Mohammadi, Ramjugernath.-Rapid Model For The Estimation of Dew Point Pressures in Gas Condensate Systems

See
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/282611028
Rapid Model for the Estimation of Dew Point

Pressures in Gas Condensate Systems.
ARTICLE in JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS · OCTOBER 2015

Impact Factor: 3
READS
52
5 AUTHORS, INCLUDING:
Amir H. Mohammadi Deresh Ramjugernath

546 PUBLICATIONS 4,561 CITATIONS University of KwaZulu-Natal
331 PUBLICATIONS 1,900 CITATIONS
SEE PROFILE
SEE PROFILE
All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Amir H. Mohammadi
letting you access and read them immediately. Retrieved on: 26 December 2015
JID: JTICE
ARTICLE IN PRESS [m5G;October 30, 2015;14:12]
Journal of the Taiwan Institute of Chemical Engineers xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Journal of the Taiwan Institute of Chemical Engineers

journal homepage: www.elsevier.com/locate/jtice
Rapid model for the estimation of dew point pressures in gas

condensate systems
Q1 Arash Kamari a, Mehdi Sattari a,b, Amir H. Mohammadi a,c,∗, Deresh Ramjugernath a
a
Thermodynamics Research Unit, School of Engineering, University of KwaZulu-Natal, Howard College Campus, King George V Avenue, Durban 4041, South Africa
b
Department of Chemical Engineering, Buinzahra Branch, Islamic Azad University, Buinzahra, Iran
c
Institut de Recherche en Génie Chimique et Pétrolier (IRGCP), Paris Cedex, France
a r t i c l e i n f o a b s t r a c t
Article history: The production of condensate, in addition to gas can improve the recovery factor of gas condensate reservoirs,
Received 16 April 2015 as well as increase the economic feasibility of the reservoir. Dew point pressure (DPP) is regarded as one of
Revised 8 September 2015
the vital parameters for characterizing a gas condensate reservoir. The accurate estimation of DPP is however
Accepted 6 October 2015
still a major challenge for reservoir engineers. In this study, a consistent, accurate, and simple-to-use model
Available online xxx
is proposed for the prediction of DPP in gas condensate reservoirs using a reliable soft-computing approach
Keywords: known as gene expression programming (GEP). The computational approach utilizes a comprehensive dataset
Dew point pressure of DPP, as well as properties of C7+ , reservoir temperature, and hydrocarbon and non-hydrocarbon reservoir
Temperature fluid compositions. The model proposed is compared to three well-known empirical correlations. The pro-
Gene expression programming (GEP) posed model produces an average absolute relative deviation of approximately 7.88% and is clearly superior
Empirical correlation to previously published methods for the prediction of dew point pressure in gas condensate reservoirs.
Gas condensate reservoir
© 2015 Taiwan Institute of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
1 1. Introduction pressure. In this type of DPP, with a decrease in dew point pressure, 21
at a specific pressure liquids start to form. At such a condition, the 22
2 There is a growing realization of the importance of gas condensate single phase fluid transforms into a liquid and gas phase. This pres- 23
3 reservoirs as a considerable hydrocarbon resource in terms of energy sure is often called the dew point or saturation pressure. By further 24
4 supply. In these reservoirs, well deliverability often reduces when the decreasing the pressure, the volume of liquid reaches a maximum. 25
5 bottom-hole pressure drops below the dew point pressure (DPP). As a Further decrease of pressure causes vaporization of the condensates 26
6 definition, dew point pressure is the pressure at which a considerably and consequently a decrease in the amount of liquid. Such reservoirs 27
7 larger amount of the gas phase is in equilibrium with a significantly are called retrograde condensate gas reservoirs [4]. The phase dia- 28
8 smaller amount of liquid phase [1]. As a result, the dew point pressure gram for gas condensate reservoirs clearly shows this behavior (Fig. 29
9 plays a significant role in hydrocarbon reservoir engineering, result- 1). Therefore, it can be concluded that the first type of dew point pres- 30
10 ing in the reservoirs being classified using dew point pressure. Gas sure does not matter in the performance of gas reservoirs, while the 31
11 condensate reservoirs are different in their thermodynamic and flow second type is very important in gas well performance. 32
12 behavior compared to common gas reservoirs. As a consequence, the accurate prediction of dew point pressure 33
13 In the operation of gas reservoirs, there are two kinds of dew in gas condensate reservoirs is important to evaluate their perfor- 34
14 points that engineers are focused on [2,3]. The first type is the nor- mance because of a reduction in the rate of gas condensate produc- 35
15 mal dew point pressure which occurs at low pressures when dry gas tion with an increase of liquid [5]. A number of researchers have stud- 36
16 is compressed to a point at which the first droplets begin to form. This ied the effect of dew point pressure on well productivity, e.g. Fevang 37
17 type of dew point pressure is not important for engineers because [6], Afidick and Bette. [7], Fan et al. [8], Barnum et al. [9], and Eil- 38
Q2
18 the pressure to achieve this dew point pressure is lower than atmo- erts and Smits [10]. The studies conclude that there is a considerable 39
19 spheric pressure, while the pressure of reservoirs are greater than at- reduction in well generation in gas condensate wells under certain 40
20 mospheric pressure [3,4]. The second type is retrograde dew point conditions, e.g. near wellbore condensate aggregation. The determi- 41
nation of the dew point pressure in gas condensate reservoirs has 42
∗
been investigated by several researchers who have attempted to de- 43
Corresponding author.
E-mail addresses: arash.kamari@yahoo.com (A. Kamari), a.h.m@irgcp.fr,
termine this important property either experimentally or theoreti- 44
amir_h_mohammadi@yahoo.com (A.H. Mohammadi), ramjuger@ukzn.ac.za cally. For the determination of the DPP experimentally, the constant 45
(D. Ramjugernath). composition expansion (CCE) and constant volume depletion (CVD) 46
http://dx.doi.org/10.1016/j.jtice.2015.10.011
1876-1070/© 2015 Taiwan Institute of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
Please cite this article as: A. Kamari et al., Rapid model for the estimation of dew point pressures in gas condensate systems, Journal of the
Taiwan Institute of Chemical Engineers (2015), http://dx.doi.org/10.1016/j.jtice.2015.10.011
JID: JTICE
2 A. Kamari et al. / Journal of the Taiwan Institute of Chemical Engineers xxx (2015) xxx–xxx
Table 1
Ranges, averages and units of the variables implemented for the development of the
Nomenclature
GEP-based model for the prediction of dew point pressures.
DPP dew point pressure Property Unit Min. Max. Avg.

CCE constant composition expansion
Dew-point pressure, DPP Psia 1405 10,790 4747.222
CVD constant volume depletion Reservoir temperature, TR °F 40 320 205.146
GA genetic algorithm Molecular weight, MW C7+ – 106 235 148.189
GEP gene expression programming Specific gravity, SGC7+ – 0.7330 0.8681 0.788
Nitrogen, N2 Mole fraction 0.0000 0.4322 0.010
GP gene programming
Carbon dioxide, CO2 Mole fraction 0.0000 0.9192 0.015
ET expression tree Hydrogen sulfide, H2 S Mole fraction 0.0000 0.2986 0.006
EoS equations of state Methane, C1 Mole fraction 0.0349 0.9668 0.802
MWC7+ molecular weight of pentane plus fraction Ethane, C2 Mole fraction 0.0037 0.1513 0.057
SPC7+ specific gravity of pentane plus fraction Propane, C3 Mole fraction 0.0011 0.1090 0.030
Butanes, C4 Mole fraction 0.0017 0.2030 0.020
TR reservoir temperature,°F
Pentanes, C5 Mole fraction 0.0006 0.0631 0.012
Pd dew point pressure, Psia Hexanes, C6 Mole fraction 0.0004 0.0510 0.009
Ei percent relative deviation Heptane-plus, C7+ Mole fraction 0.0019 0.1356 0.037
ARD average relative deviation
AARD average absolute relative deviation
RMSE root mean square error
SD standard deviation for the model development. Eilerts and Smith [10] proposed a rela- 69
R2 coefficient of determination tionship between temperature, pressure, composition, boiling point 70
of the fluid, and gas oil ratio based on research in the Palam field. In 71
47 are the two most commonly-used laboratory measurement methods 1945, Olds and Lacey [13] developed a correlation to predict the dew 72
48 [1]. Laboratory measurement of DPP is reliable, however, it is expen- point pressure (in graphical and tabular forms) by using the charac- 73
49 sive and time-consuming. Hence, there is a preference to determine teristics of oil and gas samples obtained from the primary separator 74
50 DPP using empirical methods equations of state (EoS) [1]. of a well in the Paloma field. They also studied the impact of the elim- 75
51 EoSs are usually not able to accurately simulate the phase behav- ination of intermediate molecular weight on DPP. They showed that 76
52 ior of light oil and gas condensate reservoirs, particularly in the retro- the intermediate molecular weight components have a significant in- 77
53 grade region [2]. Hence, a consistent, accurate, efficient, and simple- fluence on DPP. 78
54 to-use model is proposed in this study for the determination of dew Olds and Lacey [14] experimentally studied the volumetric behav- 79
55 point pressures of retrograde gas condensate reservoirs. To this end, ior for various mixtures of gas condensate samples which were col- 80
56 the gene expression programming (GEP) [11] computational scheme lected from the San Joaquin Valley field. The correlation developed by 81
57 was implemented to develop the model using a database of 562 ex- them provided a relationship between the retrograde DPP and gas– 82
58 perimental data points from CVD tests. The model developed is com- oil ratio, temperature, and stock tank API oil gravity. The results ob- 83
59 pared to three well-known empirically derived correlations. An anal- tained showed that the effect of temperature was minimal in compar- 84
60 ysis is also conducted to detect the suspended and/or outlier data ison with the influence of modifying the compositions. Modification 85
61 points existing in the dataset of DPP. of the composition was undertaken by eliminating the intermediate 86
components [15]. In 1950, Reamer [16] investigated existing correla- 87
62 2. Literature survey tions, with respect to higher gas–oil ratio samples by combining five 88
different pairs of fluids from a typical field in Louisiana. In their study, 89
63 Over the years, many research studies have been conducted to the effect of temperature and gas–oil ratio on DPP was investigated. 90
64 propose a global model for the prediction of DPP in gas condensate They concluded that the complexity of the effect of composition on 91
65 systems, on the basis of temperature, hydrocarbon composition, and DPP is the main reason for a lack of a global model for predicting DPP. 92
66 C7+ . In 1942, Kurata [12] developed a correlation to predict the criti- In 1952, Organick [17] studied the dew point pressure in conden- 93
67 cal properties of volatile hydrocarbon mixtures. To this end, they ne- sate gas and volatile-oil mixtures. They introduced a simple corre- 94
68 glected the effect of composition due to a limited number of DPP data lation in the form of working charts which had an error of approxi- 95
Fig. 1. A representative phase diagram of the gas condensate fluid [8].
JID: JTICE
A. Kamari et al. / Journal of the Taiwan Institute of Chemical Engineers xxx (2015) xxx–xxx 3
temperature, primary separator gas–oil ratio, pseudo reduced pres- 108

sure and temperature, relative densities of separator, and heptane- 109
plus fraction. 110
In 1996, Carlson [21] studied the effect of H2 S on DPP. They con- 111
cluded that an increase in H2 S content decreases the volume of liq- 112
uid drop out. Marruffo et al. [22] proposed a correlation which used 113
146 PVT analyses data from Western Venezuela (Anaco) fields to 114
determine the DPP in gas condensate reservoirs. In 2002, Elsharkawy 115
[15] developed a relationship between DPP of the gas condensate 116
reservoirs and some properties, including temperature, molecular 117
Fig. 2. A typical two-gene chromosome with its corresponding mathematical weight, composition of hydrocarbon fluid and specific gravity of 118
expression. the C7+ components. They used 340 experimentally measured data 119
within a pressure range of 1560–11830 psi and temperature rang- 120
96 mately 8% [15]. The correlation developed was not able to describe ing from 40 to 340°F. Gonzales and Startzman [23] used neural net- 121
97 some materials like pure components and non-complex mixtures. work modeling to predict DPP. The reservoir temperature, hydrocar- 122
98 Nemeth [18] developed an extended relationship between dew point bon and non-hydrocarbon composition, molecular weight and spe- 123
99 pressure, temperature, composition, and characteristics of the C7+ cific gravity of the C7+ were applied as the inputs of the network de- 124
100 fraction of the hydrocarbon fluid. They used multiple-variable regres- veloped. In 2008, Shokir [1] proposed a model using mathematical 125
101 sion analysis in their correlation development. In 1996, Crogh [19] genetic programming, neglecting the effect of specific gravity of C7+ . 126
102 improved the Nemeth[18] correlation. Their correlation enabled bet- The study showed that the model developed is more acceptable (pre- 127
103 ter prediction of DPP, neglecting reservoir temperature. Humoud and cise) compared to previous methods. Sarkar and Todd [24] proposed 128
104 Al-Marhoun [20] developed an empirical model using different gas equations of state to model the behavior of the reservoir fluid phase, 129
105 condensate samples extracted from the Middle East. The correlation but they could not accurately simulate the complex hydrocarbon be- 130
106 developed is a relationship between the DPP of a gas condensate fluid havior, such as the retrograde gas condensate reservoir phase. 131
107 with its reservoir temperature, the primary separator pressure and
Fig. 3. Comparison between the data calculated by Eq. (1) and actual magnitudes of dew point pressures with regard to line Y = X and residual error percentage.
JID: JTICE
Fig. 4. Comparison between the data calculated by the Nemeth and Kennedy’s model [42] and actual magnitudes of dew point pressures with regard to line Y = X and residual
error percentage.
132 3. Dataset chemical engineering disciplines. The GA works on the same princi- 153
ples as the human brain and body’s genetic coding to define individu- 154
133 The DPP in gas condensate reservoirs is strongly influenced by the als, and Darwin’s rule to solve the desired problem [28]. Replacement 155
134 properties of C7+ (molecular and specific gravity), reservoir tempera- of the present populations related to a solution with new solutions 156
Q3
135 ture, and compositions of hydrocarbon and non-hydrocarbon com- is done by progressing the optimization process. A major difference 157
136 ponents [1,15,18,23,25,26]. An extensive dataset is thus required if between genetic algorithm modeling and numerical modeling is the 158
137 one want to develop a good correlative model. A comprehensive arrangement of variables; the arrangement of variables in GA is the 159
138 databank consisting of 562 experimental data [27] for DPP obtained genotype while in the numerical modeling it is the phenotype [29]. 160
139 from CVD tests was collected. The database comprises the values for Over the years, the GA evolutionary mathematical approach has 161
140 DPP (Pd , Psia), molecular weight for heptane plus fractions (MWC7+ ), been modified and a new form, namely genetic programming (GP), 162
141 reservoir temperature (TR ,°F), specific gravity for heptane plus frac- introduced. The GP approach has been increasingly employed to solve 163
142 tions (SGC7+ ), compositions of hydrocarbons including methane (C1 ), complex and/or relatively simple problems in engineering targets 164
143 ethane (C2 ), propane (C3 ), butanes (C4 ), pentanes (C5 ), hexanes (C6 ), [30–32]. Later, the GEP evolutionary soft-computing technique was 165
144 heptane-plus (C7+ ), and compositions of non-hydrocarbons including presented by Ferreira [11]. The GP approach uses parse trees to rep- 166
145 nitrogen (N2 ), carbon dioxide (CO2 ), and hydrogen sulfide (H2 S). Table resent the solutions, while they are characterized as numbers in GA 167
146 1 summarizes the ranges, averages, and units of the abovementioned techniques [33]. As a result, the population individuals in the GP com- 168
147 parameters. As can be seen in Table 1, the parameters presented cover putational scheme are symbolic expression trees (ETs) [34], while 169
148 a wide range of DPP, reservoir temperature, etc. they are linear chromosomes in the GEP approach, which are later 170
translated into the ETs [35]. Functions (arithmetic operations) and 171
149 4. Methodology for model development terminals (variables) and constants are two main elements of a gene 172
for chromosomes in GEP approaches, involving a head and tail [36]. 173
150 In recent years, genetic algorithms (GA) have been widely imple- In other words, the GEP approach processes the problems by using 174
151 mented as a reliable intelligent optimization method to solve regres- two entities including the ETs and the chromosomes. To demonstrate 175
152 sion and classification problems, in particular in the petroleum and the GEP approach, an example is provided where w, j and v can be 176
JID: JTICE
Fig. 5. Comparison between the data calculated by the Shokir’s model [1] and actual magnitudes of dew point pressures with regard to line Y = X and residual error percentage.
177 considered as the terminals, as well as –, +, / can be expressed for tion). Therefore, about 80% of the available data points were assigned 202
178 the functions. Fig. 2 indicates a chromosome built of two genes with for the training, learning and model development stage and the re- 203
179 algebraic expression considering the Karva language. maining 20% of the data was categorized as validation or test data. To 204
180 In order to produce the new generation, with a minimum ex- evaluate the quality of a desired solution, an appropriate fitness func- 205
181 pected deviation using the GEP algorithm, the most commonly- tion is set during the prediction process by GEP algorithm [37]. Two 206
182 used genetic operators include selection, replacement, mutation, and reliable fitness functions, namely R-squared error and average abso- 207
183 crossover [37]. The selection operator in the GEP algorithm is applied lute relative deviation (AARD) were used to evaluate the performance 208
184 for increasing the possible survival, using the fitness of individuals of the GEP algorithm in the prediction of DPP. 209
185 [38]. In other words, it helps to improve the solving of the problem
186 at hand. The mutation operator is applied to prevent the convergence 5. Result and discussion 210
187 problem in the first steps of running the algorithm and also to im-
188 prove the variety of the expressions [39]. Additionally, the crossover 5.1. The proposed model 211
189 operator influences the structures in the populations and converts
190 the arrangements of instructions between two tournament winners As indicated earlier, there are a number of different factors which 212
191 [40]. Replacement enables timeworn populations which have a high affect the prediction capability of GEP-based models. Hence, many 213
192 deviation in order to be replaced with new individuals. The process attempts and computation run on a trial and error basis were per- 214
193 of effecting genetic operations continues until an acceptable and ac- formed to obtain an optimum and acceptable model in terms of preci- 215
194 curate result is reached. sion and simplicity. During the development of the GEP-based model 216
195 In the development of a model for DPP in gas condensate reser- for the prediction of DPP in gas condensate reservoirs, it was noticed 217
196 voirs by applying the GEP methodology, the collected dataset is di- that an increase in the number of permitted genes in the applied in- 218
197 vided into two groups of data, namely the training and test sets. dividual, and also the maximum depth of ET, has an effect on the do- 219
198 The training dataset is employed in the development/learning of the main size of the solution space; the complexity of the evolved func- 220
199 model (i.e. to fit the GEP model evolution). The test or validation tion rises and the run-time of the process also increases. The preci- 221
200 dataset is utilized for evaluating the capability and the prediction sion of models developed based on the GEP evolutionary approach 222
201 performance of the learned GEP-based model (i.e. external evalua- normally also depends on those factors, in that precision is increased 223
JID: JTICE
Fig. 6. Comparison between the data calculated by the Elsharkawy’s model [41] and actual magnitudes of dew point pressures with regard to line Y = X and residual error
percentage.
Fig. 7. The performance of Eq. (1), as well as the comparative methods in predicting dew point pressures with respect to AARD.
224 with an increase in the number of genes and the depth of ET. Conse- anes, heptane-plus, and compositions of non-hydrocarbons including 230
225 quently, we applied two genes, AARD and R-squared as fitness func- nitrogen, carbon dioxide, hydrogen sulfide were used as input vari- 231
226 tions, and a function set including ∗, +, –, ln and / in order to develop ables. The optimum GEP-based model obtained for the estimation of 232
227 the GEP-based model. Specific gravity and molecular weight for hep- DPP in gas condensate reservoirs is as follows: 233
228 tane plus fractions, reservoir temperature, compositions of hydrocar-
229 bons including methane, ethane, propane, butanes, pentanes, hex-
Pd = A + B (1)
13.145 − 4.942 ZC1 + 1961.7 ZC2 − 6212.71 ZC4 + 39335.07 (ZC4 ) + 2097 ZC5 − 3451.17 ZC6 + 201.93ZH2 S − 0.065224TR
2
A= (2)
0.0031904 TR + 0.094398
234
JID: JTICE
Fig. 8. The performance of Eq. (1), as well as the comparative methods in predicting dew point pressures with respect to the study of the influence of temperature on dew point
pressure trend.
Fig. 9. Determination of doubtful data points in the dew point pressure dataset, collected during development of the GEP-based scheme, using Hat values.
1367.4 + 9.98 ZC1 MWC7+ − 1697.6 ZC3 − 5096.8 ZC7 + 358.07 ln (ZC7+ ) + 933.35 ZCO2 + 1909.7 ZN2
B= (3)
1.0214 − SGC7+
235 where Pd (Psia) denotes the DPP in gas condensate reservoirs, Eq. (1), as well as a relative error distribution plot is presented. Table 2 248
236 MWC7+ is the molecular weight of the heptane-plus fraction, SGC7+ summarizes the calculated errors for data predicted by the GEP-based 249
237 is the specific gravity of heptane plus fraction, TR is reservoir tem- model developed for DPP in gas condensate reservoirs. The proposed 250
238 perature (°F) and ZC1 , ZC2 , ZC3 , ZC4 , ZC5 , ZC6 , ZC7 + , ZCO2 , ZN2 , ZH2 S are model based on the GEP approach has an overall AARD of 7.8% and an 251
239 compositions of methane, ethane, propane, butanes, pentanes, hex- R-squared equal to 0.89. These values are acceptable for the proposed 252
240 anes, heptane-plus, carbon dioxide, nitrogen, and hydrogen sulfide, model taking into account the dataset used in its development. Fig. 253
241 respectively. 3 shows the cross-plot of the actual data against the predicted val- 254
242 5.2. Capability and precision evaluation ues using Eq. (1), as well as the relative error distribution plot. As can 255
be seen in Fig. 3, there is good agreement between the actual data 256
243 In order to evaluate the capability of the model in predicting DPP, and values calculated using the GEP-based model proposed in this 257
244 we used some important error functions including AARD, R-squared study. There is an acceptable match, as observed in the top panel of 258
245 (R2 ), average relative deviation (ARD), root mean square error (RMSE) Fig. 3 between the reported and DPP values estimated using the GEP 259
246 and standard deviation (SD). Furthermore, a graphical analysis, in model. Additionally, the bottom panel of Fig. 3 illustrates satisfactory 260
247 terms of cross-plot of actual data against the predicted values, using distribution of relative deviation around the zero line. 261
JID: JTICE
Table 2 6. Conclusions 309

The overall performance of the model developed, for the training and testing phases,
as well as the comparison with other models in terms of statistical error analysis.
In this study a novel model for the accurate calculation of dew 310
Method AARDa % ARDb % SDc RMSEd R2e point pressure in gas condensate reservoirs which is based on the 311
Elsharkawy’s model [41] 15.3 -9.667 0.201 891.075 0.759

gene expression programming evolutionary approach is introduced. 312
Shokir’s model [1] 11.0 -0.728 0.151 704.177 0.818 A comprehensive databank which included dew point pressure, 313
Nemeth and Kennedy’s model [42] 8.6 4.794 0.095 611.369 0.886 reservoir temperature, characteristics of the heptane-plus fraction, 314
This study, overall 7.8 0.871 0.088 549.219 0.890 as well as compositions of hydrocarbons and non-hydrocarbons was 315
This study, training set 7.8 0.759 0.078 524.082 0.891
collated from literature. A comparative analysis was undertaken in or- 316
This study, testing set 8.0 1.321 0.040 640.348 0.882

der to determine the accuracy and capability of the model developed 317
X −Xrep./pred
a
AARD% = 1n ni=1 |Ei %| where Ei % = [ exp Xexp ] × 100 ⇒ i = 1, 2, 3, . . . , n in this study against three well-known published correlations. It was 318

b
ARD% = 1n ni=1 Ei % observed that the model developed performs superior to the other
319

RMSE = 1n ni=1 (Xi exp − Xi rep./pred )
2
c correlations with respect to calculated statistical error parameters. 320

i=1 (X(i) exp −X(i )rep./pred )
N 2
The proposed model has an overall AARD of 7.88% and an R-squared 321
d
R2 = 1 − N
i (X(i)rep./pred −averageX(i)rep/pred )2 equal to 0.89. 322
Appendix A. Equations for the comparative models used 323

262 A performance of Eq. (1) was further evaluated by comparing the in this study 324
263 results obtained with three well-known empirically derived correla-
264 tions, viz. Elsharkawy [41], Shokir [1], and Nemeth and Kennedy [42]. The models developed by Elsharkawy [41], Shokir [1], and Nemeth 325
265 The equations for these empirical methods are presented in Appendix and Kennedy [42] were used in this study in order to compare the re- 326
266 A. Table 2 also provides a systematic comparison between the actual sults obtained using the proposed GEP-based model. The correlations 327
267 DPP data, the output values of the Elsharkawy [41], Shokir [1], and associated with the comparative models are as follows: 328
268 Nemeth and Kennedy [42] methods, as well as the results of GEP- Elsharkawy [41] 329
269 based model. Moreover, Figs. 4–6 illustrate the results obtained by
270 Nemeth and Kennedy [42], Shokir [1], and Elsharkawy [41] methods, Pd = A0 + A1 T + A2 zH2 S + A3 zCO2 + A4 zN2 + A5 zC1 + A6 zC2
271 respectively. As can be seen in these figures, the agreement between + A7 zC3 + A8 zC4 + A9 zC5 + A10 zC6 + A11 zC7+
272 the values predicted by these comparative methods and actual data
MWC7+
273 of DPP is not as good as with the GEP-based model developed in this + A12 MWC7+ + A13 SGC7+ + A14 (zC7+ MWC7+ ) + A15
274 study. Fig. 7 shows the calculated AARD for the comparative meth- SGC7+

275 ods as well as the GEP-based model developed. It can be observed zC7+ MWC 7+ zC7+
276 from Fig. 7 that the GEP-based model has better performance than + A16 + A17
SGC7+ zC1 + zC2
277 the other correlations with respect to the calculated AARD values. The
278 AARDs for the Elsharkawy [41], Shokir [1], and Nemeth and Kennedy zC7+
+ A18 , (4)
279 [42] methods are 15.3, 11.0 and 8.6, respectively, while it is 7.8% for zC3 + zC4 + zC5 + zC6
280 Eq. (1).
Shokir [1] 330
281 Fig. 8 shows a further comparison of the DPP values predicted
282 using Eq. (1) with actual data, as well as the calculated DPP values Pd = B1 + B2 + B3 + B4 , (5)
283 using the comparative techniques, taking into account the influence
where 331
284 of reservoir temperature on DPP. This figure also shows that the val-
285 ues predicted using Eq. (1) are closer to the actual data compared to B1 = 201.875481 (zC7+ (((Tr (((zC3 − (zH2 S − ZCO2 ))
286 the other methods. Furthermore, it illustrates that DPP has an incre- − (zC6 − (ZCO2 − zC4 ))) − ((zC2 )) − ((zC4 (((ZCO2 − zC4 )
287 mental trend with an increase in temperature. The comparative anal-
− (MWC7+ − zN2 )) − (MWC7+ 2 zC5 ))) − zC7+ ))
288 ysis confirms that the GEP-based model proposed is able to calcu-
289 late the desired parameter (i.e. the DPP in gas condensate reservoirs) − (zH2 S − ((zN2 (T (zC1 2 − zC7+ )))
290 with greater accuracy and consistency. The model developed in this − (MWC7+ − (zC2 − zH2 S)))))) + 38456.87953 zC5
291 study also has a smaller number of adjustable parameters in compar-
292 ison with the comparative methods, implying that model optimiza- B2 = 0.000007((T ((((zCO2 − MWC7+ ) − zC7+ )((T − MWC7+ )
293 tion and development for GEP-based models are faster, less labori- − (zCO2 − T ))) − ((zH2 S − T )((MWC7+ − zC3 )MWC7+ ))))zN2 )
294 ous, and less costly. In other words, the GEP-based model proposed + 225500.9399zC5
295 for the accurate estimation of DPP in gas condensate reservoirs can
B3 = 120586.9719(zC1 ((((zH2 S zC3 ) − (zC5 − zC7+ ))zH2 S )
296 be considered as more simple-to-use than the comparable methods.
− (((zC7+ − zC1 )(zC7+ − zC6 )) − (zH2 S zN2 2 ))))
297 5.3. Detection of the outlier DPP data points available in the databank + 72.6908MWC7+
B4 = −1962.40851(zC5 (MWC7+ − zC1 2 ))
298 The Leverage scheme was used to detect outlier DPP data points
− 253385.67764((zC7+ ((zCO2 zC3 )
299 existing in the database used. To find doubtful DPP data points, the
300 Williams’ plot is drawn by calculation of the Hat values, as illustrated − (zC4 − zC7+ )))(zCO2 (zC3 (zC3 − MWC7+ ))))
301 in Fig. 9. The GEP-based model proposed in this study for the predic- − 13358.59271zC4 + 4676.933602zC2 − 6567.9
302 tion of DPP in gas condensate reservoirs displays higher performance
Nemeth and Kennedy [42] 332
303 statistically, as the main portions of the data are located within the
304 domains of 0 < H < 0.07473 and –3 < R < 3. It can be observed, based ln (Pd ) = A1 (zC2 + zCO2 + zH2 S + zC6 + 2(zC3 + zC4 ) + zC5
305 on the Leverage scheme as illustrated in Fig. 9 that there are only

zC1
306 13 data points (among a total of 562 data points) outside of the ac- + 0.4zC1 + 0.2zN2 ) + A2 SGC7+ + A3
zC7+ + 0.002
307 ceptable range of the proposed technique. Further details on outlier
308 detection are available in literature [43–45]. + A4 T + A5 (zC7+ MWC7+ ) + A6 (zC7+ MWC7+ )2
JID: JTICE
Table A1 [11] Ferreira C. Gene expression programming: a new adaptive algorithm for solving 371
List of the adjustable parameters applied in the Elsharkawy [41], and Nemeth and problems. Complex Syst. 2001;13:87–129. 372
Kennedy [42] methods. [12] Kurata DLVKF. Critical properties of volatile hydrocarbon mixtures. Am. Inst. 373
Chem. Eng. AICHE J. 1942;38:995–1021. 374
Parameter Elsharkawy’s model [41] Nemeth and Kennedy’s model [42] [13] Olds BSR, Lacey W. Volumetric and phase behavior of oil and gas from Paloma 375
Field. Trans, AIME 1945;160:77–99. 376
A0 4268.85 [14] Olds BSR, Lacey W. Volumetric and viscosity studies of oil and gas from a San 377
A1 0.094056 –2.0623054 Joaquin valleyfield. Trans. AIME 1949;179:287. 378
A2 –7157.87 6.6259728 [15] Elsharkawy AM. Predicting the dew point pressure for gas condensate reservoirs: 379
A3 –4540.58 –4.4670559 × 10−3 empirical models and equations of state. Fluid Phase Equilib. 2002;193:147–65. 380
A4 –4663.55 1.0448346 × 10−4 [16] Reamer BSH. Volumetric behavior of oil and gas from a Louisiana field. Trans. 381
A5 –1357.56 3.2673714 × 10−2 AIME 1950;189:261–8. 382
A6 –7776.1 –3.6453277 × 10−3 [17] Organick BGE. Prediction of saturation pressures for condensate-gas and volatile- 383
A7 –9967.99 7.4299951 × 10−5 oil mixtures. Trans. AIME 1952;195:135. 384
A8 –4257.1 –1.1381195 × 10−1 [18] Nemeth HKL. A correlation of dewpoint pressure with fluid composition and tem- 385
perature. SPE J. 1967;7:99–104. 386
A9 –1417.1 6.2476497 × 10−4
[19] Crogh A. Improved correlations for retrograde gases. College Station: Texas A & M 387
A10 691.5298 –1.0716866 × 10−6
University; 1996. 388
A11 40660.36 1.746622 × 10 + 1
[20] Humoud AA, Al-Marhoun MA. A new correlation for gas-condensate dewpoint 389
A12 205.26 pressure prediction. In: SPE Middle East Oil Show. Bahrain: Society of Petroleum 390
A13 –7260.32 Engineers Inc.; 2001. 391
A14 –352.413 [21] Carlson WBCMR. Obtaining PVT data for very sour retrograde condensate gas and 392
A15 –114.519 volatile oil reservoirs: a multi-disciplinary approach. Calgary, Alberta, Canada: 393
A16 8.133 SPE Gas Technology Symposium; 1996. 394
A17 94.916 [22] Marruffo JMI, Him J, Rojas G. Statistical forecast models to determine retrograde 395
A18 238.252 dew pressure and C7+ percentage of gas condensates on basis of production 396
test data of eastern Venezuelan reservoirs. In: SPE Latin American and Caribbean 397
Petroleum Engineering Conference, Buenos Aires, Argentina; 2001. 398
[23] González MABA, Startzman R. Improved neural-network model predicts dew-
point pressure of retrograde gases. J. Petrol. Sci. Eng. 2003;37:183–94.
399
400
MWC7+
+ A7 (zC7+ MWC7+ ) + A8
3 [24] Sarkar ADR, Todd A. Phase behavior modeling of gas-condensatefluids using an 401
SGC7+ + 0.0001 equation of state. In: SPE Annual Technical Conference and Exhibition; 1991. 402
2 [25] Akbari M, Jalali F. Dew point pressure estimation of gas condensate reservoirs, 403
MWC7+ using Artificial Neural Network (ANN). In: EUROPEC/EAGE Conference and Exhi- 404
+ A9 bition, Society of Petroleum Engineers, London, UK; 2007. 405
SGC7+ + 0.0001 [26] Nowroozi S, Ranjbar M, Hashemipour H, Schaffie M. Development of a neural 406
3 fuzzy system for advanced prediction of dew point pressure in gas condensate 407
MWC7+ reservoirs. Fuel Process. Technol. 2009;90:452–7. 408
+ A10 + A11 , (6) [27] Nemeth LK. A correlation of dew-point pressure with reservoir fluid composition 409
SGC7+ + 0.0001 and temperature. Texas A & M University; 1966. 410
[28] Romero C, Carter J. Using genetic algorithms for reservoir characterisation. J. 411
333 where in the equations Pd denotes the dew point pressure Petrol. Sci. Eng. 2001;31:113–23. 412
334 in gas condensate reservoirs, MWC7+ is the molecular weight [29] Xue Y, Cheng L, Mou J, Zhao W. A new fracture prediction method by combining 413
335 of the heptane-plus fraction, SGC7+ is the specific gravity of genetic algorithm with neural network in low-permeability reservoirs. J. Petrol. 414
Sci. Eng. 2014;121:159–66. 415
336 the heptane plus fraction, TR is the reservoir temperature and [30] Alavi AH, Gandomi AH, Nejad HC, Mollahasani A, Rashed A. Design equations for 416
337 ZC1 , ZC2 , ZC3 , ZC4 , ZC5 , ZC6 , ZC7 + , ZCO2 , ZN2 , ZH2 S are compositions of prediction of pressuremeter soil deformation moduli utilizing expression pro- 417
338 methane, ethane, propane, butanes, pentanes, hexanes, heptane-plus, gramming systems. Neural Comput. Appl. 2013;23:1771–86. 418
[31] Azamathulla HM, Ghani AA. Genetic programming to predict river pipeline scour. 419
339 carbon dioxide, nitrogen, and hydrogen sulfide, respectively. Further- J. Pipeline Syst. Eng. Pract. 2010;1:127–32. 420
340 more, the adjustable parameters of Elsharkawy [41], and Nemeth and [32] Azamathulla HM, Ghani AA. Genetic programming for predicting longitudinal 421
341 Kennedy [42] methods are presented in Table A1. dispersion coefficients in streams. Water Res. Manage. 2011;25:1537–44. 422
[33] Kaydani H, Mohebbi A, Eftekhari M. Permeability estimation in heterogeneous oil 423
reservoirs by multi-gene genetic programming algorithm. J. Petrol. Sci. Eng. 2014. 424
342 References
[34] Ferreira C. Gene expression programming: mathematical modeling by an artificial 425
intelligence. 2nd ed. Springer-Verlag New York, Inc.; 2006. 426
343 [1] Shokir EME-M. Dewpoint pressure model for gas condensate reservoirs based on [35] Gharagheizi F, Ilani-Kashkouli P, Farahani N, Mohammadi AH. Gene expres- 427
344 genetic programming. Energy Fuels 2008;22:3194–200. sion programming strategy for estimation of flash point temperature of non- 428
345 [2] Hadi Rostami-Hosseinkhani FE, Mowla Dariush. Application of expert systems for electrolyte organic compounds. Fluid Phase Equilib. 2012;329:71–7. 429
346 accurate determination of dew-point pressure of gas condensate reservoirs. J. Nat. [36] Teodorescu L, Sherwood D. High energy physics event selection with gene expres- 430
347 Gas Sci. Eng. 2014;18:296e303. sion programming. Comput. Phys. Commun. 2008;178:409–19. 431
348 [3] Milad Arabloo AS, Farhad Gharagheizi, Mohammadi Amir H. Toward a predictive [37] Kaydani H, Najafzadeh M, Hajizadeh A. A new correlation for calculating carbon 432
349 model for estimating dew point pressure in gas condensate systems. Fuel Process. dioxide minimum miscibility pressure based on multi-gene genetic program- 433
350 Technol. 2013;116:317–24. ming. J. Nat. Gas Sci. Eng. 2014;21:625–30. 434
351 [4] Nemeth HTKLK. A correlation of dewpoint pressure with fluid composition and [38] Cranganu C, Bautu E. Using gene expression programming to estimate sonic log 435
352 temperature. SPE J. 1967;7:99–104. distributions based on the natural gamma ray and deep resistivity logs: a case 436
353 [5] Seyed Mohammad Javad Majidi AS, Arabloo Milad, Mahdikhani- study from the Anadarko Basin, Oklahoma. J. Petrol. Sci. Eng. 2010;70:243–55. 437
Q4 354 Soleymanloo Ramin, Masihi Mohsen . Evolving an accurate model based on [39] Seavey KC, Jones AT, Kordon AK, Smits GF. Hybrid genetic programming−first- 438
355 machine learning approach for prediction of dew-point pressure in gas conden- principles approach to process and product modeling. Ind. Eng. Chem. Res. 439
356 sate reservoirs. Chem. Eng. Res. Des. 2013:1347–59. 2010;49:2273–85. 440
357 [6] Fevang CHWØ. Modeling gas-condensate well deliverability. SPE Reservoir Eng. [40] Gandomi AH, Alavi AH, Sahab MG. New formulation for compressive strength 441
358 1996;11:221–30. of CFRP confined concrete cylinders using linear genetic programming. Mater. 442
359 [7] Afidick NJKD, Bette S. Productionperformanceofaretrogradegasreservoir: a case Struct. 2010;43:963–83. 443
360 study of the Arunfield. In: SPE Asia Pacific Oil and Gas conference, Melbourne, [41] Elsharkawy AM. Predicting the dew point pressure for gas condensate reservoirs: 444
361 Australia. Society of Petroleum Engineers, Inc.; 1994 Copyright 1994, 1994. empirical models and equations of state. Fluid Phase Equilib. 2002;193:147–65. 445
362 [8] Fan L, Harris, Jamaluddin BW, Kamath AJ, Mott J, Pope R, Shandrygin GA, A, Whit- [42] Nemeth L, Kennedy H. A correlation of dewpoint pressure with fluid composition 446
363 son CH. Understanding gas-condensate reservoirs. Oilfield Rev. 1998;10:16–25. and temperature. SPE J. 1967;7:99–104. 447
364 [9] Barnum RS, Brinkman FP, Richardson TW, Spillette AG. Gas condensate reservoir [43] Gramatica P. Principles of QSAR models validation: internal and external. QSAR 448
365 behaviour: productivity and recovery reduction due to condensation. In: SPE An- Comb. Sci. 2007;26:694–701. 449
366 nual Technical Conference and Exhibition, Dallas, Texas. Society of Petroleum En- [44] Goodall CR. 13 Computation using the QR decomposition. Handbook Stat. 450
367 gineers, Inc.; 1995. 1993;9:467–508. 451
368 [10] Eilerts K, Smith RV. Specific volumes and phase-boundary properties of [45] Mohammadi AH, Eslamimanesh A, Gharagheizi F, Richon D. A novel method 452
369 separator-gas and liquid-hydrocarbon mixtures. Bartlesville, Okla. (USA): Bureau for evaluation of asphaltene precipitation titration data. Chem. Eng. Sci. 453
370 of Mines; 1942. p. 74. 2012;78:181–5. 454

Mohammadi, Ramjugernath.-Rapid Model For The Estimation of Dew Point Pressures in Gas Condensate Systems

Uploaded by

Copyright:

Available Formats

You might also like

Mohammadi, Ramjugernath.-Rapid Model For The Estimation of Dew Point Pressures in Gas Condensate Systems

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mohammadi, Ramjugernath.-Rapid Model For The Estimation of Dew Point Pressures in Gas Condensate Systems

Uploaded by

Copyright:

Available Formats

See

Rapid Model for the Estimation of Dew Point

ARTICLE in JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS · OCTOBER 2015

Amir H. Mohammadi Deresh Ramjugernath

Journal of the Taiwan Institute of Chemical Engineers xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of the Taiwan Institute of Chemical Engineers

Rapid model for the estimation of dew point pressures in gas

DPP dew point pressure Property Unit Min. Max. Avg.

Fig. 1. A representative phase diagram of the gas condensate ﬂuid [8].

temperature, primary separator gas–oil ratio, pseudo reduced pres- 108

Table 2 6. Conclusions 309

Elsharkawy’s model [41] 15.3 -9.667 0.201 891.075 0.759

Appendix A. Equations for the comparative models used 323

You might also like