SPE-201170-MS

A Novel Correlation to Predict Gas Flow Rates Utilizing Artificial

Intelligence: An Industrial 4.0 Approach

Shams Kalam, Mohammad Rasheed Khan, and Zeeshan Tariq, King Fahd University of Petroleum & Minerals;
Faisal Anwar Siddique, Pakistan Petroleum Limited; Abdulazeez Abdulraheem and Rizwan Ahmed Khan, King
Fahd University of Petroleum & Minerals

Copyright 2019, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE/PAPG Pakistan Section Annual Technical Symposium and Exhibition held in Islamabad, Pakistan, 18 - 20
November 2019.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Reservoir and production engineers rely heavily on well production rates to optimize well activities such as
ensuring optimum reservoir monitoring. Individual gas well rates are not readily available, rather, they can
be estimated thru multi-phase flow meter (MPFM) and well test analysis. These methods are associated with
certain limitations such as high cost, high uncertainty, and technically elaborate calculations. Consequently,
empirical and numerical calculations are employed with well test data to calculate daily rates. These
practices lead to inaccurate gas rate estimations.
A model with an ability to provide accurate estimates of gas rates for a gas reservoir can serve as a
handy tool for the subsurface engineers in addressing well and reservoir optimization strategies. This work
presents artificial intelligence models to estimate gas rates in a gas field containing ten wells. The aim is to
develop a correlation that is simple and easy to incorporate yet providing robust answers on a global scale.
Multiple machine learning tools are employed. These include; Artificial Neural Network (ANN), Functional
Network (FN), and Adaptive Neuro Fuzzy Inference System (ANFIS).
Production data from a dry gas field X was used for the model development. Data cleaning and data
reduction steps were carried out to ensure the input parameters for the proposed model are physically
relevant and accurate. Missing these steps would result in the development of an erroneous correlation, i.e.,
garbage -in garbage-out (GIGO). This led to finalization of certain basic well-head parameters which are
available at any typical well and had direct impact on the output production rate. The target parameter for
model training is the gas rate. A rigorous comparison between the investigated artificial intelligence models
was conducted by calculating average absolute percentage error (AAPE) and coefficient of determination.
The comparative analysis shows that the intelligent model is able to predict the gas rate in condensate wells
with accuracy in excess of 90%. Examples of such large accuracy has not been reported previously.
ANN performs a step ahead as compared to the various intelligent algorithms used in this study. This
paper sheds light on the potential of the Industrial Revolution 4.0 for the Pakistani Oil and Gas Sector.
Data-driven artificial intelligent models are capable of validating the well test and multiphase flow meter
2 SPE-201170-MS

results. In addition, it can prove to be a vital tool in an engineer's tool-kit to reduce uncertainties in gas
rate measurements.

Introduction
Controlling hydrocarbon production from wells using chokes is a norm in the petroleum industry. This helps
in maintaining stable pressures at the downstream, apply the requisite backpressure, and limit formation
damage (Boyan Guo and Ghalambor 2014). Chokes are employed both at the wellhead and also in the
subsurface to prevent abnormal pressure fluctuations from damaging the surface and sub-surface equipment
(Nasriani and Kalantari 2011; Mirzaei-Paiaman and Salavati 2013). Flow through chokes can be of two types
depending upon the downstream to upstream pressure ratio, termed as critical ratio (Nejatian et al. 2014;
Khan, Tariq, et al. 2018b). Various studies in the literature are focused to predict the flow of hydrocarbons
through a choke. Mostly, nozzles, fixed orifice, and adjustable types are investigated (Morris 1996). Tangren
et al. (1949) made the first real attempt to study flow through restrictions by introducing a generalized
approach using analytical techniques. Following this work, many correlations were developed, which are
available in the literature and are either theoretical or empirical in nature (Khan et al. 2019).
The theoretical approach started with Tangren et al. (1949), but Fortunati (1972) was the first to present
two correlations for critical and sub-critical flow regimes. Based on Tangren's assumptions and extending
Ross's (1960) work, Ashford (1974) presented a model for critical two-phase flow across chokes. Various
aspects of flow across chokes such as the critical pressure ratio, mass flow-rate, and choke discharge
coefficient among others were addressed by different researchers (Ashford and Pierce 1975; Pilehvari 1980,
1981; Sachdeva et al. 1986; Perkins 1993; Boyun Guo et al. 2002). Although lacking universal applications,
empirical methods are preferred by the industry due to its robustness and simplicity in usage as compared
to theoretical techniques.
Gilbert (1954) was the pioneer who introduced a generalized form of the empirical equation for the flow
through chokes utilizing an extensive database of well tests from an American field flowing in critical regime
and therefore, does not include the downstream pressure term. This generalized form is reproduced below:

Where,

• q = Liquid Flow Rate, STB/Day

• Pwh = Well Head Pressure, psia

• D = Choke Size, 64th inch

• R = Gas-Liquid-Ratio, Mscf/Stb

• A1, A2, A3 are constants

This work was the foundation of several subsequent models (Baxendell 1958; Ros 1960; Achong
1961) that modified the coefficients (A1, A2, A3) of the generalized form of Gilbert's equation to yield
new correlations applicable to their specific cases. Work on subcritical flow was carried out by Ross
(1961), which was then extended by Poettmann and Beck (1963). Correlations for water/gas system in
dimensionless form was proposed by Omana et al. (1969). Major focus has been on the critical flow regime
and a detailed reading on the correlations is available in the literature (Al-Attar and Abdul-Majeed 1988;
Osman and Dokla 1990; Abdul-Majeed and Maha 1991; Elgibaly and Nashawi 1998; Alrumah and Bizanti
2007; Al-attar and Arab 2009; Bairamzadeh and Ghanaatpisheh 2015). Rawlins and Schellhardt (1935
Ref. 47) presented equation for gas flow through chokes for critical flow conditions which is based on the
SPE-201170-MS 3

upstream pressure only. While Szilas (1975 Ref. 48 and 49) published another form of the gas flow equation
for field units, which is applicable for both at and above critical pressure ratio.
Most of the developed correlations in the literature are based on the Gilbert's model, which has its own
inherent limitations (Ghareeb and Elgaghah 2007). Moreover, the published models lack the universality
sought by the industry, however, key identifications can be derived from each study to result in a better
understanding of multi-phase flow (Khan, Tariq, et al. 2018b). Well production data is available readily and
in abundance, therefore can be utilized to come up with better models. Recently, the use of such massive
amounts of data and soft-computing techniques have been applied to enhance the ability to estimate flow-
rates in real-time (Gholgheysari Gorjaei et al. 2015; Seidi and Sayahi 2015; Ghorbani et al. 2017; Rashid
et al. 2019). Artificial intelligence has been aptly utilized for numerous petroleum engineering related
problems (Ahmadi and Golshadi 2012; Ahmadi et al. 2014; Ahmadi and Bahadori 2015; Tariq et al. 2016;
Khan et al. 2018; Khan et al. 2018; Khan et al. 2019; Kalam et al. 2016; Mahboob et al. 2016).
The aim of this work is to develop a new correlation utilizing artificial intelligence techniques that
performs to provide precise gas flowrate estimates and can be used universally, eliminating any complexities
involved in the black-box concept of the artificial intelligence algorithms.

Methodology and Results

Reservoir Description
Reservoir formation was deposited in and around field area where it is shallow marine to Fluvial-Deltaic
sands. The structure comprises a large north-south orientated anticline. The reservoir is relatively uniform
thickness across the entire field structure. The thickness averages 45m and comprises an area of more than
170 km2. The trap is three-way closure anticline with fault bounded one side. Gas in-place are in range of
2.5-3.0 Tcf. Gas composition is mainly methane with 20% inerts. The porosity and permeability averages
at 15% and 50mD respectively.

Artificial Neural Networks (ANN)

Neural Networks mimic the architecture and functioning of a human nervous setup. Multiple "neurons"
make up the structure of the neural network. ANN performs to yield a meaningful result from the input
data. A standalone ANN model consists of three layers in series (input/hidden/output). The data is fed into
the input layer which transfers it to the hidden layer for processing and finally, generating the data into
the output layer. Throughout, the layers consist of various neurons that do all the data processing, which
is dictated by an activation function.
ANN Model Development. ANN technique was applied to build a new empirical model that evaluates
the gas flowrate (Qgas) dictated by the flowing well-head pressure, upstream temperature, choke size, and
flowrate of condensate and water. These variables are input neurons, while there is only one hidden layer
comprising of 19 neurons, as a result of the sensitivity analysis on the number of neurons. Different transfer
functions were employed for the hidden and output layers. Tan-sigmoid was utilized for the former case,
whereas, Linear was used in the latter case. After analyzing various algorithms for training the ANN model,
back-propagation Levenberg-Marquardt was applied.
17097 data points from 7 wells were used to train a feedforward neural network (FFNN) model, while
4846 data points from 3 wells were used to test the correlation. Various combination of weights/biases were
tested through multiple realizations to avoid the local minima. The result of these realizations was judged
based on the least mean average absolute percentage error.
ANN Model Results. The optimum weights and bias values for the gas flow rate are listed in Table A
shown in the appendix. The ANN topology of the new model is displayed in Fig. A, can be found in the
4 SPE-201170-MS

appendix. Mathematical description of the model is presented in equations 1 to 3. Table 1 shows minimum
and maximum values of input and output parameters used for the training.

Table 1—Range for input and target parameters incorporated for the model development.

Parameters Minimum Value Maximum Value

Output Parameter

Qgas, MMScfd 1.82 87.71

Input Parameter

PFWH, Psig 47.62 1590

TFWH, °F 0 147

CH, % 4 100

QCond, BCPD 0 711.16

QWater, BWPD 0 1508.92

(1)

(2)

(3)
Where,
 j = Number of hidden layer neurons
 i = Number of input layer neurons
 xi = Input parameters (normalized)
 bj = Bias for hidden layer
 bk = Bias for output layer
Wij = Weights between input and hidden layer
Wjk = Weights between hidden and output Layer
f = Transfer function
N = Subscript ‘N’ shows normalized parameter
Nh = Total number of neurons in hidden layer
Ni = Total number of inputs
 nhj = jth neuron in hidden layer
Qgas = Gas flow rate, MMScfd
PFWH = Flowing wellhead pressure, psig
TFWH = Flowing wellhead temperature, °F
CH = Choke size, %
QCond = Condensate flowrate, BCPD
QWater = Water flowrate. BWTD
Procedure to apply the new correlation
1. The input parameters should be normalized for the new correlation using equation 4.

(4)
SPE-201170-MS 5

Use -1 for minimum value of ‘y’ and +1 for the maximum value. The maximum and minimum
values of ‘x’ which will be the input parameter can be found in Table 1. Normalized equations for
each input parameters are presented in equations 5 to 9.
(5)
(6)

(7)

(8)

(9)
2. The output denormalization is to be done using equation 10:

(10)

For the output equation, -1 for minimum value of ‘x’ and +1 for the maximum value. The maximum and
minimum values of ‘y’ which will be the target parameter can be found in Table 1.
After substituting values from the respective tables in equation 10, the resultant output from the ANN-
model developed in this work for gas Q is presented by equation 11:

(11)

Generally, to establish quality of a developed/produced correlation the mean/average absolute Error (M/
AAPE) and the correlation-coefficient of determination (R2) are strong statistical measures. These are
provided in Table 2. Scatter plot for training and testing data are presented in Figs. 1 and 2 respectively
which shows the good quality of the ANN model for Qgas.

Table 2—Statistical measures for the training and testing regimes.

Data Type AAPE, % R2

Training (Seen) 5.844 0.9661

Testing (Unseen) 7.386 0.9532

6 SPE-201170-MS

Figure 1—Scatter plot for training data.

Figure 2—Scatter plot for testing data.

Functional Network (FN)

Functional networks are relatively a recent but more prevailing alternate to neural networks (Castillo, 1998).
They are capable of using a combination of domain and data knowledge as compared to neural networks
where only data knowledge is utilized (Castillo et al., 1999). Initialization of the network topology can be
done through modeling of real-world properties. Using this derived and initialized topology, a much simpler
yet comparable topology can be obtained through functional equations. Functional networks can yield
optimum results with data only as well, however, they are prime candidates for solving problems consisting
of both data and domain knowledge (Castillo et al., 2000). Moreover, they have been succesfully used
SPE-201170-MS 7

for classification as well as prediction problems. Furthermore, they are capable of handling a large range
of function approximations in engineering and statistical problems. In functional network there are five
methods, namely the Exhaustive Search, forward-method, backward-method, backward--forward method
and the forward--backward method. Details of aforementioned methods can be further studied in Castillo et
al. (1999). Application of functional network is widely found in petroleum engineering to predict petroleum
reservoir parameters such as permeability, porosity, and water saturation (Anifowose et al., 2013). A general-
framework for the utilization of FN models for time-series modeling/prediction is discussed in literature
(Castillo et al., 2001; Castillo and Gutiérrez, 1998).
FN is widely applied in petroleum engineering applications. Some of the examples but not limited to
are well log prediction (Tariq et al., 2018a, 2018c), petrophysical parameters estimation (Anifowose et al.,
2013; El-Sebakhy et al., 2012), well performance prediction (Bello and Asafa, 2014; Tariq et al., 2018b),
Pressure Volume Temperature (PVT) predictions (Khoukhi et al., 2011) among others.

FN Model Development
FN model was developed to predict gas flowrate using Backward Elimination (BE) learning algorithm. A
total of four learning algorithms were tested namely backward elimination (BE), forward-backward (FB),
exhaustive search (ES), and backward-forward (BF). Among them BE resulted in a better model. Gas
flowrate was set as the target parameter while flowing well-head pressure, upstream temperature, choke
size, flowrate of condensate and water were set as the input variables. Same dataset was used to train and
test the FN BE model. The result of these realizations was judged based on the least mean average absolute
percentage error and highest R2. During training FN BE resulted in 8.9% error with R2 of 0.98, while during
testing the model resulted in 12% error and R2 of 0.91 as shown in Fig. 3.

Figure 3—Training and Testing cross-plots using FNBE model.

8 SPE-201170-MS

Adaptive Neuro Fuzzy Inference System (ANFIS)

ANFIS is a combination of ANN and Fuzzy logic. It was developed in late 1990's. The application of
ANFIS is widely found in various fields of engineering. In this study, an ANFIS model was developed
to predict flowrate of gas using Substractive Clustering (SC) learning algorithm. A total of two different
learning algorithms were tested namely Substractive Clustering, and Grid Partitioning (GP). Among them
SC resulted in a better model. Gas flow rate (Qgas) was predicted as a function of flowing well head pressure
and temperature, choke size, flow rate of condensate and water. The result of multiple realizations was
judged based on the least mean average absolute percentage error and highest R2. During training ANFIS-
SC resulted in 7.1% error with R2 of 0.985, while during testing it resulted in 14% error and R2 of 0.95
as presented in Fig. 4.

Figure 4—Training and Testing cross-plots using ANFIS-SC model.

Comparison
AAPE and R2 were calculated for both training and testing data. These are reported for different artificial
intelligence tools i.e. ANN, ANFIS and FN as shown in Figs. 5 and 6. A comparative analysis shows that
ANN is found to be the best tool among all since it gives lowest AAPE for both testing and training data.
Fig. 5 shows that training data R2 for ANN, FN and ANFIS are almost similar, while testing data R2 for FN
is the lowest as depicted in Fig. 6. ANN and ANFIS result in similar R2 for testing data. In a nut shell, ANN
provides best results as well as mathematical equations for the newly developed correlation.
SPE-201170-MS 9

Figure 5—AAPE and R2 for training data.

Figure 6—AAPE and R2 for testing data.

A comparative analysis was performed among actual field data, ANN, ANFIS and FN shown in Fig. 7.
For this purpose, randomly selected 100 blind data points were selected from the actual field data. Average
values for those inputs are shown in Table 3. ANN predicted better than ANFIS and FN depicted in Fig. 7.

Table 3—Average values of input parameters for comparation.

Input Parameter Average Value

PFWH, Psig 116.22

TFWH, °F 121.54

CH, % 95.00

QCond, BCPD 18.87

QWater, BWPD 40.86

10 SPE-201170-MS

Figure 7—Comparison among Qgas, ANN, ANFIS and FN.

Conclusions
Following conclusions are drawn from this study:
1. Three different artificial intelligence tools (ANN, FN and ANFIS) were applied on the data set to
estimate gas rate.
2. ANN was found to be the best tool as it gives lowest AAPE, highest R2, and provides the mathematical
model which can be used without knowledge of AI.
3. Newly developed ANN model can be used as an empirical correlation to estimate gas rate as a function
of wellhead flowing temperature and pressure, choke size, and flow rate of condensate and water.
4. The optimization was carried out in finding the proper number of hidden layer neurons as well as
weights and biases.
5. The newly developed correlation will help as a quick tool to predict well gas rates of a dry gas field
X without using multi-phase flow meter or well test analysis.

References
1. Abdul-Majeed, Ghassan H., and R. Abdul-Ameer Maha. 1991. "Correlations Developed
To Predict Two-Phase Flow Through Wellhead Chokes." Journal of Canadian Petroleum
Technology 30 (06). https://doi.org/10.2118/91-06-05.
2. Achong I. 1961. "Revised Bean Performance Formula for Lake Maracaibo Wells."
3. Ahmadi, Mohammad-Ali, and Alireza Bahadori. 2015. "A LSSVM Approach for Determining
Well Placement and Conning Phenomena in Horizontal Wells." Fuel 153 (August): 276–83.
https://doi.org/10.1016/j.fuel.2015.02.094.
4. Ahmadi, Mohammad Ali, Sohrab Zendehboudi, Lesley A. James, Ali Elkamel, Maurice
Dusseault, Ioannis Chatzis, and Ali Lohi. 2014. "New Tools to Determine Bubble Point
Pressure of Crude Oils: Experimental and Modeling Study." Journal of Petroleum Science and
Engineering 123 (November): 207–16. https://doi.org/10.1016/j.petrol.2014.08.018.
SPE-201170-MS 11

5. Al-Attar, H.H., and G.H. Abdul-Majeed. 1988. "Revised Bean Performance Equation for East
Baghdad Oil Wells." SPE Production Engineering 3 (01): 127–31. https://doi.org/10.2118/13742-
PA.
6. Al-attar, Hazim H, and United Arab. 2009. "SPE 120788 New Correlations for Critical and
Subcritical Two-Phase Flow Through Surface Chokes in High-Rate Oil Wells," no. 1949: 1–17.
https://doi.org/10.2118/120788-PA.
7. Ali Ahmadi, Mohammad, and Mohammad Golshadi. 2012. "Neural Network Based Swarm
Concept for Prediction Asphaltene Precipitation Due to Natural Depletion." Journal of Petroleum
Science and Engineering 98–99 (November): 40–49. https://doi.org/10.1016/j.petrol.2012.08.011.
8. Alrumah, Muhammad Khalid, and Mohamed S. Bizanti. 2007. "New Multiphase Choke
Correlations for Kuwait Fields." In SPE Middle East Oil and Gas Show and Conference. Society
of Petroleum Engineers. https://doi.org/10.2118/105103-MS.
9. Ashford, F.E. 1974. "An Evaluation of Critical Multiphase Flow Performance Through Wellhead
Chokes." Journal of Petroleum Technology 26 (08): 843–50. https://doi.org/10.2118/4541-PA.
10. Ashford, F.E., and P.E. Pierce. 1975. "Determining Multiphase Pressure Drops and Flow
Capacities in Down-Hole Safety Valves." Journal of Petroleum Technology 27 (09): 1145–52.
https://doi.org/10.2118/5161-PA.
11. Bairamzadeh, Sina, and Ehsan Ghanaatpisheh. 2015. "A New Choke Correlation to Predict
Liquid Flow Rate." Science International 27 (1): 271–74.
12. Baxendell, P B. 1958. "Producing Wells on Casing Flow-An Analysis of Flowing Pressure
Gradients." Petroleum Transactions, AIME 213: 202–6.
13. Elgibaly, A.A.M., and I.S. Nashawi. 1998. "New Correlations For Critical And Subcritical Two-
Phase Flow Through Wellhead Chokes." Journal of Canadian Petroleum Technology 37 (06).
https://doi.org/10.2118/98-06-04.
14. Fortunati, F. 1972. "Two-Phase Flow through Wellhead Chokes." In SPE European Spring
Meeting. Society of Petroleum Engineers. https://doi.org/10.2118/3742-MS.
15. Ghareeb, Mohamed, and Shedid Elgaghah. 2007. "A New Correlation for Calculating Wellhead
Production Considering Influences of Temperature, GOR, and Water-Cut for Artificially Lifted
Wells." In International Petroleum Technology Conference. International Petroleum Technology
Conference. https://doi.org/10.2523/IPTC-11101-MS.
16. Gholgheysari Gorjaei, Reza, Reza Songolzadeh, Mohammad Torkaman, Mohsen Safari, and
Ghassem Zargar. 2015. "A Novel PSO-LSSVM Model for Predicting Liquid Rate of Two Phase
Flow through Wellhead Chokes." Journal of Natural Gas Science and Engineering 24 (May):
228–37. https://doi.org/10.1016/j.jngse.2015.03.013.
17. Ghorbani, Hamzeh, Jamshid Moghadasi, and David A. Wood. 2017. "Prediction of Gas Flow
Rates from Gas Condensate Reservoirs through Wellhead Chokes Using a Firefly Optimization
Algorithm." Journal of Natural Gas Science and Engineering 45 (September): 256–71. https://
doi.org/10.1016/j.jngse.2017.04.034.
18. Gilbert, WE. 1954. "Flowing and Gas-Lift Well Performance." API Drilling and Production
Practice 20 (1954): 126–57. https://doi.org/10.2118.
19. Guo, Boyan, and Ali Ghalambor. 2014. Natural Gas Engineering Handbook. Elsevier.
20. Guo, Boyun, Ali Soud Al-Bemani, and Ali Ghalambor. 2002. "Applicability of
Sachdeva&Apos;s Choke Flow Model in Southwest Louisiana Gas Condensate Wells." In SPE
Gas Technology Symposium. Society of Petroleum Engineers. https://doi.org/10.2118/75507-MS.
21. Khan, Mohammad Rasheed, Sami Alnuaim, Zeeshan Tariq, and Abdulazeez Abdulraheem.
2019. "Machine Learning Application for Oil Rate Prediction in Artificial Gas Lift Wells." In
12 SPE-201170-MS

SPE Middle East Oil and Gas Show and Conference. Society of Petroleum Engineers. https://
doi.org/10.2118/194713-MS.
22. Khan, Mohammad Rasheed, Ahmed Sadeed, Abdul Asad, Zeeshan Tariq, and Muhammad
Tauqeer. 2018. "Maximizing Oil Recovery in a Naturally Fractured Carbonate Reservoir Using
Computational Intelligence Based on Particle Swarm Optimization." In PAPG/SPE Pakistan
Section Annual Technical Conference and Exhibition. Society of Petroleum Engineers. https://
doi.org/10.2118/195664-MS.
23. Khan, Mohammad Rasheed, Zeeshan Tariq, and Abdulazeez Abdulraheem. 2018a. "Machine
Learning Derived Correlation to Determine Water Saturation in Complex Lithologies." In SPE
Kingdom of Saudi Arabia Annual Technical Symposium and Exhibition. Society of Petroleum
Engineers. https://doi.org/10.2118/192307-MS.
24. ———. 2018b. "Utilizing State of the Art Computational Intelligence to Estimate Oil Flow Rate
in Artificial Lift Wells." In SPE Kingdom of Saudi Arabia Annual Technical Symposium and
Exhibition. Society of Petroleum Engineers. https://doi.org/10.2118/192321-MS.
25. Mirzaei-Paiaman, A., and S. Salavati. 2013. "A New Empirical Correlation for Sonic
Simultaneous Flow of Oil and Gas through Wellhead Chokes for Persian Oil Fields." Energy
Sources, Part A: Recovery, Utilization, and Environmental Effects 35 (9): 817–25. https://
doi.org/10.1080/15567031003773304.
26. Morris, S.D. 1996. "Choke Pressure in Pipeline Restrictions." Journal of Hazardous Materials 50
(1): 65–69. https://doi.org/10.1016/0304-3894(96)01786-4.
27. Nasriani, Hamid Reza, and Azim Kalantari ASL. 2011. "Two-Phase Flow Choke Performance in
High Rate Gas Condensate Wells." In SPE Asia Pacific Oil and Gas Conference and Exhibition.
Society of Petroleum Engineers. https://doi.org/10.2118/145576-MS.
28. Nejatian, Ibrahim, Mojtaba Kanani, Milad Arabloo, Alireza Bahadori, and Sohrab Zendehboudi.
2014. "Prediction of Natural Gas Flow through Chokes Using Support Vector Machine
Algorithm." Journal of Natural Gas Science and Engineering 18 (May): 155–63. https://
doi.org/10.1016/j.jngse.2014.02.008.
29. Omana, R., C. Houssiere, Kermit E. Brown, James P. Brill, and Richard E. Thompson. 1969.
"Multiphase Flow Through Chokes." In Fall Meeting of the Society of Petroleum Engineers of
AIME. Society of Petroleum Engineers. https://doi.org/10.2118/2682-MS.
30. Osman, Mohammed E., and Mahmoud E. Dokla. 1990. "Gas Condensate Flow Through
Chokes." In European Petroleum Conference. Society of Petroleum Engineers. https://
doi.org/10.2118/20988-MS.
31. Perkins, T.K. 1993. "Critical and Subcritical Flow of Multiphase Mixtures Through Chokes."
SPE Drilling & Completion 8 (04): 271–76. https://doi.org/10.2118/20633-PA.
32. Pilehvari, Ali A. 1980. "Experimental Study of Subcritical Two-Phase Flow through Wellhead
Chokes." University of Tulsa.
33. ———. 1981. "Experimental Study of Subcritical Two-Phase Flow through Wellhead Chokes."
University of Tulsa.
34. Poettmann, F H, and R L Beck. 1963. "New Charts Developed to Predict Gas-Liquid Flow
through Chokes." World Oil 184 (3): 95–100.
35. Rashid, Saeed, Ali Ghamartale, Jassem Abbasi, Hoda Darvish, and Afshin Tatar. 2019.
"Prediction of Multiphase Critical Choke Flow Behavior by Using a Rigorous Artificial
Neural Network Method." Flow Measurement and Instrumentation, June, 101579. https://
doi.org/10.1016/j.flowmeasinst.2019.101579.
SPE-201170-MS 13

36. Ros, N. C. J. 1960. "An Analysis of Critical Simultaneous Gas/Liquid Flow through a Restriction
and Its Application to Flowmetering." Applied Scientific Research 9 (1): 374–88. https://
doi.org/10.1007/BF00382215.
37. ———. 1961. "Flow-Meter Formulae for Critical Gas/Liquid Flow through a Restriction."
Applied Scientific Research 10 (1): 295–96. https://doi.org/10.1007/BF00411922.
38. Sachdeva, R., Z. Schmidt, J.P. Brill, and R.M. Blais. 1986. "Two-Phase Flow Through Chokes."
In SPE Annual Technical Conference and Exhibition. Society of Petroleum Engineers. https://
doi.org/10.2118/15657-MS.
39. Seidi, Siamak, and Tofigh Sayahi. 2015. "A New Correlation for Prediction of Sub-Critical Two-
Phase Flow Pressure Drop through Large-Sized Wellhead Chokes." Journal of Natural Gas
Science and Engineering 26 (September): 264–78. https://doi.org/10.1016/j.jngse.2015.06.025.
40. Tangren, R. F., C. H. Dodge, and H. S. Seifert. 1949. "Compressibility Effects in Two-Phase
Flow." Journal of Applied Physics 20 (7): 637–45. https://doi.org/10.1063/1.1698449.
41. Zeeshan, Tariq, Sami Alnuaim, Abdulazeez Abdulraheem, and Mohammad Rasheed Khan. 2016.
"New Methodology to Quantify Productivity of Vertical Wells in Naturally Fractured Solution
Gas Drive Reservoirs with Dual Porosity and Dual Permeability." In PAPG/SPE Pakistan
Section Annual Technical Conference and Exhibition. Society of Petroleum Engineers. https://
doi.org/10.2118/185314-MS.
42. Tariq, Z., Abdulraheem, A., Mahmoud, M., Ahmed, A., 2018a. A Rigorous Data-Driven
Approach to Predict Poisson's Ratio of Carbonate Rocks Using a Functional Network.
Petrophysics – SPWLA J. Form. Eval. Reserv. Descr. 59, 761–777. https://doi.org/10.30632/
PJV59N6-2018a2
43. Tariq, Z., Mahmoud, M., Abdulraheem, A., Al-Shehri, D., Khan, M.R., Janjua, A.N., 2018b. An
Intelligent Solution To Forecast Pressure Drop in a Vertical Well Having Multiphase Flow Using
Functional Network Technique, in: PAPG/SPE Pakistan Section Annual Technical Conference
and Exhibition. Society of Petroleum Engineers. https://doi.org/10.2118/195656-MS
44. Tariq, Z., Mahmoud, M.A., Abdulraheem, A., Al-Shehri, D.A., 2018c. On Utilizing Functional
Network to Develop Mathematical Model for Poisson's Ratio Determination, in: 52nd U.S.
Rock Mechanics/Geomechanics Symposium. American Rock Mechanics Association, Seattle,
Washington
45. Kalam, S., Alnuaim, S.A. and Rammay, M.H., 2016, March. Application of Artificial Intelligence
for Water Coning Problem in Hydraulically Fractured Tight Oil Reservoirs. In Offshore
Technology Conference Asia. Offshore Technology Conference. https://doi.org/10.4043/26450-
MS
46. Mahboob, A., Al-Nuaim, S. and Rammay, M.H., 2016, April. A Rigorous Correlation for
Quantitative Prediction of Water Cresting in Multi-Fractured Horizontal Wells. In SPE Kingdom
of Saudi Arabia Annual Technical Symposium and Exhibition. Society of Petroleum Engineers.
https://doi.org/10.2118/182778-MS
47. Rawlins, E.L. and Schellhardt, M.A. 1935. Backpressure Data on Natural Gas Wells and Their
Application to Production Practices, 7. Monograph Series, U.S. Bureau of Mines.
48. Economides M.J, Hill A.D. & Ehlig-Economides C., Petroleum Production Systems; PTR
Prentice Hall Inc. Englewood Cliffs, NJ, 1994, Chapter 10
49. Spitzer D.W.; Industrial Flow Measurement 3rd Edition, ISA, Chapter 2
14 SPE-201170-MS

APPENDIX
Table A—Values extracted from the Qgas ANN Model of the biases and weights

Weights between Input Layer and Hidden Layers (wij)

Input Layer Neurons (i)

Hidden Layer Neurons (j)
1 2 3 4 5

1 -0.47182 -3.04711 0.55972 -0.99494 -0.26728

2 3.07848 -4.12522 0.44482 -3.67250 -3.31422

3 2.15151 0.12710 0.01530 -4.34448 -5.14373

4 1.84075 -1.19843 0.22399 -2.24419 -1.14757

5 -1.47893 -0.22136 -0.02638 5.74723 5.03328

6 1.31938 0.51855 0.08150 -7.50562 -6.27374

7 0.84695 -0.27158 1.23939 10.61928 -11.34801

8 0.95065 0.63582 -1.36010 1.63761 -3.66825

9 -5.32842 -0.20016 -0.43209 -10.95542 0.75512

10 2.37113 -0.12736 0.24278 -0.94727 -1.65319

11 0.87098 0.93441 1.31514 -2.80233 -0.40243

12 -0.68254 0.27290 -0.98859 -8.39053 9.65041

13 -1.34927 -0.97848 1.46822 1.17722 4.92867

14 -4.64287 -0.02855 -0.36158 -8.23596 -0.26570

15 -14.79737 2.05551 -1.42522 11.33440 2.15983

16 1.31555 -0.26081 0.16694 -1.46558 -1.73384

17 -0.13742 1.89490 -3.32064 14.62426 8.25559

18 0.84991 0.42941 -0.86399 1.90222 2.86409

19 -0.28944 2.18831 -0.42329 -0.21409 1.56352

Bias Values for Hidden Layer Neurons (bj) Weights between Hidden Layer and Output Layers (Wjk)

Hidden Layer Neurons (j) Bias (bj) Hidden Layer Neurons (j) Output One Neuron

1 2.27018 1 -2.07453

2 -5.04933 2 -1.00187

3 -9.01761 3 -5.96680

4 -3.22574 4 7.35610

5 10.10285 5 -16.22056

6 -12.99152 6 -7.32168

7 -2.24822 7 -3.34798

8 0.29865 8 -9.30208

9 -11.71141 9 3.65225

10 -2.54458 10 6.87129

11 -1.96595 11 1.56522

12 2.44557 12 -5.04501

13 3.52526 13 -6.66635

14 -10.01104 14 -4.81992
SPE-201170-MS 15

Bias Values for Hidden Layer Neurons (bj) Weights between Hidden Layer and Output Layers (Wjk)

Hidden Layer Neurons (j) Bias (bj) Hidden Layer Neurons (j) Output One Neuron

15 -0.49549 15 -0.19419

16 -3.02967 16 -19.88461

17 23.44084 17 0.45697

18 -1.97609 18 0.89238

19 -1.85808 19 -6.04418

Bias Values for Output Layer Neuron (bk)

Output Layer Neuron Bias Value (bk)

1 -1.69765

Figure A—The ANN topology for Qgas