You are on page 1of 8

SPE-197879-MS

Utilization of Artificial Neural Networking for Real-Time Oil Production Rate


Estimation

Mohammad Al Kadem, Mohammad Al Dabbous, Ali Al Mashhad, and Hassan Al Sadah, Saudi Aramco

Copyright 2019, Society of Petroleum Engineers

This paper was prepared for presentation at the Abu Dhabi International Petroleum Exhibition & Conference held in Abu Dhabi, UAE, 11-14 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
Oil flow rate testing is a crucial concept in oil fields where several methods facilitate well rate testing and
measurement. Hundreds of multiphase flow meters (MPFMs) have been utilized to enhance the accuracy
of testing measurement and provide reliable data for all fields. Even though these meters are of paramount
importance, they require frequent preventive maintenance, calibration and manpower. In this paper, an
Artificial Neural Network (ANN) model is developed as a backup tool to replace MPFM measurements
when the device becomes defective or inoperable.
Several correlations have been established to facilitate oil well testing at minimal cost; relying on surface
production parameters to allow enumerating the oil flow rate without installing expensive equipment. An
ANN model was developed using real-time wellhead parameters measured from equipment installed at the
surface to designate properties and characteristics per reservoir. The ANN model was calibrated, tested and
validated to achieve the most accurate results. The model was further optimized to attain a reliable tool for
real-time rate estimation.
In this paper, assessment of various correlations was conducted to compare the accuracy of each ANN-
related empirical equation in five different datasets. The assessment also covers a wide range of data. More
than thousands of data points from MPFM were compared to Towailib, Marhoun and Gilbert correlations,
and showed highly deviated values with an average relative error of more than 40%. The same sets of data
were tested using the newly developed optimized ANN model. The results from the model resulted in an
average relative error of 3.7% compared with the MPFM rate measurements. Therefore, the new ANN
model presented in this paper shows highly accurate results.
The developed model contributed to enhancing testing efficiency and optimizing production. Indeed,
utilizing this model is an essential practice for production engineers to validate well tests if prompt outcomes
are desired and another reliable tool to estimate real-time rate production when metering device is down.

Introduction
Choke valve is one of the essential components utilized in surface production equipment to restrict oil,
water and gas flow rates. Therefore, formation damage, back pressure and slugging patterns can be easily
controlled5. The most mutual types of choke valves are the positive and adjustable ones. The positive valve
2 SPE-197879-MS

has fixed orifice and is dissembled to change beans4. Unlikely, the adjustable valve provides various orifice
sizes and are vulnerable to washouts from high velocity-particles5. Several attempts have been established
to come up with correlations that can estimate oil flow rates with marginal errors. These equations rely
heavily on choke measurements and surface equipment parameters3.
In the literature, there are many studies have been published on the estimation of the flow rates from
surface pressure and chokes such as Tangren et al. in 1949, Gilbert in 1954, Baxendall in 1957, Ros in
1960, Achong in 1961, Omana in 1969, Fortuati in 1972, Secen in 1976, Pilehvari in 1981, Ashford in 1975,
Shachdev in 1985, Osman 1988, Elgibaly in 1996, Mesallati in 2000, Alrumah in 2007, Ghareeb in 2007
and Al-Attar, 2009. One of the pioneer researches is the study done by Gilbert in 1954 who developed an
empirical correlation for critical flow across a short restriction5. He used 268 rate tests from the Ten Section
Field of California to develop his equation, which is applicable for tubing pressures at least 70% greater
than the line pressure to avoid the pressure fluctuations. The applicable ranges are at choke size of (7-13)
64th of inch, at flow rate of 500 bbl/day and static downhole pressure of 2500 psi and GOR of 2,000 cf/bbl5,
the general form for of his equations is given by:

Equ - 1

Other studies expanded based on Gilbert's approach including Surbey et al. in 1989, Ghassan et al in
1991, Al-Towailib et al. in 1992 and Safar et al. in 2011. Surbey et al. developed a correlation experimentally
using air & water to predict the critical flow transition through an iterative method in order to account
for choke flow geometries9. Ghassan presented another correlation to estimate the two-phase flow rates by
using upstream pressure, GOR and API oil gravity and tested it against 210 measured rate tests1. Ghassan's
correlation is given by:
Equ - 2
Al-Towailib et al. and Al-Marhoun et al. presented another correlation for calculate the two phase flow
rate through chokes, wellhead pressure and density mixture by using 3,930 production tests from 10 fields
in the middle east10. With utilizing nonlinear multiple regression for oil rate ranging between 172 – 33847
bbl/day, GOR of 12 – 5026 scf/STB, wellhead pressure between 97 to 1880 psi and choke size of 16 – 160
of 64th of inch. his correlation is given by:

Equ - 3

Safar et al. developed a correlation for multiphase flow through using 748 actual data set from an offshore
Iranian field with high flow rate and water cut3. He utilized Levenberg-Marquardt algorithm method to solve
the nonlinear function and finalize his equation form that is given by:

Equ - 4

Methods of Rate Testing Measurements


Generally, rate testing for individual wells is an important tool used by production and reservoir engineers
that helps in material balance, well monitoring and reservoir management. Two widely used testing
mechanisms in oil fields are test separators and multiphase flow meters (MPFMs). Both of those two testing
mechanisms can be installed as a permanent testing facility at the desired location or can be brought to
location as a portable testing mechanism7/8.
For most of permanent testing facilities in oil fields, number of wells are connected to a testing flow line
leading to the permanent testing facility at a location where those wells are tested one well at a time. The
SPE-197879-MS 3

frequency at which a well can be tested will depend on the number of wells connected to this testing facility,
the type of testing facility, and whether the location is onshore or offshore. Those permanent local testing
facilities are mainly dominated by MPFM and conventional separator testing. However, those permanent
testing facilities do not cover all fields or all the sites within a field. Therefore, portable testing facilities
are also widely used. However, testing frequencies for wells tested with a portable testing facility are much
lower than those with a permanent one.
Furthermore, testing wells with slug flow, high GOR or high hydrogen sulfide content using MPFMs is
very challenging. Frequent calibration and correction factors will be needed for more accurate rate tests.
Ultimately, empirical model will not require the production engineer to consider high hydrogen sulfide wells
when calulating the rate from the available surface parameters only. For those wells with harsh conditions,
portable test separators will be more accurate than MPFMs, with much lower frequency due to logistical
concerns. Though, MPFMs can supply real time flow tests unlike test separators which only provide average
rates locally. Moreover, operating and maintaining portable testing separator is much more complicated
than the MPFM where it is simpler and most of the process are automated.

Artificial Neural Network in Petroleum Engineering


The oil and gas industry is currently in a need to reduce and optimize various operational costs. Petroleum
engineers are required daily to process numerous load of data to come out with the most optimum plans.
Advance tools and technologies are needed to enable such data processing requirements and optimize the
exportation, production, and resource management costs.
Artificial Neural Networks (ANNs) are one of the latest technologies that is used by the industry. Different
kinds of problems are solved using a wide set of algorithms. Neural networks are made of two main elements
which are processing elements and interconnections. A processing element receives an input, process it and
generate an output via the hidden layers. The interconnections send this output to various other processing
elements2.
These artificial neural networks provide petroleum engineers with a great data processing and prediction
capability to automate tasks like formation lithology analysis and well test interpretation. The automation
of such daily task offers petroleum engineers with the tools required to keep up with the industry cost
optimization requirements6.

New Empirical Equation for Oil Flow Rate using ANN


Artificial Neural Network (ANN) model has been developed to help engineers to calculate flow rate across
the surface choke in oil producers. Developing this method is aimed to back-up the MPFM measurements
when the device requires calibration or becomes out of service. The model has been optimized via changing
training functions, number of hidden layers and number of neurons in the hidden layer with coding using
MATLAB™. Final optimized features of the model shown in Table-1. This ANN model is based on a
relationship between different variables. Three affecting parameters have been combined in one equation to
calculate oil flow rate with a coefficient for each individual parameter. This empirical equation is built based
on observations and a combination between the three selected parameters to fit the measurement and hence
no theoretical ground applied. The general ANN equation combines all components of Neural Network;
weights, inputs and bias as in equation-5:
Equ - 5
4 SPE-197879-MS

Table 1—ANN Model Features

The selected parameters are taken directly from a validated well test data which are flowing wellhead
pressure (Pwh), and choke size (S), and gas oil ratio (GOR) wherein are direct outputs of any conventional
rate test measured using multiphase flowmeter. This is considered as an advantage of this method versus any
other correlation where no need to calculate any additional parameters except obtaining the measurements
of the wellhead parameters along with actual. The code was run to get the best and least error between
calculated and actual oil flow rates by changing the number of neurons in the hidden layer. The ANN model
has been applied on three varied data sets, dataset#1, dataset#2 and dataset#3 with data set of 1,854 points
covering 78 oil wells completed in different reservoirs. The three datasets are selected based on diverse
conditions to cover a wide range of data and from any other field.

New Model Applicability and Limitations


The developed formula has pertinent combination of the selected parameters. Though, one limitation that
could affect its applicability wherein each dataset will have its own assigned coefficients as adding more
datasets using the same set formula. This ANN empirical equation is applied only to naturally flow oil wells
in which it is not applicable to artificially lifted wells where more parameters shall be considered. Table-2
is showing the statistical analysis of the worked dataset for the three input parameters and output parameter
(oil flow rate).

Table 2—Statistical Analysis for the three datasets

Results and Discussion


The oil flow rate is estimated using new developed model utilizing ANN for the three datasets. After
optimizing the model, the average absolute error (Equation-6) is 3.7 % indeed compared to the actual rate
from the multiphase flowmeters. This was achieved with the best optimization results of utilizing the three
parameters (Pwh, choke size (S) and GOR).

Equ - 6
SPE-197879-MS 5

As shown clearly in Table-3, the proposed new model has been compared against only Al-Towailib's and
Gilbert's since other correlations gave outrange results. The new ANN equation (Equation-7) fits the data
set with an average absolute error of 3.7% among the three datasets. However, for Towailib and Marhoun
correlation, has attained a deviated error of 89%. Gilbert's correlation is showing a moderate deviation with
39% from the actual data set but not matching them overall.

Equ - 7

b1: Bias between input and the hidden layer of the neural network.
b2: Bias between hidden and the output layer of the neural network.
w1: Weights vector between input and the hidden layer of the neural network.
w2: Weights vector between the hidden and the output layer of the neural network.

Table 3—Average Absolute Error Comparison between the new method and other correlations

Figure 1 shows the relationship between the actual rate measured from multiphase flow meters and
the estimated rate using the new ANN model. Clearly, all calculated rates in the three datasets are almost
matching with the actual measurements. This magnifies the realization of ANN applicability to stand as a
backup solution when the multiphase meter becomes down or requiring calibration. Moreover, this minor
error percentage alleviates a similar accuracy that could be gotten from the multiphase flow meters. The
same graphical comparison was applied to Figure-2 and 3 wherein the actual measurements were plotted
againsts the calculated rates from Towailib & Marhoun and Gilbert's correlations. Points on the 45 degree
separation line indicate accurate rates, above it indicate underestimated rates, and points below the line
indicate overestimated rates. As discussed above, a high deviation of more than 40% in error to the new
model was observed for both correlations giving one conclusion that proves the applicability of the new
ANN model to the working data set.
6 SPE-197879-MS

Figure 1—Graphical Comparison between Actual and Calculated Rates using the new ANN model for dataset#1

Figure 2—Graphical Comparison between Actual and Calculated Rates using the new ANN model for dataset#2

The deviation between the calculated obtained from the new ANN method and measured rates is
exhibiting sources of error that are mainly related to one of the below:

• The error could come from rate test itself in which some parameters are questionable such as venturi
differential pressure and gamma counts.
• Type of the meters used in the field utilizing different measurement mechanisms and that cause
some uncertainties.
SPE-197879-MS 7

• Moreover, if other parameters are incorporated in the new model, they may reduce the error such
are fluid and gas density, water cut and fluid hold up.

Figure 3—Graphical Comparison between Actual and Calculated Rates using the new ANN model for dataset#3

Conclusions
• ANN model has been developed covering a wide range of flow envelopes.

• It stands a back-up virtual metering tool to estimate oil flow rates when the MPFM is down,
defective or out of service. Also, it is good to be used for newly drilled wells where no MPFM
device installed yet.
• Direct implementation of data from MPFMs into the new equation without the need of calculating
any extra parameters.
• The ANN empirical equation is a customized model that works mainly for non-ESP fields.

• Additionally, the ANN model calculated rates could be used to validate the rate tests.

Nomenclature
API : crude gravity, degrees
β : ratio of cross sectional areas of upstream to that of downstream venturi meter,
dimensionless
BW&S : free water cut, percentage
Cd : discharge coefficient, dimensionless
d : choke size converted to inches, inch
D : pipe inside diameter, inch
Ea : absolute average error, dimensionless
γm : mixture specific gravity, dimensionless
gc : conversion factor (32.2 (lbm ft) / (lbfs2))
8 SPE-197879-MS

GLR : gas liquid ratio, Mscf/bbl


GOR : gas oil ratio, scf/STB
Psc : pressure at standard conditions, 14.65, psi
Pwh : flowing wellhead pressure, psi
Qactual : the actual or measured oil flow rate from multiphase flow meter, STB/D
Qestimated : the estimated or calculated oil flow rate, STB/D
QL : the liquid rate, bbl/D
Qo : oil flow rate, STB/D
ρo : oil density, pcf
S : choke size, 1/64th of inch, fraction
Tsc : temperature at standard conditions, 60, °F, R

Superscript
TM : trademark

References
1. Abdul-Majeed, G.H., 1986: "Correlations Developed To Predict Two-Phase Flow Through
Wellhead Chokes," Paper SPE 15839, Source: Society of Petroleum Engineers, 1986.
2. Ali, J.K. 1994. "Neural Networks: A New Tool for the Petroleum Industry?". Paper SPE 27561 at
the European Petroleum Computer Conference held in Aberdeen, U.K., 15-17 March 1994.
3. Beiranvand, M. S., Mohammadmoradi, P., Aminshahidy, B., Fazelabdolabadi, B. and
Aghahoseini, S., "New Multiphase Choke Correlations for a High Flow Rate Iranian Oil Field",
Published June 2012, Retrieved December 2015, from http://www.mech-sci.net/3/43/2012/
ms-3-43-2012.pdf
4. Elgibaly, A.A. and Nashawi, I.S., 1998: "New Correlations For Critical And Subcritical Two-
phase Flow Through Wellhead Chokes," Paper Journal PETSOC-98-06-04, Source: Journal of
Canadian Petroleum Technology, Volume 37, Issue 6, June 1998.
5. Gilbert, W.E., 1954: "Flowing and Gas-lift Well Performance," Paper API-54-126, Source:
Drilling and Production Practice, 1 January 1954, New York, US.
6. King, George E., "Chokes", Retrieved December 2015, from http://gekengineering.com/
Downloads/Free_Downloads/Choke.pdf
7. Mohaghegh, S., and Ameri, S. 1995. "Artificial Neural Network As A Valuable Tool For
Petroleum Engineers". Paper SPE 29220 prepared as an unsolisited paper for Society of
Petroleum Engineers.
8. Rao, S.R. and David, R.M., 2015: "Integrated Production Testing Framework to Improve Next
Generation Production Workflows," Paper SPE 177940, presented at Abu Dhabi International
Petroleum Conference and Exhibition held in Abu Dhabi, UAE, 9-12 November 2015.
9. Surbey, D.W., Kelbar, B.G. and Brill, J.P., 1989: "Study of Multiphase Critical Flow Through
Wellhead Chokes," Paper SPE 15140, Source: SPE Production Engineering Journal, Volume 4,
Issue 2, May 1989.
10. Towailib, A.I. and Marhoun, M.A., 1994: "A New Correlation For Two-phase Flow Through
Chokes," Paper Journal PETSOC-94-05-03, Source: Journal of Canadian Petroleum Technology,
Volume 33, Issue 5, May 1994.

You might also like