Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978

https://doi.org/10.1007/s13202-019-0697-7

Developing an- PRODUCTION

ORIGINAL PAPER improved ENGINEERING
approach to solving a new gas lift
optimization problem
Hamed Namdar1

Received: 15 March 2019 / Accepted: 28 May 2019 / Published online: 1 June 2019
© The Author(s) 2019

Abstract
The increased speed and accuracy in solving optimization problems of gas allocation in the gas lift process are of high
importance. Solving gas allocation optimization problems generally involves two steps: (1) The gas lift performance curve
(GLPC) fitting (gas lift modeling) and (2) optimizing the allocation of gas between wells. Therefore, in order to increase
the speed and accuracy of solving gas allocation optimization problems, both steps need to be improved. In order to
increase the accuracy of the first step, a new correlation was proposed in which, in addition to increasing the accuracy of
fit, the optimi- zation speed was improved by decreasing the number of constants used in the correlation. Besides, in order
to improve the performance of the second step, water cycle optimization algorithm was used and the results obtained from
this algorithm were compared with the results obtained from previous studies on teaching–learning-based optimization
(TLBO) algorithm, continuous ant colony (CACO) algorithm, genetic algorithm (GA) and particle swarm optimization
algorithm (PSO) for solving the five-well Nishikiori index problem. The results suggested that the water cycle
optimization algorithm has a very good performance in terms of convergence rate, non-capture at local optimum points
and repeatability. Finally, as a new problem, the gas allocation between the wells of one of the heavy oil fields in the
southwest of Iran was optimized with predetermined oil production rates. The goal of optimization was to obtain the
minimum amount of gas required to produce the predetermined oil rates using the water cycle optimization algorithm. The
results showed that optimization is of higher importance in lower oil production targets, resulting in higher additional oil
production.

Keywords Gas lift optimization · GLPC correlation · Water cycle optimization · GA · PSO

Introduction
process. In this method, the pressured gas is injected in the
When the natural energy of the reservoir is not able to bottom of the tubing, and the oil is produced through two
trans- fer the fluid to surface and to overcome the weight mechanisms of pushing through the expansion of the gas
of the fluid column in the well, one of the artificial lift and reducing the hydrostatic pressure of the fluid column
techniques should be used to produce the fluid in an inside the well. However, the excessive increase in the
economically efficient way. Artificial lift techniques amount of gas injected leads to the increased frictional
reduce the pressure from the fluid column in the well and, pressure drop and, consequently, a decrease in the
as a result, reduce the pressure at the well bottom, causing production increase due to the reduction in hydrostatic
a large pressure differ- ence between the reservoir and the pressure (Economides et al. 2013). Therefore, determining
well bottom, resulting in the transfer of the fluid produced the optimal amount of injected gas between network of
to the surface. One of the most commonly used artificial production wells is one of the main gas lift challenges and
lift techniques is the gas lift is referred to as the gas lift allocation problem (Miresmaeili
et al. 2015).

 Hamed Namdar Gas lift allocation optimization

[email protected]
1
Faculty of Petroleum and Natural Gas Engineering, In general, the gas lift allocation optimization is discussed
Sahand Oil and Gas Research Institute (SOGRI), Sahand in two scenarios: (1) optimizing the allocation of gas
University of Technology, P.O. Box 51335/1996,
between wells to maximize oil production in conditions
Sahand New City, Tabriz, Iran
where a lim- ited amount of gas is available (Hamedi et al.
2011) and (2)

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978

optimizing the allocation of gas between wells and

minimiz- ing gas injections in a situation where a certain also an optimum performance point in the gas lift perfor-
amount of oil production is pre-planned. All studies mance curve (GLPC) beyond which the gas injection
conducted so far on gas lift allocation optimization have further increases the frictional pressure gradient and
focused on the first case and explored it from different decreases the production level. Therefore, in both
perspectives (Hamedi et al. 2011; Miresmaeili et al. 2015; scenarios, the follow- ing constraints are expressed in
Mahdiani and Khamehchi addition to the constraints specific to each state (Hamedi et
al. 2011):
2015; Miresmaeili et al. 2019). However, gas lift allocation
Qgi ≥ Qgi−min (5)
optimization using the second technique is not studied yet
and addressed for the first time in the present study. The Qgi ≤ Qgi−max (6)
objective function typically used when a limited amount of
gas is available to maximize the oil production level (sce- where Qgi−max is the minimum amount of gas injection
required to start the oil production and Qgi−max is the
nario 1) is expressed as follows (Hamedi et al. 2011):
amount of gas injected in the peak of gas lift performance
, curve (optimum performance point).
n Qoi = f Qg1, Qg2, … , , i = 1, 2, … , (1)
max Qgi n
1
Literature review
in which the total amount of injected gas to the wells must
be equal to or less than the amount of available gas and is Considering the importance of gas allocation optimiza-
expressed as a linear inequality constraint (Hamedi et al. tion, many studies have been conducted to explore it.
2011): For instance, Kanu et al. (1981) used the equal slope
method to properly distribute gas rates between wells.
,n Later, Nishikiori et al. (1995) used the nonlinear con-
Qgi ≤ A (2) strained formulation with a stochastic quasi-Newton
1
method to find optimal solutions. Fang and Lo (1996)
where n represents the number of wells, Qoi is the amount departed from the nonlinear programming process and
of oil produced per well, Qgi is the amount of gas injected proposed the piecewise linearization of the well perfor-
per well and A is the amount of available gas (Hamedi et mance curves and thus changed the problem into a linear
al. 2011). In the second scenario which aims to minimize programming problem. Afterward, Buitrago et al. (1996)
the amount of gas injected into wells to produce a predeter- combined stochastic search and heuristic descent direc-
mined quantity of oil, the objective function is expressed tion and called their method Ex-In. Alarcon et al. (2002)
as follows: improved the method proposed by Nishikiori et al. (1995)
by replacing the quasi-Newton algorithm with sequential
,n quadratic programming (SQP). Wang et al. (2002) pro-
min Qgi, i = 1, 2, … , (3) posed a mixed-integer nonlinear programming (MINLP)
1 n
techniques for integrating the previous methods. Their
in which the total oil produced from the well is equal to a model included the allocation of gas rates, the produc-
predetermined amount and expressed as follows: tion rate of wells and equipment constraints. Nakashima
and Camponogara (2006) developed the recursive algo-
,n rithm for gas rate allocation. They were the first to study
Qoi = f Qg1, Qg2, … , Qgi = (4) the discontinuity of the well performance curve in detail.
1 B Ray and Sarker (2006) used the piecewise linearization
where B is the predetermined oil quantity for production. In addition, an initial gas injection is required in some
The important point is that in the second scenario the prob- wells in order to start the production of oil, and there is
lem constraint turns into a nonlinear equality constraint
com- pared to the first case with a linear inequality
constraint, since the production of each well is itself a
nonlinear func- tion of the amount of gas injection,
resulting in a nonlinear function of the amount of gas
injected. Therefore, in the second scenario, handling the
problem constraints will be more complicated.

1
Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
method and genetic algorithm (GA) to optimize the gas
allocation rate. Camponogara and Conto (2009) improved
the piecewise linearization formulation that was previ-
ously developed. Zerafat et al. (2009) solved the gas rate
distribution using two genetic and ant colony (ACO)
algorithms. Hamedi et al. (2011) used the particle swarm
optimization algorithm (PSO) to optimize the gas rate
allocation in the wells of an Iranian oil field. Sharma and
Glemmestad (2013) used the generalized reduced
gradi- ent (GRG) technique, self-optimizing control
structure and multi-start technique to optimize gas
allocation in the gas lift process. Ghaedi et al. (2014)
employed the

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
continuous ant colony (CACO) algorithm to explore gas
allocation optimization and compared their results with
previous studies. Ghassemzadeh and Pourafshary (2015)
proposed a new method for considering the time factor
in optimizing the gas allocation process. They used a
piecewise cubic hermite function for modeling the gas
lift performance and genetic algorithm for optimization.
Miresmaeili et al. (2015) used a multi-objective optimiza-
tion algorithm, Gaussian Bayesian networks and Gauss-
ian kernels to solve the gas allocation problem in the
gas lift process. Mahdiani and Khamehchi (2015) used
the genetic algorithm to investigate instability in the gas
allocation optimization problem. Tavakoli et al. (2017)
used the artificial neural networks (ANNs) to the gas lift
modeling and then studied the gas allocation optimiza-
tion using the genetic algorithm. Also, Miresmaeili et al.
(2019) employed the artificial neural networks and used
Levenberg–Marquardt (LM) and Bayesian regulariza-
tion (BR) algorithms to model gas lift operation and then
used the teaching–learning-based optimization (TLBO)
algorithm to optimize gas allocation. Moreover, Namdar
and Shahmohammadi (2019), with the help of a simple Fig. 1 Schematic diagram of water cycle in the nature (Sadollah et al.
2015)
method without programming, optimized the gas alloca-
tion by the excel solver optimization tool.
In this study, in order to increase the speed and accu- Water cycle algorithm
racy of solving gas lift allocation optimization problems,
at the first a new correlation was proposed for modeling This new meta-heuristic algorithm has been inspired by the
the gas lift process and GLPC curve fitting. The modeling behavior of the water cycle in nature (Fig. 1) and has not
results were compared with those obtained through other yet been used in the field of petroleum engineering. Similar
modeling techniques. In order to improve the optimiza- to other meta-heuristic algorithms, the water cycle
tion performance, a water cycle optimization algorithm algorithm begins with the initial population of so-called
(WCA) was used and the results of its implementation streams cre- ated after rain. To simulate this process,
were compared with the results obtained from previous initially, a primitive population Npop is created randomly,
studies on TLBO, CACO, GA and PSO algorithms for each stream having Nvar design variables (Sadollah et al.
solving the five-well Nishikiori (1989) index problem. 2015).
Also, given the effect of the gas lift process modeling on Then, the value of the objective function of each stream
the optimization results, and since previous studies have is calculated and the best solution is selected as the sea.
used various modeling techniques, we compared the per- Afterward, some of the streams that are best after the sea
formance of the water cycle optimization algorithm with are chosen as rivers, and the remaining streams are
the two well-known PSO and GA algorithms in terms of remained as streams and flow to the rivers or directly to
convergence rate, non-capture at local optimum points and the sea. NSR is a parameter whose value is determined by
repeatability with the same correlation for modeling gas the user and the number of rivers, and a sea is
lift and with equal number of iterations in order to evalu- determined through Eq. 7. As a result, the number of
ate the performance of the algorithm itself individually streams (NStreams), which is the remainder of the population,
is calculated from Eq. 8 (Sadol- lah et al. 2015):

and eliminate the modeling accuracy effects. Finally, given zation algorithm. =
that the optimization problem of scenario 2 has not been
investigated so far, this study attempted to optimize the gas
+
allocation between the wells of one of the heavy oil fields
in the southwest of Iran with predetermined quantities of
oil production. In this scenario, the optimization goal was ⌃
to obtain the minimum amount of gas required to produce 1
the predetermined oil levels using the water cycle optimi-

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978

(7)

NStream = Npop − NSR

(8)

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Thereafter, the number of streams flowing into specific
rivers or the sea is determined. The allocation of streams sea is exchanged (that is, the sea changes into the river and
is based on the intensity of the flow of rivers and the sea the river into the sea). This exchange can also occur for the
(or the values of their objective functions) through Eqs. 9 streams and the sea as well as the streams and the rivers
and 10 (Sadollah et al. 2015): (Sadollah et al. 2015).
In order to prevent the rapid convergence of the algo-
Cn = Costn − CostN rithms (immature convergence) and increase the explora-
+1, n = 1, 2, 3, … , (9)
NSR tion ability of the algorithms, a new chance commensurate
j
S

with the distance of the rivers and streams from the sea
j
NS = round j Cn j , n = 1, 2, 3, … ,
N is given to them. This concept in the WCA algorithm is
j j× N
n Stream SR
j j ∑ SR
j j applied under evaporation condition and raining process.
Cost i
j i=1 j (10) If the Euclidean distance of the rivers and streams from the
where NSn is the number of streams that flow into specific sea is less than a predetermined insignificant value (near
rivers or the sea. After that, the surface movement of zero) ( dmax), the conditions for evaporation from the sea
streams and rivers to the sea or the movement of streams to are fulfilled and the raining process starts, and the rivers
the riv- ers and their new positions are determined. The and streams are reformed (Sadollah et al. 2015).
movement of the streams directly flowing into the sea is The conditions for evaporation from the sea for rivers
calculated by Eq. 11, the movement of the streams joining and streams are expressed by pseudo-codes 14 and 15,
the rivers is determined through Eq. 12 and the movement respec- tively (Sadollah et al. 2015):
of the rivers to the sea is estimated through Eq. 13
(Sadollah et al. 2015):

(14)

(15)

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978

where dmax sets the intensity of the search around the sea
X⃗Stream
i t X⃗ i (t) + rand × C × X⃗ i t X⃗ i (t) ,
( +1 Stream Sea ( Stream and decreases by Eq. 16 in each step (Sadollah et al. 2015):
i = 1, 2, … , dmax(t)
(t + 1) = d max (t) − ,
NStream
(11 dmax max Iteration
(16)
)
t X⃗ i (t) + rand × C t X⃗ i t = 1, 2, 3, … , Max Iteration
X⃗Stream
i
( +1 Stream X⃗River
i
( Stream
(t) ,
If a condition for evaporation is fulfilled, new streams are
×
i = 1, 2, … , formed randomly in the search space due to raining, and
NStream (12) the new location of these streams is determined by the
following equation (Sadollah et al. 2015):
⃗ i (t + 1) = X⃗ (t) + rand × C × X (t) i− (t) ,
X
t ⃗ i X⃗
River River Sea River X
i = 1, 2, … , ⃗New = L���B�⃗ + U����B�⃗ −
– 1) (17)
(NSR Stream
rand × L���B�⃗

where t is the iteration number, and C (13)

where UB and LB are upper and lower bound vectors that
is a constant whose
are defined by the problem. Figure 2 shows the developed
value varies from 1 to 2 and its best value equals two. Also,
rand is a uniformly distributed random number ranging WCA optimization process. The circles, stars and the dia-
from zero to one. These equations act as exploration at the mond refer to streams, rivers and sea, respectively, and the
initial stages and as exploitations at the final stages white circles and stars mark the new positions of streams
(Sadollah et al. 2015). and rivers, respectively.
If the solution presented by a river is more effective
than that of the sea, then the position of the river and the

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
problems generally involve two steps:

Fig. 2 Schematic view of the used methods in the WCA (Sadollah

et al. 2015)

The application of constraints

Usually, in gas lift allocation optimization problems, a

penalty function is used to address the constraints of the
problem and prevent violation of them (Sukarno et al.
2009; Rashid 2010; Hamedi et al. 2011). The main
problem with this method is that the coefficient of the
penalty function itself needs to be optimized and if it is not
selected prop- erly, the slightest violation of the
constraints may result in a large penalty for the objective
function. In order to apply the constraints and to reach
feasible solutions, Deb (2000) approach was used in this
study. This approach is based on the penalty function, with
the difference that there is no need for the coefficient of the
penalty function and its optimiza- tion. This method is used
for the genetic algorithm by Deb (2000) and is used in this
study for use in the water cycle algorithm. The rules used
in this method (for the minimiza- tion problem) are as
follows:

1. Of two feasible solutions, the solution whose objective

function has a lower value is selected.
2. Of a feasible solution and a non-feasible solution, the
feasible solution is preferred.
3. Of the two non-feasible solutions, a solution that has
the slightest violation of the constraints is selected as
the feasible solution.

Methodology

The increased speed and accuracy in solving optimization

problems of gas allocation in the gas lift process are of
high importance. Solving gas allocation optimization

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
1. Modeling the gas lift process and fitting the oil methods have been used to model the gas lift process and
produc- tion data against gas injections by one of the the fitting of GLPCs. In general, these
correlations presented.
2. Implementation of the objective function of the gas
allocation optimization problem and its constraints
and solving it by one of the optimization algorithms.

Therefore, in order to increase the speed and accuracy

of solving gas allocation optimization problems, each of
these steps needs to be improved individually. In order to
increase the accuracy of the first stage, first, a new
correlation was proposed to model the gas lift process,
and then the results of modeling with it were compared
with the results obtained from other methods for different
wells. In order to improve the performance of the second
stage of the optimization pro- cess, an optimization
algorithm was used and the results of this algorithm were
compared with the results obtained from previous studies
on TLBO, CACO, GA and PSO algorithms for solving
the five-well Nishikiori (1989) index problem. Also,
given the effect of the gas lift process modeling on the
optimization results, we compared the performance of the
water cycle optimization algorithm with the two well-
known PSO and GA algorithms in terms of convergence
rate, non- capture at local optimum points and
repeatability with the same correlation for modeling gas
injections and with equal number of iterations in order to
evaluate the performance of the algorithm itself
individually and eliminate the modeling accuracy effects.
Finally, this study attempted to optimize the gas allocation
between the wells of one of the heavy oil fields in the
southwest of Iran with predetermined quantities of oil
production. In this scenario, the optimization goal was to
obtain the minimum amount of gas required to produce
the predetermined oil levels using the water cycle
optimiza- tion algorithm. In other words, the aim was to
divide the oil production among the wells in a way that it
requires the min- imum gas rate for injection into the
wells. Figure 3 shows the proposed procedure used in this
study to solve the gas lift optimization problems.

Results and discussion

Gas lift modeling process

The formation of the gas lift performance curve (GLPC)

is the first step in modeling the gas lift process. Obtaining
an accurate GLPC has a significant effect on optimizing
the allocation of injected gas into wells because if the
curves are not sufficiently accurate, despite the high
performance of the optimization algorithm, the
optimization results will not be accurate and the produced
oil overestimated or underes- timated. So far, many

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
In addition to the previously proposed models, the
author developed the following model fitting the
operational data of oil production against gas injection by
adjusting the numbers in the previously proposed
correlations and removing some of the less important
terms:
Q = a + b × Q + c × Q0.7
0.6
o g g + d × Ln Qg + + e × exp −Qg
0.9 (21)
where a, b, c, d, e and f are constant coefficients of correla-
tions, Qg is the injected gas flow rate and Qo is the produced
oil flow rate. Among the previous correlations proposed for
modeling the gas lift process, the correlation proposed by
Behjoomanesh et al. (2015), despite the high accuracy in
fitting, has some basic problems, which reduces its
general- izability to be used for all wells. Therefore, it has
been tried to solve these problems in the model proposed
by the author, while maintaining the accuracy of the model
proposed by Behjoomanesh et al. (2015). The fitting of
operational oil production data against gas injection
becomes convergent for some wells in the model proposed
by Behjoomanesh et al. (2015), and it is necessary to limit
the range of variations of the coefficients of this model. For
example, operational data on oil production versus gas
Fig. 3 Graphical demonstration of the procedure used in this study to injections presented in studies conducted by Nishikiori
solve the gas lift optimization problems (1989) and Jung and Lim (2016) cannot be fitted by this
equation for the mentioned reasons. Also, the operational
data fitting using the model proposed by Behjoomanesh et
methods can be divided into two groups: modeling al. (2015) would result in overestimat- ing or
methods by fitting through correlation and modeling underestimating the oil production in some wells. Fig- ure
methods based on artificial neural networks. To date, 4 shows a comparison of the fitting of the operational data
various correlations have been proposed for GLPC fitting. for six wells studied by Kanu et al. (1981) using the model
The quadratic polyno- mial model is one of these models developed by the author of the present study and the model
(Nishikiori 1989). proposed by Behjoomanesh et al. (2015).
However, since most of the GLPCs are not symmetrical, As it can be seen, the GLPC fitting curve trend in the
the model does not have a high efficiency for modeling. In model proposed by Behjoomanesh et al. (2015) is in a way
order to increase the accuracy of the quadratic polynomial that leads to the overestimation or underestimation of oil
model, Alarcon et al. (2002) added a logarithmic term to it production in some parts of the fitting curve. Generally, the
and proposed the following model: GLPC curve should follow a downward trend on the right
side, while, as shown in Fig. 4, in wells 2, 3 and 5, the
model
Qo = a + b × Qg + c × Q2 + d × Ln(Qg + (18) proposed by Behjoomanesh et al. (2015) moves upward in
g
1)

By replacing the quadratic term of the quadratic poly- the right. Also, however, the downward trend of the curve
nomial model with a squared term, Hamedi et al. (2011) for well 6 is suddenly bulged and these behaviors generally
proposed the following model: invalidate the oil production values estimated by this
model. In order to assess the accuracy of the proposed
models and to select the best one for modeling the gas lift
process,
Qo = a + b × Qg + c × Qg (19) the fitting of operational data for oil production versus gas

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
With the linear combination of the previously proposed injection of wells in the previous studies was calculated
models, Behjoomanesh et al. (2015) developed a model as and the results were compared with the results of the model
follows: pro- posed in this study. In order to evaluate the accuracy
of the models, the indices of correlation factor (R2) and
the root-
Q = a + b × Q + c × Q2 + d × Ln Q + mean-square error (RMSE) were used for fitting the opera-
1o g g
g
, (20 tional data. The correlation factor (R2) shows the correlation
)
+e× Qg + f × exp Qg between the model and the data and the closer values to one

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Fig. 4 Comparison of opera-
tional data fitting for wells
studied by Kanu et al. (1981)
using the author’s model and
the Behjoomanesh et al.’s
(2015) model

are more optimal. This coefficient can be defined as

As before, n is the number of observation data, yi is the
follows (MATLAB Manual 2018): ith observed value, ȳ i is the mean of all observed data
SSE and wi is the weight used for each data point, which is
usually 1.
R2 = 1 − (22) Root-mean-square error (RMSE) represents the average
TSS
difference between the values predicted by the model and the
where SSE is the sum of squared errors and is estimated as actual values, and the closer the value to zero, the model fit
follows: is more accurate. This index is defined as follows
(MATLAB Manual 2018):
,n
SSE = wi yi − ŷ i 2 (23) RMSE = MSE (25)
i=1
MSE is the mean squared error and is calculated as
where n is the number of observation data, yi is the ith follows:
observed value, ŷ i is the estimated value of yi from fit 1
and
wi is the weight used for each data point, which is usually 1. MSE = n SSE (26)
TSS is the total sum of squares and is calculated as follows:
Table 1 shows the accuracy of the proposed model by

TSS = i=1
,n 2 wi yi − ȳ ii

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
(24) a ) and
u Alarcon
t et al.
h (2002) in
o terms of
r fitting
the
a operation
n al data in
d the
studies
t conducte
h d by
e Nishikior
i (1989)
m and Jung
o
d
e
l
s

p
r
e
s
e
n
t
e
d

b
y

H
a
m
e
d
i

e
t

a
l
.

(
2
0
1
1

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Table 1 Results of author’s
model, Alarcon et al.’s References Well no. Alarcon et al.’s (2002) Hamedi et al.’s Author’s model
(2002) model and Hamedi et model (2011) model
al.’s
(2011) model for fitting R2 RMSE R2 RMSE R2 RMSE
operational data from Nishikiori
(1989) and Jung and Lim (2016) Nishikiori (1989) 1 0.987848 12.15155 0.997918 4.591787 0.999817 1.924778
2 0.991633 27.43855 0.995863 17.86276 0.999992 1.013721
3 0.988812 48.66602 0.997012 23.28292 0.99998 2.509012
4 0.996597 7.947265 0.999726 2.522818 0.999965 1.27366
5 0.996583 16.63779 0.980617 37.06711 0.999939 2.634642
Ave. 0.992295 22.56823 0.994227 17.06548 0.999939 1.871163
Jung and Lim (2016) 1 0.998931 20.69668 0.978634 86.55026 0.999973 3.563994
2 0.999433 14.29458 0.975063 88.70493 0.999993 1.701711
3 0.995613 56.26503 0.987708 88.10038 0.999979 4.198918
4 0.999499 12.88447 0.974472 86.01951 0.999996 1.241376
Ave. 0.998369 26.03519 0.978969 87.34377 0.999985 2.6765

Table 2 Results of author’s

Reference Well no. Alarcon et al.’s (2002) Hamedi et al.’s Author’s model
model, Behjoomanesh et al.’s
model (2011) model
(2015) model, Hamedi et
al.’s
(2011) model and Alarcon R2 RMSE R2 RMSE R2 RMSE
et al.’s (2002) model for
fitting operational data from Vieira (2015) 1 0.9999 4.279252 0.988402 42.10739 0.9999173 5.028378
Vieira (2015) 2 0.999659 15.47992 0.998197 31.83748 0.9999833 4.847634
3 0.999802 6.364839 0.987168 47.4466 0.9999904 1.720824
4 0.974538 21.61731 0.981893 22.95411 0.9999996 0.127734
5 0.975266 18.99519 0.96637 19.8109 0.9999997 0.091736
Ave. 0.989833 13.3473 0.984406 32.8313 0.9999780 2.363261

Table 3 Results of author’s

model, ANN method, Hamedi Error index Well no. Alarcon et al.’s Hamedi et ANN method Author’s model
et al.’s (2011) model and model (2002) al.’s model
Alarcon et al.’s (2002) model (2011)
for fitting operational data R2 5 0.97450 0.98630 0.99445 0.999492
from Tavakoli et al. (2017)
4 0.99350 0.97990 0.99980 0.999745
3 0.99720 0.98350 0.99920 0.999687
2 0.99900 0.97210 0.99950 0.999078
1 0.99510 0.98290 0.99980 0.994151
Ave. 0.99186 0.98094 0.99855 0.998431

and Lim (2016). As shown in Table 1, the proposed model Behjoomanesh et al. (2015), it has higher R2 values and
by author yields better results than the other two models. lower RMSE than the other three models and thus
Table 2 also shows the results of the fitting of the opera- provides more accurate
tional data in the studies of Vieira (2015) by the proposed
model by author and the models presented by
Behjoomanesh et al. (2015), Hamedi et al. (2011) and
Alarcon et al. (2002). As shown in Table 2, the proposed
model by author has higher R2 values and lower RMSE,
thus having more accu- rate fitting results than the other
three models.
As shown in Table 2, although the proposed model has
one constant less than the model proposed by

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
results in terms of the operational data fitting. The lower
number of constants while maintaining the fitting
accuracy makes it possible to reduce the time of
computation by main- taining accuracy in cases where it is
necessary to optimize the gas allocation between a large
numbers of wells.
In addition, Tavakoli et al. (2017) fitted the operational
data with artificial neural networks and compared their
results with models presented by Hamedi et al. (2011) and
Alarcon et al. (2002) in terms of R2 values. Table 3 shows
the results of fitting the data in the study conducted by
Tava- koli et al. (2017) using the proposed model, in
comparison with the artificial neural network and the
models proposed by Hamedi et al. (2011) and Alarcon et
al. (2002).

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
As it is shown, the accuracy of the model proposed in
the present study is within the limits of the artificial neu-

1085.
330.3

721.5

579.7
ral network approach. Accordingly, it can be suggested that

35
Qo

8
the proposed model possesses a high accuracy in terms of

Qo (STB/D)
the operational data fitting. In the following section, the
proposed model is used for the gas lift performance curve

241.

550.

874.

574.
(GLPC) fitting.

74

35

74
W
Performance evaluation of the water cycle

Qo (STB/D)
algorithm

1085.
316.5

734.1

578.2
Available gas = 3000

57

24
9
In order to assess and validate the performance of the water
cycle algorithm, it was attempted to optimize the five-well
Nishikiori (1989) index problem using the water cycle

MSCF/D
algorithm and the obtained results were compared with the

Table 4 Comparison of the optimization results for the five-well Nishikiori (1989) index problem by using the TLBO, CACO, GA and PSO

158.

590.

913.

575.
results obtained from previous studies. The five-well

75

45

48
Qg
Nishi- kiori (1989) index problem is a gas allocation

Qo (STB/D)
optimiza- tion problem of scenario 1 and has been

3.64E+

757.68
1142.2

632.02
optimized for the available gas quantities of 4600
MMSCF/D by the TLBO (Miresmaeili et al. 2019), CACO

02

1
(Ghaedi et al. 2014) and GA (Ghassemzadeh and
Pourafshary 2015) algorithms and 3000 MMSCF/D by
PSO algorithm (Hamedi et al. 2011). Table 4 illustrates the

1211.
444.1

764.5

887.3
optimization results for the five-well Nishikiori (1989)

73
W
Qg
index problem by using the water cycle optimization

7
algorithm and compares them with the TLBO, CACO, GA

Qo (STB/D)
and PSO algorithms. As shown in Table 4, only the oil
rates generated by the TLBO algorithm are more than the

367.

758.

633.
114
1.1
values obtained by the WCA algorithm, but a thor- ough

9
examination of the data obtained for the gas allocation
Qg (MSCF/D)

among the wells by the TLBO algorithm shows that the oil
production rate by this algorithm has been overestimated.
394.

672.

918.
An investigation of initial operational data for oil
116
7.5
TL

3
production versus gas injection for well no. 1 in Nishikiori
Qo (STB/D)

(1989) stud- ies indicates that only 490.2 MSCF/D gas is

required to pro- duce 367.5 STB/D oil. However, the
1132.
366.7

758.0

631.3
amount of gas required to produce 367.4 STB/D oil is
38
2

estimated by the TLBO algo- rithm to be 394.3 MSCF/D.

In other words, the amount of oil producible by this
amount of gas is overestimated.
Also investigation of initial operational data for oil pro-
duction versus gas injection for well no. 5 in Nishikiori
(1989) studies indicates that by injecting 1667.3 MSCF/D
1132.
482.2

768.5

898.7
CA

41

gas into the well no. 5, the oil production rate will be 813.6
Qg

STB/D. Therefore, in order to produce 839.3 STB/D oil,

Qo (STB/D)
WellAvailable gas = 4600 MSCF/D

over 1667.3 MSCF/D gas needs to be injected into the

365.93

759.46

1131.0

633.47

well. However, the required amount of the injected gas to

produce the same amount of oil as estimated by this
01

12

02

algorithm is less than 1667.3 MSCF/D (1447.7 MSCF/D).

Therefore, it can be said that the oil production rate is
(MSCF/D)

overestimated by the TLBO algorithm, while the results

obtained by the WCA algorithm are more reasonable than
GA

1473.5147

2785.8542
31134.233
4918.9989
51287.399
Qg

the initial operational data of the Nishikiori’s (1989) study.

Also, given the effect of the gas lift process modeling

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
on the optimization results, and since previous studies have

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
3700
500
450 GA WCA
3650
400 PSO
GA WCA
PSO 350
Optimum Oil Production

3600

CPU Run Time

300
3550 250
200
3500
150
100
3450
50
0
3400
0 200 400 600 800 1000 1200
0
200 400 600 800 1000 1200 Population
Population

Fig. 6 Optimization CPU run time at different populations for WCA,

Fig. 5 Optimized oil production rate in different populations as esti- PSO and GA algorithms
mated by WCA, PSO and GA algorithms

used various modeling techniques for solving the five-well than the GA algorithm in terms of speed because the PSO
Nishikiori (1989) index problem, the performance of the algorithm converges to the optimal point at 24.81 s with a
water cycle optimization algorithm was compared with
the two well-known PSO and GA algorithms in terms of
convergence rate, non-capture at local optimum points and
repeatability with the same correlation for modeling gas
lift process and with equal number of iterations in order to
evaluate the performance of the algorithm itself individu-
ally and eliminate the modeling accuracy effects. For this
purpose, in all three algorithms, the correlation proposed
by the author was used in the present study for gas lift
fitting and modeling. Also, following previous studies for
solving Nishikiori (1989) index problem (Miresmaeili et al.
2019, Ghaedi et al. 2014, Hamedi et al. 2011), the number
of itera- tions was considered equal to 100 in all three
algorithms.
The WCA, PSO and GA algorithms are all three popula-
tion-based algorithms, in which the population in these
three algorithms is expressed in terms of the number of
streams, chromosomes and particles. Figure 5 shows
optimized oil production rates in different populations
using the three WCA, PSO and GA algorithms.
As it is shown Fig. 5, the exploration and exploitation
phase in the search space in the genetic algorithm is highly
dependent on the population and in the lower population,
the algorithm is trapped in the local optimum points, while
the PSO and WCA algorithms have a more powerful,
explora- tion and extraction phase in the search space and,
therefore, are not so dependent on the population and are
not trapped in the local optimum points. In addition, the
PSO and WCA algorithms have high repeatability in the
obtained results.
Figure 6 shows the optimization run time by the three
algorithms studied in different populations. As it is shown,
the PSO optimization algorithm has a better performance

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
population of 50, while the GA algorithm converges to the
optimal point at 85.85 s with a population of 650. In con-
trast, the WCA algorithm has a much better performance
in terms of speed than the PSO algorithm, so that it
converges to the optimal point at lower populations and at
higher speeds. However, as shown in Fig. 6, the PSO
algorithm has a lower performance at high populations in
terms of speed. The WCA algorithm converges to an
optimal point in 0.56 s with a population of 50. Therefore,
it can be said that the WCA algorithm has a better
performance in the opti- mization problems than GA and
PSO algorithms in terms of convergence speed, non-
capture in local optimum points and repeatability.

Case study (Scenario 2)

All studies on gas lift allocation optimization have

assumed that if a limited amount of gas is available, then
the gas should be distributed between the wells in a
manner that it would be possible to produce the maximum
amount of oil. However, an issue of interest not
investigated in the previous studies is if the goal is to
produce a predetermined amount of oil from the reservoir,
then how this predetermined amount of oil should be
distributed between the wells to require the minimum
amount of gas injection in the gas lift operation? This
section addresses this problem in one of the heavy oil
fields of southwest Iran using the water cycle algorithm.
The field in question contains four wells, of which the
well no. 2 is closed because of the high water cut. Table 5
shows the specifications of the three productive wells in
the field. The normal production of the field is 3728
STB/D without any gas lift operation. Table 6 also shows
the data of oil produc- tion against gas injection in the
wells in gas lift operation, which were obtained by using
wells modeling in Prosper software.

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Table 5 Specifications of the three productive wells in the field
Table 8 Constants of the proposed model obtained from the GLPC
Property Well 1 Well 3 Well 4 fitting
Well TVD (ft) 9730.7 9898.29 10080.8
Well MD (ft) 9730.97 12513.1 11302.5

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
Well no. a b c d e

1 5976.40 32.19 – 477.73 1129.00 –

4397.74
Reservoir pressure (psia) 4640 4640 4640 3 5138.54 – 375.23 1071.04 – 186.29 – 4289.66
Bottom hole temperature (F) 204 204 204 4 5669.25 – 280.89 661.66 367.49 – 4230.72
Well head temperature (F) 77 77 77
Well head flowing pressure (Psia) 150 150 150
Formation gas–liquid ratio (scf/STB) 137 137 137 Table 8 shows the values of the constants obtained from
API oil gravity (API) 13.11 13.11 13.11 the GLPC fitting by the proposed model. The GLPCs for
dif-
Water cut (%) 40 40 40 ferent wells suggest that the injection of 8.718, 10.017 and
PI (STB/day/psi) 9.3 8.5 10.1 10.13 MMSCF/D in the wells 1, 3 and 4 results in the pro-
Tubing O.D. (in.) 4 1/2 4 1/2 4 1/2 duction of maximum oil production, i.e., 1937.257 STB/D.
Casing O.D. (in.) 5 5 5 Four oil production targets (12500, 15000, 17500 and
Specific gravity of gas 0.91 0.91 0.91 19737.257 STB/D) were set for the oil field under analysis.

Following these targets, the gas allocation between the wells

was optimized in such a way that the minimum amount of
To solve the optimization problems, it is necessary to gas is required for the production of the targeted oil
have a correlation to calculate the amount of oil produced produc- tion rates. Furthermore, to show the effect of
for different amounts of injected gas. Therefore, to find the optimization, the worst possible gas injection scenario with
best correlation, we calculated the fitting of the data the maximum amount of gas used to produce the
presented in Table 6 by using different correlations. Table predetermined oil produc- tion was calculated.
7 shows the correlation factor (R2) and root-mean-square Table 9 shows the results for the gas allocation opti-
error (RMSE) for the GLPC fitting calculated through mization for different wells of the oil field at different oil
different correla- tions. As it was expected, the model flow rates. The Qo/Qg ratio in the table indicates the num-
proposed in this study produces better fitting results than ber of standard barrels of oil produced per injection of
other models presented in the literature. Therefore, the every one million cubic feet of the standard gas. As it was
proposed model was used to investigate gas allocation expected, in the optimal scenario (the minimum required
optimization.

Table 6 Data of oil production Well 1 Well 2 W ell 3

versus gas injection in wells
of the field Qg (MMSCF/D) Qo (STB/D) Qg (MMSCF/D) Qo (STB/D) Qo (STB/D)
Qg (MMSCF/D)
0.00E+00 1.46E+03 0.00E+00 8.68E+02 0.00E+00 1.40E+03
6.99E−01 4.19E+03 5.84E−01 3.51E+03 7.16E−01 4.30E+03
1.11E+00 4.77E+03 9.46E−01 4.06E+03 1.15E+00 4.91E+03
1.75E+00 5.35E+03 1.52E+00 4.65E+03 1.80E+00 5.52E+03
2.66E+00 5.82E+03 2.37E+00 5.19E+03 2.76E+00 6.04E+03
3.96E+00 6.18E+03 3.61E+00 5.64E+03 4.15E+00 6.48E+03
5.75E+00 6.42E+03 5.36E+00 5.98E+03 6.10E+00 6.80E+03
8.19E+00 6.53E+03 7.75E+00 6.18E+03 8.73E+00 6.96E+03
1.14E+01 6.49E+03 1.09E+01 6.22E+03 1.22E+01 6.95E+03

Table 7 Error indices values of Well no. Error index Alarcon et al.’s Hamedi et al.’s Behjoomanesh Author’s model
GLPC fitting through different (2011) model
(2002) model et al.’s (2015)
correlations
model
1 R2 0.989154 0.996935 0.999981 0.999987
RMSE 215.7698 104.7083 11.68057 8.254048
3 R2 0.979262 0.993753 0.999982 0.999993
RMSE 297.11 150.9735 10.64476 6.668139
4 R2 0.98866 0.997422 0.999976 0.999990
RMSE 241,768 105.23 14.23711 7.994826

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Table 9 Gas allocation optimization for wells of the field at different oil flow rates

Required gas Min. Max.

Determined oil rate (STB/D) Well Gas injected Oil produced Gas injected Oil
(MMSCF/D) (STB/D) (MMSCF/D) produced
(STB/D)

12500 1 0.772406 4314.469 1.20246 4874.365

3 0.724736 3746.082 10.017 6222.635
4 0.800147 4439.449 0 1403
Total 2.297289 12500 11.21946 12500
Qo/Qg (STB/MMSCF) 5441.195 1114.136
15000 1 1.455337 5117.931 8.718 6534.253
3 1.434568 4575.322 0.025065 1485.387
4 1.539948 5306.747 10.136 6980.359
Total 4.429853 15000 18.87906 15000
Qo/Qg (STB/MMSCF) 2821.764 794.5309
17500 1 2.790915 5870.636 0.761482 4297.009
3 2.985055 5445.478 10.017 6222.635
4 3.105087 6183.886 10.13 6980.356
Total 8.881057 17500 20.90848 17500
Qo/Qg (STB/MMSCF) 1970.486 836.9809
Max = 19,737.25 1 8.718 6534.253 8.718 6534.253
3 10.017 6222.635 10.017 6222.635
4 10.13 6980.359 10.13 6980.359
Total 28.865 19737.25 28.865 19737.25
Qo/Qg (STB/MMSCF) 683.7779 683.7779

5000 through the optimization can be reduced. Therefore, it can

4500 be suggested that in lower oil production targets, the opti-
per 1 MMSCF Injected
Additional STB Oil produced

4000
mization is more important and generates more extra oil.
3500
In addition to the gas lift, the method used in this paper
3000
2500
can be used in improving steam allocation management in
2000
thermal enhanced oil recovery methods such as SAGD.
1500 One of the challenges in these methods is excessive water
1000 produc- tion which is due to an improper steam injection
500 plan. Thus, for a given volume of steam, steam allocation
0
12500 15000 17500 19737.25
between wells should be managed in a manner to delay
Determined Oil Rate (STB/D) the water break- through in producer wells and, as a
result, improves sweep
efficiency and increases the oil recovery.
Fig. 7 Surplus oil rate produced in the optimal scenario (the mini-
mum required gas) compared to the worst possible scenario (the max-
imum required gas)
Conclusion
gas), a higher number of standard barrels of oil can be 1. In the gas lift process modeling, the correlation
produced per injection of every one million cubic feet of proposed by Behjoomanesh et al. (2015), despite the
gas than the worst scenario (the maximum required gas). high preci- sion, cannot be applied to all wells because
Figure 7 also shows the surplus oil rate produced in the firstly, in some wells, convergence requires limiting the
optimal scenario (the minimum required gas) compared to variations of constant coefficients, and secondly, the oil
the worst possible scenario (the maximum required gas). production rates in some oil wells are overestimated or
As it can be seen, the closer the targeted oil production underesti- mated
rates to the maximum oil production rate in the gas lift 2. The model proposed in this study, despite the reduc-
process (1937.257 STB/D), the surplus oil producible

1
2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978
tion of a constant compared to the model presented
by

1
 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978 2
Behjoomanesh et al. (2015), has a higher fitting accu-
racy and is also free from the limitations of the men- Economides MJ, Hill AD, Economides CE, Zhu D (2013) Petroleum
production systems. Prentice Hall Publication, Englewood Cliffs
tioned model. Reducing a constant coefficient in this Fang W, Lo K (1996) A generalized well-management scheme for
model will increase the speed of optimization. Also, reservoir simulation, Society of Petroleum Engineers. https://
the correlation proposed in this study is more accurate doi.org/10.2118/29124-pa
Ghaedi M, Ghotbi C, Aminshahidy B (2014) The optimization of
than the Hamedi et al. (2011) and Alarcon et al.
gas allocation to a group of wells in a gas lift using an effi-
(2002).cor- relations. cient ant colony algorithm (ACO). Energy Sources Part A
3. The GA algorithm is highly dependent on population Recovery Util Environ Effects 36(11):1234–1248. https://doi.
and is trapped in low populations at local optimum org/10.1080/15567036.2010.536829
Ghassemzadeh S, Pourafshary P (2015) Development of an intelli-
points, while the water cycle and PSO algorithms are gent economic model to optimize the initiation time of gas lift
not so population dependent and have good operation. J Pet Explor Prod Technol 5(3):315–320. https://doi.
repeatability. org/10.1007/s13202-014-0140-z
4. In terms of the convergence rate, the water cycle algo- Hamedi H, Rashidi F, Khamehchi E (2011) A Novel approach to
the gas-lift allocation optimization problem. J Pet Sci Technol
rithm has a much better performance than the PSO and 29(4):418–427. https://doi.org/10.1080/10916460903394110
GA algorithms. In high populations, the performance Jung SY, Lim JS (2016) Optimization of gas lift allocation for
of the PSO and GA algorithms is reduced significantly improved oil production under facilities constraints. Geosyst
in terms of speed. The PSO algorithm converges to Eng 19(1):39–47. https://doi.org/10.1080/12269328.2015.10848
95
the optimal response faster than the GA algorithm in a Kanu EP, Mach J, Brown KE (1981) Economic approach to oil pro-
smaller population and more rapidly. duction and gas allocation in continuous gas lift. J Pet Technol.
5. In optimization problems that aim to produce a prede- https://doi.org/10.2118/9084-pa
Mahdiani MR, Khamehchi E (2015) Stabilizing gas lift optimization
termined amount of oil, the closer the targeted oil pro-
with different amounts of available lift gas. J Nat Gas Sci Eng
duction rates to the maximum oil production rate in the 26(1):18–27. https://doi.org/10.1016/j.jngse.2015.05.020
gas lift process, the surplus oil producible through the MATLAB Manual, Version 9.5.0.944444 (R2018b), 2018. Natick,
optimization can be reduced. As a result, in lower oil Massachusetts: The MathWorks Inc
Miresmaeili SOH, Pourafshary P, Jalali Farahani F (2015) A novel
production targets, the optimization is more important multi-objective estimation of distribution algorithm for solving
and generates more extra oil. gas lift allocation problem. J Nat Gas Sci Eng 23:272–280. https
://doi.org/10.1016/j.jngse.2015.02.003
Miresmaeili SOH, Zoveidavianpoor M, Jalilavi M, Gerami S, Rajabi
Open Access This article is distributed under the terms of the Crea- A (2019) An improved optimization method in gas allocation for
tive Commons Attribution 4.0 International License (http://creativeco continuous flow gas-lift system. J Pet Sci Eng 172(1):819–830.
mmons.org/licenses/by/4.0/), which permits unrestricted use, distribu- https://doi.org/10.1016/j.petrol.2018.08.076
tion, and reproduction in any medium, provided you give appropriate Nakashima P, Camponogara E (2006) Solving a gas-lift optimi-
credit to the original author(s) and the source, provide a link to the zation problem by dynamic programming. IEEE Trans Syst
Creative Commons license, and indicate if changes were made. Man Cyber Part A 36(2):407–414. https://doi.org/10.1016/j.
ejor.2005.03.004
Namdar H, Shahmohammadi M (2019) Optimization of production
and lift-gas allocation to producing wells by a new developed
GLPC correlation and a simple optimization method. J Energy
References Sources Part A: Recovery, Util Environ Effects. https://doi.
org/10.1080/15567036.2019.1568635
Alarcon G, Torres C, Gomez L (2002) Global optimization of gas Nishikiori N (1989) Gas allocation optimization for continues flow
allocation to a group of wells in artificial lift using non-linear gas lift systems, M.S. Thesis, University of Tulsa
constrained programming. ASME J Energy Resour Technol Nishikiori NN, Redner RA, Doty DR, Schmidt ZZ (1995) An
124(4):262–268. https://doi.org/10.1115/1.1488172 improved method for gas lift allocation optimization. ASME J
Behjoomanesh M, Keyhani M, Ganji-Azad E, Izadmehr M, Riahi S Energy Resour Technol 1995. https://doi.org/10.2118/19711-ms
(2015) Assessment of total oil production in gas-lift process of Rashid K (2010) Optimal Allocation Procedure for Gas-Lift Opti-
wells using Box-Behnken design of experiments in comparison mization. Ind Eng Chem Res 49(5):2286–2294. https://doi.
with traditional approach. J Nat Gas Sci Eng 27:1455–1461. org/10.1021/ie900867r
https://doi.org/10.1016/j.jngse.2015.10.008 Ray T, Sarker R (2006) Multiobjective evolutionary approach to
Buitrago S, Rodrıguez E, Espin D (1996) Global optimization the solution of gas lift optimization problems, IEEE Congress
techniques in gas allocation for continuous flow gas lift sys- on Evolutionary Computation, pp 3182–3188. https://doi.
tems. In: Presented at the SPE gas technology symposium, org/10.1109/cec.2006.1688712
Calgary, Alberta, Canada, SPE Number 35616. https://doi. Sadollah A, Eskandar H, Bahreininejad A, Kim JH (2015) Water
org/10.2118/35616-ms cycle algorithm with evaporation rate for solving constrained
Camponogara E, Conto AM (2009) Lift-gas allocation under prec- and unconstrained optimization problems. Appl Soft Comput
edence constraints: MILP formulation and computational anal- 30:58–71. https://doi.org/10.1016/j.asoc.2015.01.050
ysis. IEEE Trans Autom Sci Eng 6(3):544–551. https://doi. Sharma R, Glemmestad B (2013) A novel multi-objective estimation
org/10.1109/TASE.2009.2021333 of distribution algorithm for solving gas lift allocation problem.
Deb K (2000) An efficient constraint handling method for genetic J Process Control 23(8):1129–1140. https://doi.org/10.1016/j.
algorithms. Comput Methods Appl Mech Eng 186(2/4):311– jngse.2015.02.003
338. https://doi.org/10.1016/S0045-7825(99)00389-8

1
 2 Journal of Petroleum Exploration and Production Technology (2019) 9:2965–2978

Sukarno P, Saepudin D, Dewi S, Soewono E, Sidarto K, Gunawan

A (2009) Optimization of gas injection allocation in a dual gas and Exhibition, San Antonio, Texas’, Society of Petroleum
lift well system. ASME, J Energy Resour Technol. https://doi. Engi- neers, 29 September-2 October.
org/10.1115/1.3185345 https://doi.org/10.2118/77658
Tavakoli R, Daryasafar A, Keyhani M, Behjoomanesh M (2017) -ms
Optimization of gas lift allocation using different mod- Zerafat MM, Ayatollahi S, Roosta AA (2009) Genetic algorithm and
els. Recent Adv Petrochem Sci 1(2):555559. https://doi. ant colony approach for gas-lift allocation optimization. J Jpn
org/10.19080/RAPSCI.2017.01.555559 Pet Inst 52(1):102–107. https://doi.org/10.1627/jpi.52.102
Vieira CRG (2015) Model-based optimization of production systems,
M.S. Thesis, Norwegian University of Science and Technology Publisher’s Note Springer Nature remains neutral with regard to
Wang P, Litvak M, Aziz K (2002) Optimization of production opera- jurisdictional claims in published maps and institutional affiliations.
tions in petroleum fields. In: SPE Annual Technical Conference