimization and direct estimation of α. In ∞
the minimization technique, α is determined in a way that the ⁎
difference between the calculated molar distribution and the
Pav = ∫ P ⁎F (P ⁎) dP ⁎ (15)
0
mole fractions resulted from compositional analysis of the plus
fraction become minimized. This method has some limitations ⁎ ⁎
where Pav is the average value of P for the mixture. By
as it needs a full compositional analysis of the plus fraction, ⁎
substituting F (P ) from Equation (14) into the above integral,
preferably with a negligible quantity of the last fraction. In the ⁎
Pav can be calculated as:
direct estimation approach, an empirical correlation is used to
determine α. 1
Whitson[4] proposed a procedure for estimation of α in the ⁎ ⎛ A ⎞B ⎛ 1 ⎞⎟
Pav = ⎜ ⎟ Γ ⎜1 + (16)
absence of an extended analysis of the plus fraction for gamma ⎝B⎠ ⎝ B⎠
distribution method:
To estimate the lower and upper property bounds of the pseudo
α = Y−1 (0.500 087 6 + 0.164 885 2Y − 0.054 417 4Y2) (9) component i for determining the molar distribution of the plus
fraction using Equation (1), i.e., (Pi−, Pi+), Equations (2) and
where: Equation (3) can be employed. The mole or volume fraction and
the specific property of each pseudo component can be
(10) calculated by:[5]
Y = ln [(Mn + − η)/ m G]
⎛ B ⁎− ⎞ ⎛ B ⁎+ B⎞
and: x i = exp ⎜ − (Pi ) B⎟ − exp ⎜ − (Pi ) ⎟ (17)
⎝ A ⎠ ⎝ A ⎠

⎡ N ⎤1 / zn+
1 ⎛⎜ A ⎞⎟1 / B ⎡ ⎛⎜ 1 −⎞ ⎛ 1 +⎞ ⎤
m G = ⎢∏ (Mi − η) zi⎥ (11) Pav, i = − ⎢Γ 1 + , qi ⎟⎠ − Γ ⎜⎝1 + , qi ⎟⎠ ⎥ (18)
⎢⎣ i = n ⎦⎥ xi ⎝ B ⎠ ⎣ ⎝ B B ⎦

It should be mentioned that since the value of α should be where

larger than 1; there must be a constraint on Y which is shown by
Whitson[4] to be 0 < Y < 0.5772. B ⁎− B ⁎+
qi− = Pi qi+ = Pi (19)
It should also be emphasized that this relation cannot be A A
employed in the case of gas condensate and light hydrocarbon
samples,[4] which limits its applicability as a general rule for the It should be emphasized that there are no existing methods for
estimation of α. determination of the controlling parameters of the generalized

MONTH 2019 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 3

distribution function (i.e., A, B, and Po) in a way that yields
reliable distributions for molecular weight and specific gravity.
Proposed Optimization‐Based Splitting Technique
Usually, average (bulk) properties such as MW and SG of
hydrocarbon plus fractions are the only available data. Phase
behaviour and volumetric data of reservoir fluids principally
depend on their composition. Hence, it is essential to have a
reliable method to estimate the distributions of the required
properties from the available bulk properties.
If the default values of the controlling parameters of the
distribution functions are employed to characterize the plus
fractions, they become independent from the reservoir fluid type.
However, it is reported in the literature that the parameters of the
distribution functions are highly dependent on the fluid type and
the plus fraction properties.[2,29]
Furthermore, independent determination of distribution para-
meters of specific gravity and molecular weight may lead to
inconsistency issues between the molecular weight and specific
gravity of the obtained pseudo components. Hence, considering
the consistency between different properties of the resulted split
sample is an essential step in the process of plus fraction
characterization.
Before introducing the fundamental idea of the proposed
splitting method, two important facts should be stated:
1. Based on the values of physical properties of C6 to C50 SCN
groups with molecular weights ranging from 84–700,
Riazi[26] observed that the specific gravity and molecular
weight of SCN groups are not independent and they follow
the following trend:
SG = 1.07 − exp (3.560 73 − 2. 938 86 × MW 0.1) (20)
2. Volume fractions and mole fractions of the components of a
fluid sample are not independent and are related through the
specific gravity and molecular weight of the components.
Specifically, for a split sample of a plus fraction CN+, the
following relation exists between volume fraction, xvi, and
mole fraction, xmi, of the pseudo‐components:
xmi MWi
SG i
x vi = xmCN + MWCN +
(21)
SGCN +
In this work, a new splitting scheme is introduced, which is
based on a simple yet important fact: the split sample is a
representation of the actual plus fraction composition. As a result,
the split sample should satisfy three main criteria based on the
facts above:
1. The molecular weight and specific gravity of the resultant
pseudo components represent actual hydrocarbon compo-
nents, and as a result, they cannot be independent and
should follow the trend of hydrocarbon components, e.g.,
Equation (20). Although this expression is not the only
available relationship between molecular weight and specific
gravity of the SCN groups, it is employed here due to its
simplicity and generality. In fact, any other expressions that
can appropriately correlate these two parameters can be used
instead of Equation (20).
4 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING
2. The resultant split sample is a representation of an actual
hydrocarbon sample, and as a result, the mole fractions
resulted from the molecular weight distribution, and the
volume fractions generated from the specific gravity distribu-
tion should be consistent. In other words, the obtained mole
fractions (from the MW distribution) and volume fractions
(from the SG distribution) of the pseudo‐components of the
split sample should satisfy Equation (21).
3. The obtained molecular weights and specific gravities of the
split sample pseudo components should be mono-
tonic.[12,25,26]
In this paper, an optimization‐based splitting scheme is
introduced that enforces the aforementioned criteria. To this
end, an objective function is defined that quantifies the amount of
the violation of a given molecular weight and specific gravity
distribution concerning the aforementioned enforcing criteria. By
minimizing this objective function employing the controlling
parameters of the distribution functions as the design parameters,
the optimum split sample satisfying the stated criteria will be
obtained.
In this method, two distinct distributions are considered for
molecular weight and specific gravity. Although there are no
restrictions to specify different types of distribution functions for
molecular weight and specific gravity, they are considered to be
the same in this paper.
The objective function can be defined by the following
algorithm:
1. Select the distribution model (e.g., gamma or generalized).
2. Get the four input parameters for distribution functions of
molecular weight and specific gravity; two of these para-
meters are related to the molecular weight distribution
function (e.g., α and η of the MW gamma distribution) while
the other two parameters are for specific gravity distribution
(e.g., α and η of the SG gamma distribution).
3. Calculate the third parameter of the molecular weight and
specific gravity distributions using the input parameters, and
Equation (6) or Equation (16) for gamma or generalized
distribution functions, respectively.
4. Calculate the pseudo‐component molecular weights and
specific gravities separately using the specified MW and SG
distributions. Simultaneously, calculate the mole fraction and
the volume fraction of each pseudo‐component based on MW
and SG distributions, respectively. It should be noted that
these calculations are carried out by employing Equation (7)
and Equation (8) for gamma distribution and Equation (17)
and Equation (18) for generalized distribution. Here, the
evaluated mole fraction of each pseudo‐component based on
the molecular weight distribution is denoted by xmi, while xvi
is used for the calculated volume fractions based on the
specific gravity distribution.
5. Based on the calculated mole fraction (based on the MW
distribution), molecular weight, and specific gravity of
pseudo‐components in step 4, calculate the volume fraction
of each pseudo‐component using Equation (21), denoted
by x vicalc .
6. Using the calculated volume fraction (based on the SG
distribution), molecular weight, and specific gravity of
pseudo‐components in step 4, calculate the mole fraction of
each pseudo component using Equation (21), denoted
calc
by x mi .
MONTH 2019
 7. Using the molecular weight of each pseudo‐component
(obtained from the MW distribution), calculate the specific
gravity of each pseudo‐component using Equation (20),
denoted by SGicalc .
8. To enforce the stated criteria, calculate the objective function
for characterizing CN+, OF, as:
range of molecular weight and specific gravity. Properties of the
plus fractions used for evaluation of the introduced splitting
method are presented in Table 1. To examine the generality of the
proposed method for plus fraction characterization, a wide range
of reservoir fluids has been selected. As shown in Table 1, these
sets of data cover a wide range of C7+ (heavy end) mole fraction

OF (αMW , ηMW , αSG, ηSG ) Calc Calc

xvi − x vi xmi − x mi SG i − SGiCalc
or = ∑ xvi
+∑ xmi
+∑ SG i
OF (BMW , P0, MW , BSG, P0, SG ) i i i

Subject to: (22)

αMW ≥ 1 BMW > 0
αSG ≥ 1 BSG > 0
or
MWSCN −(N − 1) ≤ ηMW ≤ MWSCN −(N ) MWSCN −(N − 1) ≤ P0, MW ≤ MWSCN −(N )
SGSCN −(N − 1) ≤ ηSG ≤ SGSCN −(N ) SGSCN −(N − 1) ≤ P0, SG ≤ SGSCN −(N )

The constraints imposed on α are based on the restriction made from 0.2502–98.46 %, including gas condensates, volatile oils,
on this parameter by Whitson.[4] Since no limitation on the B black oils, and heavy oils. Data sets 1–17 are gas samples while
parameter is introduced by Riazi,[5] based on the characteristics of sets 18–43 are oil samples. Samples 1–11 are from Al‐Meshari and
molecular weight and specific gravity distributions, it is enforced McCain,[1] the data for samples 12–15 are taken from Folsta
to be larger than one. Parameter η of the gamma distribution and et al.,[31] and samples 16 and 17 are from Whitson.[4] Oil samples
parameter P0 of the general distribution are the minimum used in this study are selected as follows: data sets 18–28 are
molecular weight or specific gravity expected to occur in the taken from Al‐Meshari and McCain,[1] the data for samples 29–31
CN+ fraction,[4,5] and the recommended ranges of these para- are from Krejbjerg and Pedersen,[32] data set 32 is taken from
meters are between the molecular weight and specific gravity of
the N‐1st and Nth SCN fraction,[4,5] as enforced in the proposed
optimization problem. The algorithm for calculating the objective
function is presented in Figure 1.
In this work, the objective function is minimized using the
Nelder‐Mead optimization method.[30] The Nelder‐Mead method
is a commonly applied numerical optimization technique in
multidimensional problems. This method is suitable for nonlinear (Equation (16)).
optimization problems for which the derivatives may not be
known. The Nelder‐Mead technique is a heuristic search‐based
method with a high chance of finding the global minima. Since
the distribution parameters that should be optimized are bounded (Equation (17)).
to specific ranges, the proposed splitting algorithm can be
expressed as a constraint optimization problem that is tackled
by the Nelder‐Mead technique in this paper.
(Equation (18)).

RESULTS AND DISCUSSIONS

Experimental data on 43 different fluid samples have been used to
evaluate the performance of the proposed method comparing to
the common existing splitting techniques. The available data for
CN+ in all the considered samples are the mole fraction, molecular
weight, and specific gravity. The plus fractions of all the fluid
samples are split using the proposed and existing methods up to
the SCN of 30. This extended composition is used to investigate
the performance of the considered splitting techniques.
It should be emphasized that the proposed method is not
limited to the C7+ fraction and can be used for any plus fraction.
However, since the correlative splitting schemes (Katz and
Ahmed) are limited to C6+ and C7, these plus fractions are used
to compare the performance of the proposed technique with the
common splitting schemes.
The fluid compositions used for validation of the proposed Figure 1. The Algorithm of the proposed objective function for gamma
method are selected such that it contains plus fractions in a wide distribution (generalized distribution).

MONTH 2019 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 5

 Whitson,[4] sample 33 is from Carreón‐Calderón et al.,[8] and
Table 1. Plus fraction properties of the considered samples for samples 34–43 are taken from Jaubert et al.[33] The performance
characterization study of the introduced splitting algorithm is compared to the most
popular splitting techniques including Ahmed et al.,[20] Katz,[17]
Fluid zPF MWPF SGPF Fluid type
and Whitson et al.[22]
1 0.0430 145.2 0.7920 Gas condensate
2 0.0500 156.6 0.7950 Gas condensate Molar Distribution Prediction of the Plus Fraction
3 0.0647 163.8 0.8040 Gas condensate
4 0.0687 157.3 0.8020 Gas condensate The phase behaviour of any hydrocarbon mixture is influenced by
5 0.0723 152.2 0.8060 Gas condensate the splitting results of its plus fraction significantly.[6,34] To
6 0.0804 153.8 0.7980 Gas condensate analyze the effect of mole fraction distribution on the phase
7 0.0999 180.8 0.8310 Gas condensate behaviour of a given sample, a specific sample is split using
8 0.1087 173.3 0.8110 Gas condensate different distribution parameters for molecular weight, and the
9 0.1145 190.3 0.8220 Gas condensate resultant molar distributions and P‐T diagrams are illustrated in
10 0.1227 154.6 0.7910 Gas condensate Figure 2a and b.
11 0.1239 158.098 0.8050 Gas condensate The sensitivity analysis is performed on sample 8. The plus
12 0.002 502 144.6 0.79 Gas condensate
fraction of this sample is split employing three different α values,
13 0.022 867 119.3 0.747 Gas condensate
14 0.005 535 132.0 0.786 Gas condensate
including 1, 1.25, and 1.5, using the gamma distribution function.
15 0.021 936 145.8 0.787 Gas condensate The obtained extended compositions for these values of α are
16 0.0154 141.3 0.787 Gas condensate shown in Figure 2a. It should be mentioned that the value of η is
17 0.020 22 147.3 0.737 Gas condensate considered to be 94.6, while the β value for this sample is
18 0.1292 205 0.837 Oil determined by Equation (6). As can be seen in Figure 2a,
19 0.148 259.1 0.8723 Oil increasing the α value results in the creation of a left‐skewness in
20 0.1566 232 0.8601 Oil the distribution function, which means that the percentage of
21 0.1692 173 0.8364 Oil heavy components increases with increasing the value of alpha.
22 0.172 218 0.8294 Oil Analyzing Figure 2, one may conclude that increasing the value
23 0.19 250 0.8549 Oil
of alpha leads to an increase in the size of the phase envelope.
24 0.2053 188 0.8275 Oil
25 0.213 168 0.8067 Oil
Hence, it can be concluded that changes in the α parameter of the
26 0.2273 177 0.8137 Oil gamma distribution lead to significant alterations in the phase
27 0.237 172 0.8017 Oil diagram of the fluid sample. Therefore, correct determination of
28 0.2468 240 0.8586 Heavy oil distribution parameters is a key step in the accurate characteriza-
29 0.468 257 244.7 0.887 Heavy oil tion of hydrocarbon samples.
30 0.9846 316.9 0.933 Heavy oil As previously stated, Whitson[4] proposed an empirical relation
31 0.784 736 540.0 1.006 Heavy oil to determine the α parameter of the gamma distribution.
32 0.3684 198.7 0.841 Oil However, their proposed algorithm does not yield a reliable value
33 0.460 71 233.9 0.879 Heavy oil for α in the case of gas condensates and light hydrocarbons. They
34 0.4884 244.2 0.855 14 Oil
suggest setting the α parameter equal to one in these cases.[4]
35 0.4983 240.2 0.869 Oil
36 0.3154 207.1 0.845 Oil
Furthermore, it should be mentioned that Whitson’s method[4]
37 0.228 192.5 0.83 Oil does not yield reliable results for some heavy hydrocarbons as
38 0.460 71 233.5 0.879 Oil well.[19]
39 0.376 89 208.3 0.851 Oil Considering the gamma distribution, the α parameter for the
40 0.4115 189.1 0.828 Oil molecular weight is calculated for the considered 43 fluid samples
41 0.278 57 199.7 0.825 Oil employing the proposed splitting technique, and the results are
42 0.300 53 202.0 0.839 Oil compared to the Whitson’s method in Table 2. Based on the
43 0.283 211.7 0.853 Oil research of Riazi,[26] the molar distribution of gas condensates is
usually exponential, while for the black oil or crude oil samples, it

(a) (b)

Figure 2. Effect of the α parameter on: (a) the molar distribution of the plus fraction; and (b) the phase envelope of a gas condensate sample.

6 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING MONTH 2019

 gas condensate samples (7 and 15 in Table 1) with the available
Table 2. The obtained value for α parameter of gamma distribution experimental extended analysis of plus fractions are considered.
using the proposed scheme and Whitson method The SCN‐group mole fraction‐molecular weight distribution is
Gamma distribution parameter (α)
evaluated using the proposed optimization technique and
Whitson’s method for the four selected samples, and the results
Fluid Fluid type Proposed method Whitson[4] are shown in Figure 3. The experimental data are also plotted as
the reference points. As it can be seen, the proposed splitting
1 Gas condensate 1.0000 1.0738 method outperforms Whitson’s method in terms of yielding the
2 Gas condensate 1.0000 1.0530
molecular weight distribution of the plus fraction, noticeably in
3 Gas condensate 1.0482 1.0406
4 Gas condensate 1.0000 1.0498
cases (a) and (c), an oil and a gas condensate sample, respectively.
5 Gas condensate 1.0000 1.0565 To quantify the results, the average absolute relative errors of
6 Gas condensate 1.0000 1.0564 the resultant molar distribution with respect to the experimental
7 Gas condensate 1.0000 1.0192 data for the considered samples are evaluated for the proposed
8 Gas condensate 1.2033 1.0286 technique along with the considered existing splitting methods,
9 Gas condensate 1.4814 1.0149 and the results are shown in Table 3. As can be seen, for the
10 Gas condensate 1.0000 1.0573 considered oil samples, the proposed technique yields the
11 Gas condensate 1.0000 1.0479 minimum error, more noticeably in sample 1. For the gas
12 Gas condensate 1.0000 1.0757 condensate samples, the proposed technique yields the minimum
13 Gas condensate 1.0000 1.1631
error for sample 4, while in the other sample, Ahmed’s method
14 Gas condensate 1.0000 1.1060
15 Gas condensate 1.0000 1.0744
predicts the molar distribution of pseudo‐components more
16 Gas condensate 1.0000 1.0838 accurately. However, it should be mentioned that contrary to
17 Gas condensate 1.3121 1.0912 other techniques that lead to inconsistent results in some
18 Oil 1.4756 1.0096 samples, the proposed technique yields acceptable results in all
19 Oil 1.9333 1.0313 the considered samples. Furthermore, the results show that the
20 Oil 1.6379 1.0143 proposed method results in the lowest average relative error
21 Oil 1.0500 1.0241 considering all samples.
22 Oil 2.5415 1.0093
23 Oil 2.4504 1.0222 Monotonicity in the Split Sample
24 Oil 1.1971 1.0155
One of the issues that make the presented optimization‐based
25 Oil 1.1524 1.0349
26 Oil 1.2229 1.0248
method superior to the conventional splitting techniques is the
27 Oil 1.3182 1.0318 fact that contrary to other existing techniques, it preserves the
28 Heavy oil 1.9312 1.0175 monotonicity between the specific gravity and molecular weight
29 Heavy oil 1.4337 1.0245 through considering a relation between them during the splitting
30 Heavy oil 1.1100 1.1102 process.
31 Heavy oil 1.1500 1.4826 To examine the monotonicity in the split samples, the ratio of
32 Oil 1.2230 1.0105 the final plus fraction molecular weight (MWN+) to the
33 Heavy oil 1.3417 1.0173 molecular weight of the final SCN fraction (MWN‐1) is calculated
34 Oil 2.1327 1.1366 for all 43 considered samples, and the results are reported in
35 Oil 1.6082 1.1366
Figure 4 for different splitting techniques. It should be
36 Oil 1.4239 1.1162
37 Oil 1.3382 1.1148
emphasized that to ensure monotonicity, the molecular weight
38 Oil 1.3386 1.1336 of the final plus fraction should be larger than the molecular
39 Oil 1.3287 1.1169 weight of the last SCN fraction. As a result, if a splitting
40 Oil 1.2597 1.1152 technique yields an MWN+/MWN‐1 ratio of smaller than one,
41 Oil 1.7529 1.1146 there exists an inconsistency, and the splitting process may be
42 Oil 1.4096 1.1150 considered invalid.
43 Oil 1.3700 1.1182 As shown in Figure 4, for most cases, the Katz method yielded
an MWN+/MWN‐1 ratio of smaller than one, while in some other
cases, the value of the final plus fraction molecular weight
is a left‐skewed distribution. In other words, the logical value for became negative. Analyzing the results presented in Figure 4, we
the α parameter in gas condensate reservoirs is around one while can also conclude that the splitting results based on the Whitson
for the oil and heavy oil samples, it should be greater than one. method are not monotonic for some samples, i.e., the value of the
Analyzing the results presented in Table 2, one may observe that final plus fraction molecular weight is less than the molecular
for the proposed technique, the evaluated values for α parameter weight of the last SCN fraction. Although the results of the
are around 1.0 for gas condensate samples and larger than 1.0 for Ahmed[12] method are monotonic for the molecular weight, it
oil samples, which is a logical prediction. However, although cannot preserve monotonicity for the specific gravity, as will be
Whitson’s method yields acceptable results for gas condensates, shown below.
the estimated values for α for oil and heavy oil samples are also Figure 6 demonstrates the ratio of the final plus fraction
nearly one, which cannot be correct according to Riazi.[26] specific gravity (SGN+) to the specific gravity of the last SCN
However, the proposed method results in a wider range and fraction (SGN‐1) generated from different splitting methods in all
more reasonable values of α for the oil samples. 43 considered samples. As the results show, the specific gravities
To further analyze the performance of the proposed technique yielded from the proposed method always remain monotonic. By
in terms of yielding an accurate extended composition of the plus contrast, this monotonicity does not exist in most of the cases for
fraction, two oil samples (30 and 32 in Table 1) and two other conventional methods. In other words, the specific gravity

MONTH 2019 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 7

 (a) (b)

(c) (d)

Figure 3. The molecular weight distribution of the plus fraction resulting from the proposed optimization‐based method and Whitson’s method along
with experimental results: (a) oil sample 30; (b) oil sample 32; (c) gas condensate sample 7; and (d) gas condensate sample 15.

of the final plus fraction after the splitting process becomes less order of magnitude as the Whitson and Ahmed methods. Also, the
than the final SCN fraction specific gravity that leads to evaluated mole fractions employing Katz’s method[17] are
inconsistency and unreliability. This was predictable since the unreasonable (negative) in some cases.
conventional methods do not consider the specific gravity The negative mole fraction obtained from Katz’s method is a
distribution in the splitting process. result of its correlative nature. In this method, the mole percent of
Another interesting point is the trend of the MW ratio and each constituent of the plus fraction is correlated to the carbon
SG ratio for the different methods. For the MW ratio, the number in the fraction as:
results of the proposed method using both the distribution
functions are close to each other and similar to Ahmed’s z n = 1.382 05zpf e−0.259 03n (23)
method. Meanwhile, for the SG ratio, the consistency between
the proposed method and the method of Ahmed remains
In the case of C6+ characterization, the resulted split sample
preserved. However, Ahmed’s method does not maintain 12
employing the Katz method gives ∑n = 6 z n > zpf , hence:
monotonicity.
Finally, the mole fraction of the final plus fraction (xN+) for
12
each split sample resulted from different splitting techniques is
z13 + = zpf − ∑ zn < 0 (24)
reported in Table 4. As can be seen in this table, the mole fraction
n=6
of the final plus fraction in the proposed method is in the same

Table 3. The average absolute relative error of the proposed method in terms of predicting the mole fraction of the pseudo‐components comparing
to the existing techniques for the selected samples (the minimum error value in each row is shown in bold)
Average absolute relative error (AARE) (%)

Ahmed
Fluid zPF MWPF SGPF Fluid type Proposed method (gamma) Katz[17] et al.[20] Whitson[4]

1 0.9846 316.9 0.933 Heavy oil 17.91 138.42 35.85 48.53

2 0.3684 198.7 0.841 Oil 23.78 81.70 33.67 26.9
3 0.0999 180.8 0.831 Gas condensate 41.01 40.50 12.39 85.76
4 0.021 936 145.8 0.787 Gas condensate 20.41 43.32 52.09 22.47
Maximum relative error (%) 41.01 138.42 52.09 85.76
Minimum relative error (%) 17.91 40.50 12.39 22.47
Average relative error (%) 25.78 75.99 33.50 45.92

8 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING MONTH 2019

Figure 4. The ratio of the molecular weight of the final plus fraction to the last SCN molecular weight.
Figure 5. The ratio of the specific gravity of the final plus fraction to the last SCN specific gravity.
Figure 6. Predicted phase envelopes employing characterization results
of the proposed method and Whitson’s method for condensate sample E.
MONTH 2019
However, in the case of the C7+ plus fraction, the obtained mole
fraction of the final plus fraction employing Katz’s method does
not become negative.
Based on the presented results, it can be concluded that
all the considered conventional splitting methods lead to
inconsistent results. In contrast, the proposed optimization‐
based splitting method yields monotonic results. In other words,
for all considered samples, the yielded final plus fraction
properties, including molecular weight, specific gravity,
and mole fraction are reasonable and the corresponding
molecular weight and specific gravity distributions are
monotonic.
For most of the samples in this study, the final SCN has been
considered equal to 30 in the characterization process, which is a
reasonable value considering the previous studies in the
literature. However, it should be mentioned that the proposed
algorithm can be applied to split a plus fraction to even higher
values for the final SCN, while preserving the monotonicity and
THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 9
 Phase Behavioural Analysis
Table 4. Mole fractions of the final plus fractions evaluated after An ideal plus fraction splitting process should improve the
splitting employing different splitting methods (negative values are accuracy of the equation of state (EOS) model in predicting the
shown in bold)
fluid properties. In other words, if experimental data on the fluid
Mole fraction of the final plus fraction (xN+) phase behaviour (e.g., saturation pressure and volumetric proper-
ties of the fluid sample over the desired range of temperature and
Proposed Proposed pressure) of a considered sample is available, a good splitting
Fluid method method Ahmed scheme should yield minimum difference in fluid thermodynamic
sample (gamma) (generalized) et al.[20] Katz[17] Whitson[4] properties estimated by the EOS model comparing to the
1 0.0001 0.0001 0.0005 0.0003 0.0022 experimental data.
2 0.0002 0.0002 0.0006 0.0007 0.0022 The Peng‐Robinson (PR) EOS,[35] is used to analyze the phase
3 0.0004 0.0004 0.0008 0.0012 0.0028 behaviour of these fluid samples, which is the most popular EOS
4 0.0003 0.0003 0.0000 0.0001 0.0001 for petroleum reservoir fluids. The thermodynamic properties of
5 0.0002 0.0003 0.0009 0.0008 0.0033 light components (e.g., H2S, N2, CO2, C1, C2, …, n‐C5) and SCN
6 0.0002 0.0002 0.0010 0.0010 0.0037 groups (e.g., C6, C7, …) are obtained from the literature.[12,13]
7 0.0018 0.0018 0.0013 0.0031 0.0048 Furthermore, these properties are predicted using the Twu[25] and
8 0.0007 0.0009 0.0014 0.0027 0.0049 Edmister[36] empirical correlations for the plus fractions and
9 0.0009 0.0015 0.0015 0.0044 0.0062
pseudo components.
10 0.0004 0.0004 0.0016 0.0015 0.0056
11 0.0005 0.0005 0.0016 0.0018 0.0022 To quantitatively analyze the performance of the proposed
12 0.0000 0.0001 0.0001 –0.0007 0.0003 splitting method comparing to the other techniques, four
13 0.0000 0.0000 0.0002 –0.0062 0.0036 fluid samples with available experimental data are considered.
14 0.0000 0.0000 0.0001 –0.0015 0.0007 The compositions of three of these samples along with their
15 0.0004 0.0004 0.0008 –0.0059 0.0026 plus fraction properties are reported in Table 5. These
16 0.0001 0.0001 0.0003 0.0003 0.0009 three fluid samples are called as A, B, and C, which are
17 0.0003 0.0002 0.0008 –0.0055 0.0024 obtained from one of the Iranian oil fields. The fourth fluid
18 0.0043 0.0061 0.0101 0.0019 0.0120 sample, called D, is selected from the Jaubert’s data bank.[33]
19 0.0213 0.0193 0.0212 0.0022 0.0267 The composition of this sample is available in Jaubert et al.,[33]
20 0.0162 0.0136 0.0173 0.0023 0.0211
as sample F6. The most widely‐used splitting methods as
21 0.0052 0.0052 0.0073 0.0025 0.0096
22 0.0065 0.0060 0.0161 0.0025 0.0193 well as the proposed algorithm are employed to split the plus
23 0.0106 0.0178 0.0251 0.0028 0.0313 fractions in the considered samples up to C30+. Then, the
24 0.0053 0.0070 0.0121 0.0030 0.0145 constant composition expansion (CCE) test has been carried out
25 0.0028 0.0029 0.0081 0.0031 0.0114 on each sample at a constant temperature, and molar volumes,
26 0.0035 0.0046 0.0107 0.0033 0.0136
27 0.0036 0.0031 0.0099 0.0035 0.0132
28 0.0237 0.0224 0.0296 0.0036 0.0365
Table 5. The composition of the fluid samples used in a phase
29 0.0660 0.0565 0.0588 0.0068 0.0730
behavioural analysis of the proposed method
30 0.2623 0.2623 0.2078 0.0144 0.2689
31 0.4056 0.4056 0.3243 0.0115 0.4047 Mole fraction
32 0.0219 0.0177 0.0261 0.0054 0.0309
33 0.0600 0.0522 0.0554 –0.1279 0.0809 Component Sample A Sample B Sample C
34 0.0496 0.0407 0.0646 0.0005 0.0931
35 0.0566 0.0658 0.0636 0.0005 0.0920 H2S 0.0204 0.0043 0.0016
36 0.0195 0.0234 0.0282 0.0003 0.0439 N2 0.0022 0.0022 0.0045
37 0.0099 0.0081 0.0166 0.0002 0.0279 CO2 0.0436 0.0245 0.0155
38 0.0520 0.0597 0.0552 0.0005 0.0807 C1 0.2221 0.2582 0.2290
39 0.0257 0.0322 0.0342 0.0004 0.0530 C2 0.0684 0.0792 0.0764
40 0.0172 0.0156 0.0284 0.0004 0.0490 C3 0.0511 0.0562 0.0535
41 0.0107 0.0070 0.0225 0.0003 0.0363 i‐C4 0.0084 0.0100 0.0116
42 0.0164 0.0179 0.0251 0.0003 0.0400 n‐C4 0.0248 0.0285 0.0274
43 0.0204 0.0258 0.0268 0.0003 0.0410 i‐C5 0.0081 0.0100 0.0093
n‐C5 0.0095 0.0117 0.0112
F‐C6 0.0092 0.0131 0.0427
yielding reliable results. It should be mentioned that through the F‐C7 0.0636 0.0660 0.0429
literature, splitting plus fractions to a final SCN as high as 40 is F‐C8 0.0436 0.0459 0.0360
reported.[12] F‐C9 0.0262 0.0426 0.0331
On the contrary, methods such as Katz and Ahmed are F‐C10 0.0316 0.0332 0.0295
mathematical correlations that do not consider specific gravity F‐C11 0.0233 0.0292 0.0272
during characterization and do not preserve monotonicity. C12+ 0.3439 0.2852 0.3486
In addition, due to the correlative nature of these characterization
Plus fraction properties
techniques, the obtained results are unreliable in some test
cases, particularly for high SCN final fractions. It should be noted Fluid sample zPF MWPF SGPF
that, although for smaller values of final SCN these correlations
results may be monotonic, the fact that monotonicity does not A 0.3439 310 0.964
preserve for higher final SCNs, indicates their incorrect trend of B 0.2852 420 0.944
characterization. C 0.3486 370 0.960

10 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING MONTH 2019

 Table 6. The relative error of the EOS‐evaluated parameters with respect to the experimental data employing different splitting methods (the
minimum error value in each row is shown in bold)

Relative error (%)

Proposed
Sample Proposed Method method Ahmed
name Experiment Parameters (Gamma) (Generalized) Katz[17] et al.[20] Whitson[4]

A CCE Test (T = 43 ˚C) Vt/Vs 0.85 0.88 2.23 0.86 1.00

Bt 1.61 1.68 2.49 1.64 2.09
Co 3.15 3.36 45.13 3.15 3.94
CCE Test (T = 121 ˚C) Vt/Vs 0.57 0.61 2.14 0.58 0.82
Bt 4.51 4.53 2.32 4.48 4.95
Co 5.30 6.13 72.67 5.48 6.26
Bubble point Pressure Pb (T = 43 ˚C) 19.63 20.05 23.62 19.74 20.29
Pb (T = 121 ˚C) 11.53 12.00 20.22 11.40 12.47
B CCE Test (T = 49 ˚C) Vt/Vs 0.88 1.07 2.55 1.07 1.22
Bt 2.15 2.26 3.04 2.25 2.48
CCE test (T = 104 ˚C) Vt/Vs 0.73 0.93 2.26 0.92 1.07
Bt 3.81 3.89 2.18 3.86 4.09
Bubble point Pressure Pb (T = 49 ˚C) 16.27 19.44 20.25 19.44 19.58
Pb (T = 104 ˚C) 8.78 11.99 16.97 11.88 12.42
C CCE test (T = 121 ˚C) Vt/Vs 3.02 3.03 2.51 3.30 3.03
ρl 3.77 3.90 419.3 5.01 7.14
CCE Test (T = 145 ˚C) Vt/Vs 5.45 5.46 7.69 5.85 5.44
ρl 2.44 2.53 412.7 4.48 4.88
Bubble point pressure Pb (T = 30 ˚C) 24.43 24.47 44.42 27.85 26.35
Pb (T = 80 ˚C) 17.46 17.53 37.54 20.24 18.15
Pb (T = 121 ˚C) 14.19 14.34 32.94 16.04 17.53
Pb (T = 145 ˚C) 12.32 12.54 30.04 13.56 15.80
D CCE Test (T = 110 ˚C) Vt/Vs 1.67 2.03 3.89 1.83 1.70
ρl 0.24 0.51 14.96 7.22 0.29
Swelling Test Vt/Vs 0.62 0.64 8.58 1.34 1.24
(T = 110 ˚C) ρl 1.11 1.66 19.86 5.93 1.38
Bubble point pressure Pb (T = 110 ˚C) 11.94 13.89 20.78 11.96 9.91
Errors Maximum relative error (%) 24.43 24.47 419.3 27.85 26.35
Minimum relative error (%) 0.24 0.51 2.14 0.58 0.29
Average relative error (%) 6.61 7.09 47.16 7.83 7.61

total formation volume factor, and bubble point pressure of

each sample are evaluated. In the case of sample D, the Table 7. The composition of the fluid samples used in phase envelope
swelling test is also carried out. The relative errors of these prediction
parameters with respect to the results obtained from the
Mole fraction
corresponding experimental data are calculated for the con-
sidered splitting schemes, and the results are summarized in Component Sample E Sample F Sample G
Table 6. In order to study the performance of the considered
methods, the maximum, minimum, and average error percen- H2S 0.0 0.0239 0.0322
tage values for each method are reported at the end of Table 6. N2 0.0527 0.0017 0.0049
CO2 0.0037 0.0331 0.1137
As can be seen, the proposed method yields the best
C1 0.8644 0.8155 0.2736
performance.
C2 0.0436 0.0537 0.0941
The results show that, for most thermodynamic properties C3 0.0147 0.0223 0.0670
of the considered samples, the proposed splitting technique i‐C4 0.0033 0.0151 0.0081
yields the most accurate results. Although the other techniques n‐C4 0.0045 0.0097 0.0317
result into the minimum error in some cases, their performance is i‐C5 0.0021 0.0075 0.0122
not consistent through all cases. By contrast, the proposed n‐C5 0.0015 0.0175 0.0198
technique leads to acceptable results in all the considered cases. F‐C6 0.0022 0.0239 0.0249
Hence, it can be concluded that the proposed technique does not C7+ 0.0073 0.0017 0.3179
sacrifice accuracy to guarantee the monotonicity of the split
Plus fraction properties
sample.
Furthermore, in order to prove the advantages of the proposed Fluid sample zPF MWPF SGPF
characterization method for phase envelope prediction, three
different samples are gathered from the literature with reported E 0.0073 111.9 0.748
experimental data of bubble points and dew points. These F 0.0017 108.7 0.742
G 0.3179 248.1 0.874
hydrocarbon samples include two gas condensate samples and

MONTH 2019 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 11


Figure 7. Predicted phase envelopes employing characterization results
of the proposed method and Whitson’s method for condensate sample F.
one oil sample. Two condensate fluid samples will be denoted by
E and F, which are gathered from studies of Gharesheikhloo and
Moayyedi,[37] and Kamari and Shadizadeh,[38] respectively.
Furthermore, the oil sample G is selected from the Mohebbinia
et al. study.[39] The compositions of these three samples are
reported in Table 7.
The plus fractions of these samples are split employing the
proposed method and the well‐known Whitson method. The phase
envelopes obtained from the split sample employing the proposed
method and the Whitson method along with the experimental data
are presented in Figures 6–8. The presented results clearly show
better performance of the proposed method in predicting the phase
envelope compared to the Whitson method.
Based on these results, it can be concluded that the proposed
method yields more accurate phase envelope predictions in terms
of bubble point and dew point lines, besides leading to monotonic
split samples and predicting more accurate thermodynamic
properties. In order to accurately compare the experimental results
and the predictions shown in Figures 6–8, the average, maximum,
and minimum error percentages are calculated and reported in
Table 8. The results show that the proposed method outperforms
the results of Whitson’s method yielding lower average errors.
It should be noted that the proposed method has advantages and
limitations similar to all other methods. According to the results,
one of the main benefits of the proposed method is preserving the
Figure 8. Predicted phase envelopes employing characterization results
of the proposed method and Whitson’s method for oil sample G.
12 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING
Table 8. Error analysis for predicted phase envelopes employing
characterization results of the proposed method and Whitson’s
method for samples D, E, and F
Proposed method Whitson[4]
Sample
name AARE Max. ARE Min. ARE AARE Max. ARE Min. ARE
D 16.72 73.40 0.33 44.67 123.84 7.27
E 5.73 13.74 0.21 30.76 65.50 7.30
F 10.50 13.88 7.54 30.41 33.12 28.02
monotonicity between the molecular weight and specific gravity.
Also, the proposed method provides a more accurate extended
composition than the available methods. Moreover, the modelling
phase behaviour using the extended composition of the proposed
method is more reliable than the existing methods. However, the
proposed method, similar to all other distribution‐based techni-
ques, is limited to the mathematical models that it employs to
describe the extended compositions of plus fractions. The devel-
opment of these techniques is based on assuming a continuous
molar distribution for pseudo‐components. However, experimental
compositional data for some special samples show discontinuities
that cannot be described using these mathematical distributions.
Finally, it is worth mentioning that in the petroleum industry,
reservoir fluid modelling is one of the most important steps in
reservoir simulation and modelling of flow lines as well as surface
facilities. Having accurate models that can predict the properties and
behaviour of reservoir fluids can significantly contribute to decision
making and determination of the correct strategy for developing a
reservoir. One of the most uncertain parameters of the reservoir fluids
is the plus fraction. An accurate and consistent splitting scheme can
reduce this uncertainty and, as a result, enhances the modelling of
reservoir fluids, which was the main concern of this research.
CONCLUSIONS
In this study, a new optimization‐based method for plus fraction
splitting was proposed. The proposed method employs specific gravity
in the plus fraction splitting process to yield more consistent results.
In other words, the developed method is based on the fact that the
molecular weight is not uniquely related to single carbon number
fractions and the relationship between specific gravity and molecular
weight should also be preserved. The proposed method can be easily
coupled with widely‐used molar distribution models (i.e., exponential,
3‐parameters gamma function, and generalized model) and any other
probability distribution function.
In the proposed optimization‐based splitting technique, the
controlling parameters of the selected distributions are determined
optimally, such that the relation between the molecular weight and
the specific gravity is maintained, enforcing the relation between
the mole fraction and volume fraction of each pseudo‐component.
The performance of the proposed method was evaluated over
different fluid samples covering a wide range of reservoir fluids from
gas condensates to heavy oils. Analyzing the results, it was shown
that the proposed technique yields the most consistent molar
distribution of the pseudo‐components. The results also showed that
contrary to the conventional splitting methods, the proposed method
preserves monotonicity of molecular weight and specific gravity. For
most samples, the conventional methods yielded specific gravity
distributions that did not remain monotonic. Particularly, the final
MONTH 2019
plus fraction specific gravity was smaller than the specific gravity of
the last SCN fraction, which is obviously not reasonable.
To compare the performance of the proposed optimization‐
based technique with respect to the conventional methods in
terms of EOS modelling of the fluid samples (employing PR EOS)
and liquid and vapour properties of the corresponding split
samples were considered. The results showed that the proposed
optimization‐based splitting method yields the best performance
with respect to the experimental data. Furthermore, it was shown
that the proposed method results in more accurate and more
consistent phase envelope predictions concerning the experi-
mental data of three different samples.
It should be emphasized that the proposed method can also be
coupled with any desired distribution function, which is another
advantage of the technique. By considering any desired distribu-
tion function for molecular weight and specific gravity, the
parameters of the selected distribution can be optimized such that
the plus fraction is split in a way such that the results remain
monotonic while ensuring accuracy.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the Abdal Industrial
Project Management (MAPSA) company for providing financial
support and permission to use their PVT Pro software and publish
this paper.
REFERENCES
[1] A. A. Al‐Meshari, W. McCain Jr., Validation of Splitting the
Hydrocarbon Plus Fraction First Step in Tuning Equation-of-
State, SPE Middle East Oil and Gas Show and Conference,
SPE, Manama, 11–14 March 2007.
[2] J. Duan, W. Wang, H. Liu, J. Gong, Fluid Phase Equilibr.
2013, 345, 1.
[3] P. Luo, X. Wang, Y. Gu, Fluid Phase Equilibr. 2010, 291, 103.
[4] C. H. Whitson, Soc. Petrol. Eng. J. 1983, 23, 683.
[5] M. R. Riazi, Ind. Eng. Chem. Res. 1997, 36, 4299.
[6] X. Liang, W. Yan, K. Thomsen, G. M. Kontogeorgis, Fluid
Phase Equilibr. 2014, 375, 254.
[7] G. Moradi, A. Khoshmaram, M. Riazi, J. Petrol. Sci. Eng.
2011, 76, 57.
[8] B. Carreón‐Calderón, V. Uribe‐Vargas, E. Ramírez‐Jaramillo,
M. Ramírez‐de‐Santiago, Ind. Eng. Chem. Res. 2012, 51,
14188.
[9] P. Hosseinifar, S. Jamshidi, J. Petrol. Sci. Eng. 2016, 147, 624.
[10] I. Rodriguez, A. A. Hamouda, SPE Reserv. Eval. Eng. 2010,
13, 283.
[11] M. Riazi, H. Al‐Adwani, A. Bishara, J. Petrol. Sci. Eng. 2004,
42, 195.
[12] T. H. Ahmed, Hydrocarbon Phase Behavior, Gulf Publishing
Company, Houston 1989.
[13] A. Danesh, PVT and Phase Behaviour of Petroleum Reservoir
Fluids, Elsevier, Amsterdam 1998.
[14] P. Hosseinifar, S. Jamshidi, Fluid Phase Equilibr. 2014,
363, 189.
[15] M. Saajanlehto, V. Alopaeus, Fuel 2014, 133, 216.
MONTH 2019
[16] R. Hosein, W. D. McCain Jr., T. Jagai, J. Petrol. Sci. Eng. 2012,
100, 59.
[17] D. L. Katz, J. Petrol. Technol. 1983, 35, 205.
[18] A. Mehrabian, F. Crespo, Oil Gas Sci. Technol. 2011,
66, 1025.
[19] K. Pedersen, P. Thomassen, A. Fredenslund, Report SEP
8207, Denmark Technical University, Lyngby Phase Equili-
bria and Separation Processes, 1982.
[20] T. Ahmed, G. Cady, A. Story, A Generalized Correlation for
Characterizing the Hydrocarbon Heavy Fractions, SPE
Annual Technical Conference and Exhibition, SPE, Las Vegas,
22–26 September 1985.
[21] S. Huang, M. Radosz, Fluid Phase Equilibr. 1991, 66, 1.
[22] C. H. Whitson, I. Søreide, T. Anderson, C7+ Characterization
of Related Equilibrium Fluids using the Gamma Distribu-
tion, C7+ Fraction Characterization, L. G. Chorn, G. A.
Mansoori, Eds., Taylor & Francis, New York 1989, pp. 35–56.
[23] M. Petitfrere, D. V. Nichita, F. Montel, Fluid Phase Equilibr.
2014, 362, 365.
[24] R. L. Cotterman, J. M. Prausnitz, Ind. Eng. Chem. Proc. D.D.
1985, 24, 434.
[25] C. H. Twu, Fluid Phase Equilibr. 1984, 16, 137.
[26] M. R. Riazi, Characterization and Properties of Petroleum
Fractions, ASTM International, West Conshohocken,
PA 2005.
[27] M. R. Riazi, Ind. Eng. Chem. Res. 1989, 28, 1731.
[28] C. H. Whitson, T. F. Anderson, I. Søreide, Chem. Eng.
Commun. 1990, 96, 259.
[29] M. Ghasemi, E. Sedarat, S. Gerami, S. Ebrahimzadeh, M. A.
Emadi, J. Nat. Gas Sci. Eng. 2014, 21, 1.
[30] S. S. Rao, Engineering Optimization: Theory and Practice,
John Wiley & Sons, Hoboken 2009.
[31] K. Folsta, G. Camargo, R. Espósito, Fluid Phase Equilibr.
2010, 292, 87.
[32] K. Krejbjerg, K. Pedersen, Controlling VLLE Equilibrium
with a Cubic EOS in Heavy Oil Modeling, 7th Canadian
International Petroleum Conference, Petroleum Society of
Canada, Calgary, 13–15 June 2006.
[33] J.‐N. Jaubert, L. Avaullee, J.‐F. Souvay, J. Petrol. Sci. Eng.
2002, 34, 65.
[34] J. Y. Zuo, D. Zhang, “Plus Fraction Characterization and PVT
Data Regression for Reservoir Fluids near Critical Condi-
tions,” SPE Asia Pacific Oil and Gas Conference and
Exhibition, SPE, Brisbane, 16–18 October 2000.
[35] D.‐Y. Peng, D. B. Robinson, Ind. Eng. Chem. Fund. 1976,
15, 59.
[36] W. C. Edmister, Petrol. Refiner. 1958, 37, 173.
[37] A. A. Gharesheikhloo, M. Moayyedi, Iran. J. Chem. Chem.
Eng. 2014, 33, 83.
[38] E. Kamari, S. Shadizadeh, Petrol. Sci. Technol. 2012, 30, 2114.
[39] S. Mohebbinia, K. Sepehrnoori, R. T. Johns, A. K. N. Korrani,
J. Petrol. Sci. Eng. 2017, 158, 693.
Manuscript received October 22, 2018; revised manuscript received
January 16, 2019; accepted for publication January 26, 2019.
THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 13