Sunteți pe pagina 1din 11

Petroleum Science and Technology, 29:418427, 2011

Copyright Taylor & Francis Group, LLC


ISSN: 1091-6466 print/1532-2459 online
DOI: 10.1080/10916460903394110

A Novel Approach to the Gas-Lift Allocation


Optimization Problem

H. HAMEDI,1 F. RASHIDI,1 AND E. KHAMEHCHI1


1
Chemical Engineering Department Amirkabir University of Technology,
Tehran, Iran

Abstract Because compressed gas is a scarce and expensive resource, allocating an


optimal amount of gas injection to a group of wells to increase the oil production rate
is an important optimization problem in the gas lift operation. In this article, a particle
swarm optimization algorithm is employed to assign an optimum gas injection rate
for each individual well. Also, a new gas lift performance curve-fit that can reduce
the time and volume of the computation is suggested. Finally, the algorithm is tested
on five wells in an Iranian oil field.

Keywords allocation optimization, gas injection rate, gas lift optimization, gas lift
performance curve, particle swarm optimization

1. Introduction
Artificial lift technology is a suitable method for elongating the life of oil wells. The
pump lift and gas lift are the most common systems. In the presence of sufficient gas
volumes, the gas lift returns interesting economic feedbacks. This method has earned
much interest in modern industry. The bottom-hole pressure is reduced and reservoir
fluid is motivated to flow toward the tubing and the wellhead afterwards.
Increasing the gas injection rate will first result in a reduction in the hydrostatic
pressure and the density of the fluid in the tubing, which finally leads to a higher
production rate from the reservoir. As the gas injection rate increases, the pressure drop
term will increase due to the frictional losses and will dampen the oil production rate.
This reversal phenomenon results in the dome shape of the oil rate versus gas injection
rate curve, which is called a gas-lift performance curve (GLPC). For instance, Figure 1
shows some typical forms of GLPC. It should be mentioned that GLPC is obtained
by using nodal analysis method. In this method, the intersection of inflow performance
relationship (IPR) and outflow performance relationship (OPR) is recognized and is taken
as the operating point of the system.
In Figure 1, curve A represents a well with an uneconomic oil production rate that
gas-lift leads to increase in oil rate. Curve B is related to a well that cannot produce
without a gas-lift system. Curves C and D behave as wells not producing without an
initial amount of gas, but at this point, well C has a value for oil production rate. These
two curves make some constraints in the well allocation problem explained more in the
next section.

Address correspondence to Fariborz Rashidi, Chemical Engineering Department Amirkabir


University of Technology, 424 Hafez Ave., Tehran, Iran 15875-4413. E-mail: rashidi@aut.ac.ir

418
Gas-Lift Allocation Optimization 419

Figure 1. Typical forms of GLPC.

If there is sufficient gas for injection, the sum of the maximum point of all curves
yields the optimal oil rate. This approach does not work for an insufficient gas resource
problem. In this case, we have to look for a point with maximum summation of oil rate in
which the sum of the gas injection rates equals the amount of gas available. Mathematical
formulation of the problem is defined as below:
n
X
Max QTO D QOi D f .Qg1 ; Qg2 ; : : : ; Qgn / (1)
i D1

n
X
Subject to W Qgi  Qg avai lable i D 1; 2; : : : ; n
i D1

Qgi  Qg min i i D 1; 2; : : : ; n

Variables W Qgi i D 1; 2; : : : ; n
Qg min i for the wells not requiring the initial gas to start producing equals to zero like
curves A and B and otherwise, it must be set by a zero value.
In this work, a fitting model of GLPC was suggested, and the optimal values
of limited injection gas for each individual well were allocated using particle swarm
optimization (PSO).

2. Literature Review
Several methods for optimizing the gas injection rate have appeared in the literature.
Mayhill (1974) presented the first model to approximate the outflow of an oil well
as a function of the gas injection. Martinez et al. (1994) used a genetic algorithm to
optimize the distribution of the gas lift. Nishikiori et al. (1995) reported a quasi-Newton
method to handle the constraints of the gas compressors capacity. Buitrago et al. (1996)
employed global optimization techniques to optimize the operation of gas-lift systems.
The piecewise linearization of the oil wells outflow and the use of mixed integer program-
ming were presented by Fang and Lo (1996). Alarcon et al. (2002) developed nonlinear
programming to optimize the gas distribution in a well network. Wang et al. (2002, 2008)
used optimization software such as SNOPT to solve the nonlinear problems and developed
a sequential quadratic programming to model gas allocation. A dynamic programming al-
gorithm and piecewise linear formulation was employed by Componogara and Nakashima
420 H. Hamedi et al.

(2003, 2006) to solve gas-lift optimization problems. Zhong et al. (2004) presented a well
penalty function to solve a nonlinear optimization model of gas allocation. Kosmidis et al.
(2005) used mixed integer linear programming to optimize the gas distribution of a gas
lift system and the genetic algorithm model was applied by Ray and Sarker (2007) to
deal with gas allocation problems. Finally, Zerafat et al. (2009) applied an ant colony
optimization (ACO) method to allocate the restricted available gas to a group of wells.
Note that a derivative-based approach cannot fetch up the large-scale problems due to
their intrinsic time-consuming characteristics. Therefore, stochastic methods are desirable
techniques to solve these problems. However, each stochastic method has an individual
convergence speed. In this article, a new application of the PSO technique with high
optimum seeking capacity is presented for solving the allocation optimization problem.
The next section gives an overview of the PSO operation.

3. Particle Swarm Optimization


Particle swarm optimization is a population-based stochastic optimization technique de-
veloped by Eberhart and Kennedy in 1995 (Eberhart and Kennedy, 1995; Kennedy and
Eberhart, 1995). The method was inspired by the movement of flocking birds and their
interactions with their neighbors in the group.
In the past several years, PSO has been successfully applied in various application
areas. PSO provides better results and is faster and cheaper compared to other methods.
Consider a swarm of particles flying through the parameter space and searching
for an optimum. Each particle in the swarm begins with a randomized position (xi )
and randomized velocity (vi ) in the n-dimensional search space, where xi;j represents
the location of particle index i in the j th dimension of the search space. Candidate
solutions are optimized by flying the particles through the virtual space, with attraction
to positions in the space that yielded the best results. Each particle remembers at which

position it obtained its highest performance .xi;j /. Every particle is also a member of
some neighborhood of particles and remembers which particle achieved the best overall
position in that neighborhood (given by the index i 0 ). The equations executed by the PSO
at each step of the algorithm are

vi;j .t C 1/ D vi;j .t/ C c1 Rand./.xi;j .t/ xi;j .t//

C c2  Rand./  .xi0 ;j .t/ xi;j .t// (2)

xi;j .t C 1/ D xi;j .t/ C vi;j .t C 1/ (3)

where is the inertia weight that controls the exploration and exploitation of the search
space, c1 is the weight given to the attraction to the previous best location of the current
particle, and c2 is the weight given to the attraction to the previous best location of the
particle neighborhood. Rand() is a random number between 0 and 1.

4. GLPC Curve-Fitting
Generally, the GLPC is a unimodal function, though the function is not differentiable
and continuous. This function always presents with some points that are achievable using
a nodal analysis method that determines the intersection of IPR and OPR. One of the
known methods in use for the gas-lift allocation optimization problem is fitting of these
points as a continuous function.
Gas-Lift Allocation Optimization 421

In some studies, the second-degree polynomial was used for this purpose. Because
the GLPC is not symmetric, this curve-fitting based on least squares technique does not
work well. Also, Alarcon et al. (2002) suggested an equation defined as:
2
Qoi l D a0 C a1 Qgas C a2 Qgas C a3 Ln.Qgas C 1/ (4)

This model is more accurate than the second polynomial one, although it has an
important disadvantage. This model can cover discontinuous points of the GLPC but
fails to match the trend of curve beyond the maximum point. This fact is illustrated in
Figure 2. Curve D is a GLPC fitted by this model (mostly related to a well with a small
tubing size). But as shown in this figure, after a specific point, the fitted curve goes up
strongly.
Suppose a restricted gas source must be distributed to the wells of which GLPCs were
drawn in Figure 2. If the volume of the available gas lies between zero and maximum of
curve D, the optimum solution will be achievable without any disturbances, but assume
the rate of available gas is beyond the maximum point of the curve D. In this case,
an optimizer fails to obtain the optimal values correctly because it attracts to higher oil
production rates in right side of the curve D which is an unreal area. For instance, in
our optimization model, PSO, if a particle is generated in the mentioned range, other
particles attract to this point mistakenly.
One may suggest as a solution for solving this problem that we can add a constraint
for each well and present the following model:
n
X
Max QTO D QOi D f .Qg1 ; Qg2 ; : : : ; Qgn / (5)
i D1

n
X
Subject to W Qgi  Qg avai lable i D 1; 2; : : : ; n
i D1

Qgi  Qg min i i D 1; 2; : : : ; n

Qgi  Qg max i i D 1; 2; : : : ; n

Variables W Qgi i D 1; 2; : : : ; n

Figure 2. Some GLPCs fitted by Alarcon et al.s (2002) model.


422 H. Hamedi et al.

Figure 3. Curve-fitted for the data from Nishikiori (1989) by the proposed model.

Considering each more constraint in optimization model requires more time for find-
ing optimum values. Therefore, this model increases time of optimization significantly.
For this reason, a suitable fitting model developed in this article as follows:
p
Qoi l D a0 C a1 Qgas C a2 Qgas (6)

Not only does the model fit well with the points of the GPLC, it does not cause the
problem mentioned earlier. Figures 3 to 5 confirm these advantages.
Table 1 shows the results of curve-fitting by the three models. The proposed model
works better than the others due to the greater R2 values in all wells. Furthermore,
it keeps the decreasing trend on the right side of the curve as displayed in Figure 3;
correspondingly, as shown in Figure 4, the Alarcon et al. (2002) model goes up after
these points, and this behavior inconveniences in optimization process as demonstrated
earlier. In order to check mathematically that the suggested model does not have the

Figure 4. Curve-fitted for the data from Nishikiori (1989) by Alarcon et al.s (2002) model.
Gas-Lift Allocation Optimization 423

Figure 5. Curve-fitted for the data from Nishikiori (1989) by the polynomial quadratic model.

mentioned behavior, the differentiation of Eq. (6) must be equal to zero that expressed
as:
dQoi a1 a2
D p C a2 D 0 ! Qgi D 12 > 0 (7)
dQgi 2 Qgi 4a2

This states that Eq. (6) just has one peak matching with the maximum point of the
GLPC, and the second differentiation yields:

d 2 Qoi a1
2
D p (8)
dQgi 4Qgi Qgi

Because the second differentiation of Eq. (6) cannot be equal to zero, the curves
fitted to all GLPCs are always in a dome shape and do not increase rapidly at a specific
point.
Therefore, in this work, the proposed model was preferred for an allocation opti-
mization problem.

Table 1
Results for proposed model and Alarcon et al.s (2002) model for
fitting data from Nishikiori (1989)

R2 of the polynomial
2 2
Well R of the R of Alarcon quadratic model
number proposed model et al.s (2002) model (Nishikiori, 1989)

Well 1 0.997 0.995 0.944


Well 2 0.998 0.986 0.857
Well 3 0.997 0.954 0.789
Well 4 0.994 0.966 0.869
Well 5 0.997 0.974 0.815
424 H. Hamedi et al.

5. Generation of Particles in PSO


In this work, a computer program was established to allocate optimum values of gas
injection for each well. All steps of the optimization operation are as follows. First, the
maximum of each GLPC is obtained and then the sum of these values is computed.
Mathematical formulation of this summation is expressed as follows:

n
X
max
QgT D Qgi (9)
i D1

max a12
Qgi D i D 1; 2; : : : ; n (10)
4a22

If QgT is less than or equal to available injection gas, the answer to the problem
for each individual well is the maximum value (Qgi ) of each GLPC. Otherwise, in an
optional range of two values, n random numbers are generated between the zero and a
positive constant values like [0 a]. Next, these random numbers (Ri ) must be converted
to the values that satisfy all constraints in Eq. (1) using the following conversions:

Ri
Qig D Qg avai lable n (11)
X
Ri
i D1

If Qig < Qg min i then Ri D 0;

Ri
Qig D Qg avai lable n
X
Ri
i D1

for i D 1; : : : ; n (12)

Satisfaction of the two constraints in Eq. (1) is guaranteed by Eq. (11) and Eq. (12)
respectively.

6. Results
In order to verify the PSO model for an allocation optimization problem, we developed
a computer program and used Nishikioris (1989) five-well problem with a limit of 3.0
MMscf/D for available injection gas (as shown in Figures 3 and 4). The parameters used
in the PSO model are given in Table 2. Three methods were applied to tackle this problem
and all the results were presented in Table 3.
As shown in Table 3, our model using PSO and the proposed model of GLPC
calculates a lower value for total oil production rate, 3383.63 Stbo/D, when compared
to the result of Alarcon et al.s (2002) model, 3387.70 Stbo/D. And this is not due to
the fact that Alarcon et al.s (2002) optimization model can find a better solution for
this optimization problem. The difference between the oil production rates results from
Gas-Lift Allocation Optimization 425

Table 2
Parameters of the PSO model for solving
Nishikioris (1989) five-well problem

Number of
particles C1 C2 Iterations

10 1 0.1 0.1 90

overestimation of oil production value in Alarcon et al.s GLPC model. This claim can
be confirmed by the second case in Table 3, employing curve-fiting model of Alarcon
et al. and PSO method for GLPC and optimization model respectively. In the two last
methods of Table 3, although the both curve-fiting models are same and based on Alarcon
et al. model, the PSO method can find a higher total oil production compared to SQP
(Sequential Quadratic Programming) method.
Another example using the proposed model for allocating 10 MMscf/D of available
gas to five wells is related to an Iranian oil field. The GLPC of these wells is shown in
Figure 6 and the results are present in Table 4.
It seems rational that the most available gas is allocated to the wells which tend to
produce more oil as the volume of injection gas increases. Generally speaking, assigning
more gas volume to the wells with higher slope of GLPC leads to higher total oil
production in a well network. As expected, this model assigns the greatest volume of
available gas to the wells having a higher slope value like wells 3 and 4 in Table 4, which
are marked with square and plus signs in Figure 6, respectively. As shown in Figure 6
the curve slopes of the wells 2 and 5 are small; therefore, the less portion of the available
gas is allocated to them. And the optimum allocated gas values for these wells are around
the point at which the curve slopes are quite small.
As can be estimated in Table 4, the oil production rate will accrue to 605 stbo/D
more than the current produced oil if the allocation optimization model improves the gas
injection rate of each individual well.

Table 3
Result of Nishikioris (1989) five-well problem

Sequential quadratic
Optimization programming
method PSO PSO (Alarcon et al., 2002)

Alarcon et al.s Alarcon et al.s


(2002) model (2002) model
The proposed model Qo D c0 C c1 Qg C c2 Qg2 Qo D c0 C c1 Qg C c2 Qg2
GLPC model Qo D c0 C c1 Qg0:5 C c2 Qg Cc3 ln.Qg C 1/ Cc3 ln.Qg C 1/
Variables Qg , MMscf/D Qo , Stbo/D Qg , MMscf/D Qo , Stbo/D Qg , MMscf/D Qo , Stbo/D
Well 1 0.15875 316.557 0.19782 314.56 0.26330 325.18
Well 2 0.57527 578.21 0.59152 585.52 0.58580 581.180
Well 3 0.59045 734.19 0.56672 734.54 0.55750 721.44
Well 4 0.76204 669.44 0.82169 694.24 0.70920 677.41
Well 5 0.91348 1,085.24 0.82225 1,052.24 0.88420 1,082.48
Total 3.0 3,383.63 3.0 3,390.10 3.0 3,387.70
426 H. Hamedi et al.

Figure 6. Curve-fitted for the data from an Iranian oilfield by the proposed model.

Table 4
Results for five wells of an Iranian oilfield

Master Optimum value using


development plan the current model
Well
number Qg , MMscf/D Qo , Stbo/D Qg , MMscf/D Qo , Stbo/D

Well 1 2.1 795 0 0


Well 2 2.1 1,860 1.586266 1,720.741
Well 3 1.5 2,100 3.199817 2,945.652
Well 4 1.5 1,640 3.336355 2,541.691
Well 5 2.8 3,390 1.877563 3,182.041
Total 10.0 9,785 10.0 10,390.12

7. Conclusion
In this work, a model has been presented to fit the GPLC data points. This model covers
all the reported field data in a good agreement. Also a new application of PSO was
employed in allocation optimization of a group of wells under gas lift. Finally this model
was used for an Iranian oilfield and the optimum injection gas values for each individual
well were offered.

Acknowledgment
We gratefully acknowledge the assistance of the National Iranian Oil Company for
providing the data used in this work.

References
Alarcon, G., Torres, C., and Gomez, L. (2002). Global optimization of gas allocation to a group of
wells in artificial lift using nonlinear constrained programming. Energ. Resour. Tech. 124:262
268.
Gas-Lift Allocation Optimization 427

Buitrago, S., Rodrguez, E., and Espin, D. (1996). Global optimization techniques in gas allocation
for continuous flow gas lift systems. Paper No. SPE 35616, Presented at the Gas Technology
Conference, Calgary, Canada, April 28May 1, pp. 375383.
Camponogara, E., and Nakashima, P. H. R. (2003). Applying dynamic programming to a gas-
lift optimization problem. Proceedings of the 2nd Brazilian Conference on Research and
Development in Petroleum and Gas, Rio de Janeiro, Brazil.
Camponogara, E., and Nakashima, P. H. R. (2006). Solving a gas lift production of oil wells:
Piecewise linear formulation and computational analysis. IIE Trans. 38:173182.
Eberhart, R., and Kennedy, J. (1995). A new optimizer using particle swarm theory. Proceedings
of the Sixth International Symposium on Micro Machine and Human Science, Nagoya, Japan,
October 46, pp. 3943.
Fang, W. Y., and Lo, K. K. A. (1996). Generalized well-management scheme for reservoir simu-
lation. SPE Reservoir Eng. 11:116120.
Kennedy, J., and Eberhart, R. (1995). Particle swarm optimization. IEEE Proceedings of the
International Conference on Neural Networks IV, Perth, Australia, pp. 19421948.
Kosmidis, V., Perkins, J., and Pistikopoulos, E. (2005). A mixed integer optimization formulation
for the well scheduling problem on petroleum fields. Comput. Chem. Eng. 29:15231541.
Martinez, E. R., Moreno, W. J., Moreno, J. A., and Maggiolo, R. (1994). Application of genetic
algorithm on the distribution of gas lift injection. Paper No. SPE 26993-MS. Proceedings of
the 3rd SPE Latin American/Caribbean Petroleum Engineering Conference, Buenos Aires,
Argentina, April 2729.
Mayhill, T. D. (1974). Simplified method for gas-lift well problem identification and diagnosis.
Paper No. SPE 5151-MS, Annual Fall Meeting of the Society of Petroleum Engineers of AIME,
Houston, TX, October 69.
Nishikiori, N. (1989). Gas allocation optimization for continues flow gas lift systems. M.S. Thesis,
Tulsa, OK: University of Tulsa.
Nishikiori, N., Redner, R. A., Doty, D. R., and Schmidt, Z. (1995). Improved method for gas lift
allocation optimization. J. Energ. Resour. Tech. 117:8792.
Ray, T., and Sarker, R. (2007). Genetic algorithm for solving a gas lift optimization problem.
J. Petrol. Sci. Eng. 59:8496.
Wang, P., and Litvak, M. (2008). Gas lift optimization for long term reservoir simulations. SPE
Reservoir Eval. Eng. 11:147153.
Wang, P., Litvak, M., and Aziz, K. (2002). Optimization of production from mature fields. Paper No.
32152. Proceedings of the 17th World Petroleum Congress, Rio de Janeiro, Brazil, September
15.
Zerafat, M., Ayatollahi, S., and Rosta, A. (2009). Genetic algorithm and ant colony approach for
gas-lift allocation optimization. J. Jpn. Petrol. Inst. 102107.
Zhong, H., Li, Y., and Liu, Y. (2004). An approach for optimization of gas-lift allocation to a group
of wells and oilfield. Mathematics China 1.
Copyright of Petroleum Science & Technology is the property of Taylor & Francis Ltd and its content may not
be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.

S-ar putea să vă placă și