Australian Journal of Basic and Applied Sciences, 3(4): 3919-3929, 2009

ISSN 1991-8178

Continuous Gas Lift Optimization Using Genetic Algorithm

1
Ehsan Khamehchi, 1 Fariborz Rashidi, 2 Behrooz Karimi, 3Peyman Pourafshary, 1Mohammad Amiry

1
Faculty of Chemical Engineering, Amirkabir University of Technology, Tehran, Iran
2
Faculty of Industrial Engineering, Amirkabir University of Technology, Tehran, Iran
3
Petroleum Engineering Institute, Tehran University, Tehran, Iran

Abstract: Gas lift is one of a number of processes used to artificially lift oil or water from wells
where there is insufficient reservoir pressures to produce the well. The process involves injecting gas
through the tubing-casing annulus. Injected gas aerates the fluid to reduce its density; the formation
pressure is then able to lift the oil column and forces the fluid out of the wellbore. Gas may be
injected continuously or intermittently, depending on the producing characteristics of the well and the
arrangement of the gas-lift equipment. Being somehow an ancient tool with an age of over a century,
gas lift is though still a challenging problem when overall optimization is the concern. W hen the
injection gas is of a limited supply the problem is finding the best gas allocation scheme. However
there are increasingly emerging cases in certain geographic localities where the gas supplies are
usually unlimited. The optimization problem then totally travels to the wellbore and completion string
and fully engages with multiphase flow concepts. In the present study an intelligent genetic algorithm
has been developed to simultaneously optimize all effective factors namely, gas injection rate, injection
depth and tubing diameter towards the maximum oil production rate with the water cut and injection
pressure as the restrictions. The computations and real field data are mutually compared.

Key words: continuous gas lift; Genetic algorithm optimization; depth of injection; gas injection rate;
tubing size

INTRODUCTION

W hen oil is first found in the reservoir, it is under pressure from the natural forces that surround and trap
it. If a hole (well) is drilled into the reservoir, an opening is provided at a much lower pressure through which
the reservoir fluids can escape. The driving force which causes these pressured fluids to move out of the
reservoir and into the wellbore comes from the compression of the fluids that are stored in the reservoir. The
actual energy that causes a well to produce oil results from a reduction in pressure between the reservoir and
the producing facilities on the surface. In many wells the natural energy associated with oil will not produce
a sufficient pressure differential between the reservoir and the wellbore to cause the well to flow into the
production facilities at the surface. In some wells also, natural energy will not drive oil to the surface in
sufficient volume. The reservoir’s natural energy must then be supplemented by some form of artificial lift.
There are four basic ways of producing an oil well by artificial lift. These are Gas Lift, Sucker Rod Pumping,
Electric Submersible Pumping and Subsurface Hydraulic Pumping (Takacs, 2005). Gas lift is a widely used
method among artificial lift methods, in which gas is injected into the producing well providing energy to the
flow. Continuous gas lift being cost effective, easy to implement, very effective in a wide range of operating
conditions and requiring less maintenance in comparison to the other alternatives, is one of the most typical
forms of artificial lift in oil production (Ayatollaki et al., 2004). It is a usual technique where there is enough
natural gas resources (Taheri and Hooshmandkoochi, 2006). The basic principle is decreasing the pressure
gradient in the liquid by means of the injected gas, Figure. 1. The resulting mixture becomes less heavy than
the original oil so that it eventually starts flowing .See Figure.1&2 (http://www.answers.com/topic/gas-lift-1).
In gas lift operations, three problems are the most important ones. The first one is finding the optimal
position for injection point and the other is estimating the optimal gas injection rate. These parameters are
interrelated, the more the rate of gas injection the deeper could be the injection point. In other words, the
deeper the injection point the more gas volumes would be needed. The third one is finding the optimal tubing
(string) size. The major problem in gas lift design is the optimization. In the present study a new method is
devised for optimization of continuous gas lift with an unlimited supply.

Corresponding Author: Ehsan Khamehchi, Faculty of Chemical Engineering, Amirkabir University of Technology,
P.O. Box 15914 Tehran, Iran
3919
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Fig. 1: Schematic of a valve position in gas lift well Fig. 2: Schematic of a gas lift well

What Is Gas Lift Optim ization?

Gas being of a limited or unlimited supply completely changes the optimum gas lift system design and
operation. The limited supply problem where the gas allocation is the focal point has been the subject of many
researches, the most notable ones being summarized in Table1. In such a study the researcher is needles of
the physics of the well and the reservoir, and the solution is obtained through several mathematical approaches.

What Are the Main Challenges in Gas Lift Optim ization with Unlim ited Gas Supply (Gopugs)?
In this problem at least three variables are of importance, gas injection rate, tubing internal diameter and
injection location.The current methods have the two following weaknesses:
1- Fundamentally they do not perform optimization and rather, run a sensitivity analysis in fact. Sensitivity
analysis is the next step to optimization and should not be used instead.
2- In the repeated literature sensitivity analysis is run in a single variable fashion and not multi- variable.
This means that when one variable is changed the others are kept constant and in queue for being
analyzed for sensitivity.

Gopugs Com plexity:

An essential question may bear in mind about the reason why this optimization problem has not been yet
attacked though there have been the two aforementioned weaknesses. And the answer lies in the complexity
of multiphase flow problem. The core of nodal analysis is based on the IPR (Inflow Performance Relationship)
that formulates the reservoir response and OPR (Outflow Performance Relationship) that defines the nature of
well performance, the intersection being the operation point as shown in Figureure 3.

Fig. 3: Relation between IPR and OPR (Nodal Analysis)

3920
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

The relationships usually used for IPR are well defined and relatively simple expressions such as Darcy
and Vogel. The difficulty is in solution for OPR and simultaneous optimization of some several variables with
the maximum operating rate as the problem objective. In the multiphase flow problem we are not faced with
an explicit and exact function for which a classical approach suits. The variables are of integer and discrete
nature and relationships and restrictions also are nonlinear. Metaheuristic methods are perhaps the best suitable
way outs currently available and this superiority comes from their independence up on explicit and derivability
of the function. The possible methods are for example, Simulated Annealing, Genetic Algorithm, Tabu Search
and etc. W e used the genetic algorithm.

Solution Approach to Gopugs:

Methodology and Modeling:
Interactive user-friendly software was developed in the VB.NET2008 environment called GOPUGS. The
program is capable of predicting the performance of a single well under natural and gas lifted flows. The
software is composed of different divisions and the solution bone is nodal analysis.

Gopugs Software 3.2.1. Reservoir Control:

In this section some physical properties of the reservoir and the data needed for obtaining IPR are entered.
For the undersaturated reservoirs the solution is based on the Darcy’s equation and for the saturated reservoir
the Vogel’s equation is used.

Well Data:
The parameters entered in this section are pertinent to OPR. The entered data are then manipulated through
a Mechanistic model presented by Ansari. The reason for application of Ansari’s model is its high accuracy
and its intrinsic capability for pressure and flow regime prediction (Ansari et al., 1994; Abdul-Mashat, 2000).

Fluid Data:
In this section fluid data such as API, GOR, Pb and others are entered.

Gas Lift Parameters:

As discussed earlier the variables are gas injection rate, injection depth and tubing size. The first two are
of discrete and integer nature and the third of discrete nature but not integer. An upper limit is also entered
in this section that does not imply limited supply restriction but is taken as a boundary for solution purposes.
The user will also define the number of segmentations. The bigger number of segments used the more accuracy
is obtained in expense of increased solution times. The user can choose a batch of segments for the optimum
segment to be found among them. For choosing the tubing size the user can choose between the default option
that refers to standard IDs in the market or can enter some tubing sizes by hand.
There is another option in the software and that is capability for selection between different gases with
different specific gravities. In this case the problem will have four variables, two of which are integer and the
other two discrete. Two constraints are also considered, gas injection pressure and produced water rate. The
user can arbitrarily switch these options on or off.

Genetic Algorithm :
A genetic algorithm (GA) is a search technique used in computing to find exact or approximate solutions
to optimization and search problems. Genetic algorithms are categorized as global search heuristics. Genetic
algorithms are a particular class of evolutionary algorithms (also known as evolutionary computation) that use
techniques inspired by evolutionary biology such as inheritance, mutation, selection, and crossover (also called
recombination) (Goldberg, 1989). The genetic algorithm method was first introduced by Holland in (1975), this
technique has earned many interests among petroleum production specialists (Mohaghegh, 2000).
GA allows precise modeling of the optimization problem, although not usually providing mathematically
optimal solutions. Another advantage of using GA techniques is that there is no need to having an explicit
objective function. Moreover, when the objective function is available, it does not have to be differentiable
(Lee and El- Sharkawi, 2008).
In the present study we have developed a GA that is novel with respect to selection method. A thorough
flexibility feature is provided in the software the way that GA design parameters like Ps, Pm and Pc and cross
over types are not limited to a definite value or interval and the user is given the chance to select and vary
them according to his/her specific problem. The employed genetic algorithm has the following sections:

3921
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Representation:
After the first generation is randomly selected the following operations may be performed on them in order
for the next generation to be produced.

Chrom osom e Design:

Each chromosome is composed of four genomes. These genomes are tubing diameter, gas injection rate,
the specific gravity of injected gas and injection depth. The four genomes are discrete variables and all
operator and applied methods are also specific to discrete variables.

Initial Population:
A definite number of chromosomes are generated randomly that comprise the first generation. For the next
generations the following operators and methods are used.

Evaluation:
A rank is attributed to each of the chromosomes based on the oil production predicted by the Ansari's
model for that chromosome. If the Ansari's model does not converge or if the chromosome is formed in an
out of borders format then (-1) and (-2) ranks would be attributed to that respectively.

Constraints:
There are two conditions to constrain the chromosomes. These constraints are maximum water production
and the maximum gas injection pressure. If any of these conditions are given a zero value that constraint won't
be applied.

Generation Pool
At any step for the next generation to be produced a pool of chromosomes would participate for
reproduction based on the following rules and with their attributed rankings.

Crossover:
This determines the percentage and the way the offspring chromosomes take after each of their parents.

Parent Mutation:
This is a special case of the crossover operation that exactly duplicates the parents in the next generation.
The more the probability of crossover the longer will be the simulation run but the less will be the probability
of converging to a local optimum.

Rw Select from Parents:

This operator transfers the parents to the next generation based on the ‘Roulette W heel’ and with a
probability proportional to its rank.

Include Parents:
If this option is activated all parents would be directly copied to the generation pool and will have the
chance to compete with the offspring.

Selection:
In this section the user may be able to select between the roulette wheel, and tournament options for
selection. There is also the option for removing one chromosome from options once it is selected. This way
each chromosome would have the chance to be selected utmost once.
The members of the next generation are selected based on the ‘Roulette W heel’ technique the probability
of selection for each member is proportional to its rank. Or the tournament method in which a definite number
of pool members are selected randomly and the best ranked members are then raised to the next generation.
If ‘Use Each Generation Once’ is switched on, the chromosome transferred to the next generation is removed
from pool not to be probably selected again.

Mutation:
In this section the mutation option can be checked on and the mutation probability be entered to the
software. If the ‘Reduce’ option is checked on the mutation probability would approach zero as the run goes
forth and at the last generations the mutation phenomena almost vanishes. This way we won't roust about the
target.

3922
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Elitism :
In case this option is activated one position is always reserved for the best member of the previous
generation.

Run Section:
After all the input variables are entered the run button can be clicked and meanwhile the solution user
is able to monitor fitness and error versus generation number.

Result Section:
In this section the IPR and OPR curves and their intersection point can be viewed. The above mentioned
features were implemented in the software but there is the possibility for developing the software for taking
into account the compositional effects besides the ‘Black Oil’ formulations. Or for the well performance
calculations, we may add models other than Ansari to activate calculations for horizontal well configurations.

RESULTS AND DISCUSSIONS

For the purpose of evaluating the accuracy of the generated model for gas lift optimization, some data
were obtained from three Iranian oil wells that are under gas lift operation. To show that the conventional
optimization method is not an accurate method, the problem is analyzed here using an example. Assume a well
that produces 1000 bbl of an oil with GOR of 420 scf/stb per day with natural flow conditions. If we graph
the injected gas rate or GOR vs. produced oil rate for gas lifted conditions an illustration like Figure 4 would
appear. This Figure shows that at GOR=2000 the oil production would be maximum. As previously mentioned
this Figure is generated by sensitivity analysis on GOR. There are two shortcomings for this point. First it is
calculated based on sensitivity analysis which means that this point is not essentially the optimum point but
is the best among the tested guesses. The second is that this point is obtained based on sensitivity analysis on
only GOR and the rest of variables are kept constant. Let's consider that the optimum point is for example
GOR=1565 that is corresponding to 2420 oil rate. In table 2 the optimum GOR that is 1565 may never be
inspected due to the employed algorithm. If the performance of GOR vs. Qo is graphed we may obtain a
Figure like Figure 4. It is apparent that GOR=2000 is addressed as the optimum point. This means a greater
injected gas volume and at the same time lesser oil production rate. The wells in this study are adopted from
three Iranian oil fields with their specific reservoir and well characteristics as summarized in table 3.After the
needed data were imported to the GOPUGS, the simulations were run for the three wells separately. The results
are shown in Figure 6 to 11. The fitness function versus generation number and IPR against the best OPR are
graphed. The results are also presented in tabulated format in table 4.

Fig. 4: Gas lift Performance with Sensitivity Analysis Fig. 5: Gas lift Performance with Optimization

3923
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Table 2: Sensitivity analysis to GO R

N GOR(scf/stb Q o(stb)
1 420 1000
2 1000 2000
3 1565 2420
4 2000 2300
5 3000 1800

Table 3: Som e of the data for this study

Section Som e Param eters W ell 1 W ell 2 W ell 3
Reservoir Pr(psi) 4000 5000 6500
Pwf(psi) 3200 4650 3800
Qo 3500 4150 3780
K(m d) 10 2 7
well W ell Depth (ft) 10000 12290 14500
Tubing Size 5 5 4.5
Pwh(psi) 526 700 1000
Fluid ão 0.75 0.8 0.68
ãg 0.63 0.65 0.57
GO R(SCF/STB) 700 450 850
W ater Salinity(ppm ) 15000 2000 8500
W O R (% ) 3 1 0.5
Gas lift Q inj (M M SCF/D ) 4.5 5 6
Pinj (psi) 1800 2200 2000
Tubing Size (in) 5 5 4.5
D epth of Injection (ft) 7300 9800 12200

Table 4: Result sum m ery of using GA for 3 wells

Gas Lift Param eters W 1-SA W 2-SA W 3-SA W 1-GA W 2-GA W 3-GA
Tubing size (in) 3.5 5 4.5 4.5 4 6
D epth of Injection (ft) 7300 9800 12200 9523 7100 12250
Q ginj (M M SCF/D ) 4.5 5 6 4.25 3.81 6.21
Q o (STB) 3500 4150 3780 3900 4729 4127

Fig. 6: Error and Fitness Function versus Generation for well#1

3924
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Fig. 7: IPR versus Max OPR for well#1

Fig. 8: Error and Fitness Function versus Generation for well#2

3925
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Fig. 9: IPR versus Max OPR and for well#2

Fig. 10: Error and Fitness Function versus Generation for well#3

3926
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Fig. 11: IPR versus Max OPR for well#3

Fig. 12: IPR versus Max OPR from commercial software output with using of optimized variables with
GOPUGS for well#1

3927
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

Fig. 13: IPR versus Max OPR from commercial software output with using of optimized variables with
GOPUGS for well#2

Fig. 14: IPR versus Max OPR from commercial software output with using of optimized variables with
GOPUGS for well#3

3928
 Aust. J. Basic & Appl. Sci., 3(4): 3919-3929, 2009

To verify the simulation results there are two possibilities. First, modification of field experiences for the
suggested parameters and rerunning the commercial softwares with the given parameters as is suggested here
and comparison with other designs. This is not plausible for the available producing company authorities. The
second way was adopted and the suggested Qinj, Tubing size and Dinj were imported to the W ellflo software.
The resulting Qo values were very close to the GA outputs. The small differences in results maybe attributed
to the use of Ansari's model which is not implemented in W ellflo v.3.6.

Conclusions:
In this research a Metaheuristic method is developed for exact optimization of continuous gas lift in
unlimited gas supply conditions. For this purpose general flexible software is developed and a novel innovative
approach is presented that covers the shortcomings of currently used method. The proposed genetic algorithm
in itself is a novel format in the operators category. Extensions to this study can be compositional
thermodynamic 9 models, incorporating the horizontal well formulations and coupling with a reservoir simulator
for long term prediction purposes.

ACKNOW LEDGEM ENT

Authors would like to thank National Iranian Oil Company-Petroleum Engineering and Development
Company (NIOC-PEDEC) for its support of this research.

REFERENCES

Abdul-Majeed, G.H.and A.M. Al-M ashat ,2000. “A mechanistic model for vertical and inclined two-phase
slug flow” Journal of Petroleum Science and Engineering. 27(1-2): 59-67.
Ansari, A.M, et al. 1994. “A comprehensive mechanistic model for upward two phase flow in wellbore”
SPE. Production and Facility.
Ayatollahi, S. et al. 2004. “Intermittent gas lift in Aghajari oil field, a mathematical study” Journal of
Petroleum Science and Engineering, 42(2-4): 245-255.
http://www.answers.com/topic/gas-lift-1
Goldberg D.E., 1989. “Genetic Algorithms, in Search, Optimization & M achine Learning” Addison W esley.
Holland, J.H., 1975. “Adaptation in Natural and Artificial Systems” University of Michigan Press. Ann
Arbor.
Lee, K.Y. and M.A. El-Sharkawi, 2008. “Modern Heuristic Optimization Techniques” The Institute of
Electrical and Electronics Engineers, Inc. ISBN 978-0471-45711-4 Published by John W iley & Sons, Inc.,
Hoboken, New Jersey.
Mohaghegh, S., 2000. “Virtual Intelligence Applications in Petroleum Engineering: Part 2; Evolutionary
Computing” Journal of Petroleum Technology.
Takacs, G., 2005. “GAS LIFT MANUAL” Petroleum Engineering Department University of Miskolc,
Hungary, Tulsa, USA: PennW ell Books.
Taheri, A. and A. Hooshmandkoochi 2006. “Optimum selection of artificial-lift system for Iranian heavy-oil
fields” SPE 99912. Presented at SPE W estern Regional/AAPG Pacific Section/GSA Cordilleran Section Joint
Meeting, Anchorage, Alaska, USA

3929