SPE 90506

Gas Lift Optimization for Long-Term Reservoir Simulations

Pengju Wang, SPE, BP, Michael Litvak, SPE, BP
Copyright 2004, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the SPE Annual Technical Conference and
Exhibition held in Houston, Texas, U.S.A., 26 29 September 2004.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Socie ty of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous
acknowledgment of where and by whom the paper was prese nted. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083 -3836, U.S.A., fax 01-972-952-9435.

Abstract
Although various gas-lift optimization algorithms have been
proposed in literature, few of them is suitable for long-term
reservoir development studies, which require the gas-lift
optimizer to be highly efficient, flexible and powerful enough
to handle complicated fluid flows and operational constraints,
and have low impact on simulator convergences. This paper
investigated methods to address these important issues.
The gas-lift optimization problem considered in this paper
is to maximize the daily hydrocarbon production by selecting
optimally the well production and lift gas rates subject to
pressure and rate constraints in nodes of the surface pipeline
network and to the amount of lift gas available. The problem
is regarded as a well management problem in a commercial
reservoir simulator capable of simulating multiphase
compositional fluid flow in reservoirs, well tubing strings,
surface pipeline network systems, and separation facilities.
The problem is solved in selected iterations of a reservoir
simulation time step.
This paper proposed a method for the described gas-lift
optimization problem and investigated its performance against
multiple existing methods. Case studies showed that the new
method is capable of producing high quality results while
requires less CPU time for optimization and has smaller
impact on reservoir simulator convergence.
This paper also applied the concept of multiobjective
optimization to smooth the rate oscillation between adjacent
iterations by sacrificing a certain amount of oil production. In
certain cases, this method reduces the simulation time
significantly.
Introduction
When oil field matures, the hydrocarbon production is often
assisted by continuous lift gas injection and constrained by the
gas and/or liquid handling capacities of surface facilities. The
optimal allocation of production rates and lift gas rates subject

to reservoir deliverability and surface facility capacities can

have a big impact on facility design and other capital
investment decisions and should be captured in long term
reservoir studies. Compared to real time production
optimization, the optimal rate allocation in long-term reservoir
simulation studies poses unique problems: the rate allocation
optimizer has to be highly efficient and have low impact on
simulation convergence while is capable of generating quality
results.
The stated problem has been addressed in different ways in
existing literature. Fang and Lo 1 proposed a linear
programming technique to allocate lift gas rates and
production streams subject to multiple flow rate constraints.
The method was implemented in a reservoir simulator and
proved to be efficient in several field studies. Based on Fang
and Los work, Wang et al.2 developed a procedure to
optimally allocate the production rate, lift gas rate, and well
connections to surface pipeline systems simultaneously. The
optimization procedure is invoked at the Newton-iteration
level of a commercial reservoir simulator. Hepgular et al.3
coupled a separate commercial surface pipeline network
optimizer with a commercial reservoir simulator through an
iterative procedure. The surface network optimizer employees
a Sequential Quadratic Programming (SQP) optimization
algorithm and has the ability to perform general operation and
design optimizations. Davidson and Beckner4 presented an
integrated facility and reservoir model in which the rate
allocation problem is solved in the facility model using SQP
methods. They also presented a detailed procedure on how to
handle infeasible conditions.
Fang and Los 1 method is simple and efficient. However, it
ignores the pressure interactions among wells through
common flow lines and may result unsatisfying results.
Hepgular et al.3 and Davidson and Beckners4 methods relied
on powerful facility network optimizers that require
significant effort to develop. This paper bridges this gap by
presenting a simple yet robust and efficient rate allocation
optimization procedure. In addition, this paper applied the
concept of multiobjective optimization5 to minimize the
impact of lift gas rate oscillation on simulator convergence.
This method proved to be successful in certain cases.
Problem Statement
The optimization problem is to maximize the daily
hydrocarbon production by selecting optimally the well
production and lift gas rates subject to pressure and rate
constraints of surface facilities. The optimization problem is
considered a well management problem in VIP 6, a commercial

reservoir simulator capable of simulating multiphase

compositional fluid flow in reservoirs, well tubing strings,
surface pipeline network systems, and separation facilities.
Wang et al.2 presented a procedure to integrate the
optimization problem into the reservoir simulator. The
procedure is presented below.
1. Start with pressure and fluid compositions in reservoir grid
blocks calculated in the previous Newton iteration. Use
well lift gas rates from the previous Newton iteration as
the initial guesses.
2. Solve the surface pipeline network problem, convert
pressure constraints to flow rate constraints. This step was
presented in detail by Litvak and Darlow7.
3. Perform production and lift gas rate allocation
optimization.
4. Determine active constraints in wells and nodes. Linearize
multiphase fluid flow equations for well tubing strings and
the surface pipeline network system (numerical derivatives
are used). Add these equations to the linearized fluid flow
equations for reservoir grid blocks.
5. Solve the linearized system of equations established in
Step 4.
6. Repeat Steps 1 to 5 until convergence.
7. March to the next time step.
Wang et al.2 adopted Fang and Los 1 separable
programming (SP) method to solve the rate allocation problem
in Step 3. This method works as follows.
1. Construct a gas-lift performance curve (oil rate versus lift
gas rate curve) and inflow performance curves (oil rate
versus water rate curve and oil rate versus formation gas
rate curve) for every well on automatic gas-lift allocation.
In current implementation, a minimum gas-lift efficiency
parameter (defined as the oil rate increase for a unit of lift
gas injection) can be specified. A gas-lift performance
curve is constructed in such a way that its slope at the end
of the curve should be larger than or equal to the userspecified minimum gas-lift efficiency.
2. Approximate the gas-lift and inflow performance curves
using piecewise linear curves. Formulate the constrained
gas-lift optimization problem as a linear programming
problem. Solve the linear programming problem and
obtain the optimal lift gas rates.
A detailed description of the above method can be found in
Fang and Lo 1 and Wang8.
When Wang et al.s2 procedure was applied to several
field case studies; two major limitations of that procedure
were exposed:
1. The SP method requires a gas-lift performance curve and
two inflow performance curves for each well on gas-lift
optimization. Each curve has to be established after the
corresponding well is isolated from the surface pipeline
network (SPN) by ignoring the backpressure imposed by
other wells. Consequently, the method may produce
significantly suboptimal solutions when the flow
interactions among wells are significant.
2. The gas lift optimization problem is solved in selected
Newton iterations. Fluctuations of reservoir and operation
conditions can cause significant oscillations of lift gas rate
allocated in different iterations. These oscillations can
make the simulator hard to converge.

SPE 90506

This paper addresses the first and second problem by

presenting a simple yet powerful gas-lift optimization
algorithm. This paper also addresses the second problem by
introducing a multiobjective optimization method to minimize
the lift gas rate oscillations while maximizing the oil
production.
Gas-lift Optimization Procedure
Overall Method. The gas-lift optimization method developed
in this study takes into account the flow interactions among
wells through common surface pipelines. The method works
as follows.
1. Start with the existing lift gas rates for all wells on
automatic lift gas allocation. Solve the multiphase flow
problem in the surface pipeline network (SPN). Build a
linear programming model (described below in section
Constraint Handling) to scale production and lift gas
rates to satisfy flow rate and/or velocity constraints.
Denote the objective function value obtained in this step
as f 0 .
2. Select a well on automatic lift gas rate allocation, say well
0
i . Denote its lift gas rate at this stage as q lg,i
. Increase its
lift gas rate by q lg,i . Solve the multiphase flow problem in
the SPN with the updated lift gas rates and scale
production and lift gas rates to satisfy the flow rate
constraints. Denote the objective function value obtained
in this step as f 1 .
3. Compute the gas-lift efficiency for well i
f1 f0
e=
(1)
q lg,i
Compare the gas-lift efficiency e with a user-specified
minimum gas-lift efficiency coefficient, e min . If e e min ,
update f 0 by setting f 0 = f 1 and go to step 6 with the
increased lift gas rate for well i . If 0 e e min , reset the
0
lift gas rate of well i to q lg,i
and go to step 6. If e < 0 ,
0
reset the lift gas rate of well i to q lg,i
and go to step 4.

4. Decrease the lift gas rate of well i by q lg,i ( q lg,i > 0 ).

Solve the multiphase flow problem in the SPN with the
updated lift gas rates. Scale optimally the production rates
and lift gas rates to satisfy flow rate constraints. Denote the
objective function value obtained in this step as f 2 .
5. Compute the gas-lift efficiency for well i ,
f2 f0
e=
(2)
q lg,i
Compare this gas-lift efficiency with the user-specified
minimum gas-lift efficiency coefficient emin . If

e emin , update f 0 by setting f 0 = f 2 and go to step

6 with the decreased lift gas rate for well i . Otherwise,
0
reset the lift gas rate of well i to q lg,i
.
6. Repeat step 2-5 for every well on automatic lift gas
allocation.

SPE 90506

7. Repeat step 2-6 until no lift gas rate change can be made or
the maximum number of iterations allowed is reached.
v
Constraint Handling. Given a set of lift gas rates qlg , the
v
v
corresponding production rates (oil rates qo , water rates q w ,
v
and formation gas rates q g ) may exceed the flow rate and/or
velocity constraints and be not feasible. The new gas-lift
optimization procedure adopted a linear programming model
developed by Lo and Holden9 to scale the infeasible lift gas
and production rates to the feasible region. This linear
programming model takes a set of flow streams (either from
production wells or from satellite reservoirs) as the input and
scales them to meet the flow rate and velocity constraints in a
way that maximize the objective function. A flow stream is
represented by the unconstrained oil, water, formation gas, and
lift gas rates of a well or a satellite reservoir. For example,
suppose we want to maximize the total oil rate of a field
subject to a total gas rate constraint. The problem can be
formulated as (Problem 3)
nw
(3a)
Maximize x i q o,i
i =1

x (q
nw

Subject to

i =1

g ,i

+ qlg,i ) Qg

(3b)

(3c)
0 x i 1 , i = 1,..., n w
where n w is the number of flow streams, x i denotes the
decision variable for Problem 3, Q g is the total gas flow rate
capacity of the field, and q o,i , q g ,i , and qlg,i are the oil,
formation gas, and lift gas rate for well i , respectively. In the
optimal solution, x i = 0 indicates well i should be shut-in;
x i = 1 indicates well i should produce at rate q o,i , q g ,i , and

qlg,i ; x i (0,1) indicates well i should be choked back.

The optimal objective function value of Problem 3 is a
feasible value to the gas-lift optimization problem and is
v
regarded as the function value for the set of lift gas rates qlg in
Step 1, 2, and 4 of the overall gas-lift optimization procedure.
Discussion. The results generated by the new gas-lift
optimization procedure will be suboptimal because of the
following two facts. First, the function evaluation procedure
employed in the overall optimization algorithm uses another
optimization procedure with simplified assumptions.
Consequently the function value obtained from this procedure
is only an approximation. Secondly, the new method is a local
search method and can be stuck at a local sub-optimal point.
Fortunately, it was demonstrated in several case studies that
the new method produces quality results for long-term
development studies.
Coefficient e min is a parameter used to control how easily
a lift gas rate can escape from its current value. If e min is large,
the lift gas rate is not sensitive to small changes in reservoir
and operation conditions. Consequently the result will be less
near the true optimum but there will be less simulator
convergence problem resulted from lift gas rate oscillations.
Conversely, if e min is small, the allocated lift gas rates will be
more noisy but the solution will be closer to the true o ptimum.

To facilitate later discussions, the above overall gas-lift

optimization method will be referred to as the GLINC method.
Lift Gas Rate Damping
For constrained gas-lift optimization problem, there are cases
that multiple vastly different lift gas distributions result similar
oil rate increases. For such cases, although moving from
current gas-lift injection scenario to another gas-lift injection
scenario may increase the total oil production by a minuscule
amount, the resulting production rates for individual wells can
be significantly different, thus make the reservoir simulator
hard to converge. When gas-lift injection scenarios oscillate
frequently in different Newton iterations, the computational
efficiency of the reservoir simulator deteriorates significantly.
In the GLINC method, this problem can be mitigated by using
a large value for coefficient e min . For the separable
programming method, however, a different strategy was
employed. This strategy was described below.
To minimize the impact of lift gas oscillation on simulator
convergence, the gas-lift optimization problem in VIP was
reformulated as a multiobjective optimization problem5 with
two competing objectives:
1. Maximize the total oil production rate subject to the flow
rate and velocity constraints. This objective can be
expressed mathematically as
nw

f 1 = qo ,i

(4)

i= 1

2. Minimize the absolute change of lift gas rates between two

consecutive Newton iterations subject to the flow rate and
velocity constraints. This objective can be expressed
mathematically as
nw

0
= q lg,i q lg,
i

(5)

i =1

0
where q lg,i
is the lift gas rate of well i allocated in

previous iteration and q lg,i is the lift gas rate of well i to

be allocated in current Newton iteration.
The multiobjective optimization problem was solved by a
hierarchical optimization method10. This method allows the
decision maker to rank and optimize the objectives in
descending order of importance. For this particular gas-lift
optimization problem, the first objective f 1 is ranked the
most important; the second objective f 2 is ranked the least
important. Denote the decision variable of the optimization
v
problem as x . Denote the flow rate and velocity constraints as
v
c i ( x ) 0 , i = 1,..., m
(6)
The solution procedure for the multiobjective optimization
problem goes as follows:
v
1. Find the optimum point x 1,* for the first objective, f 1 ,
subject to the original set of constraints (Problem 7)
v
(7a)
Maximize f 1 (x )
v
(7b)
Subject to c i (x ) 0 , i = 1,..., m
1
8
As described by Fang and Lo and Wang , Problem 7 can
be reformulated to and solved as a linear programming
v
problem. Let f 1,* x 1,* denote the optimal objective
function value for Problem 7 .

( )

SPE 90506

0
be the lift gas rates of previous Newton Iteration, q lg,i
. By
adjusting the damping factor between 0 and 1, the
competition between maximizing the total oil production,
f 1 , and minimizing the discrepancy of lift gas rates

between two consecutive Newton iterations, f 2 , can be

balanced. With appropriate reformulations11, Problem 8
can also be solved as a linear programming problem.
Although two optimization problems need to be solved in
the hierarchical method, the CPU time is increased only
slightly compared to the original single objective optimization
problem provided that the separable programming method is
used to solve Problem 7 and 8. The reason is that for the
separable programming method, the majority of computational
time is spent on constructing the gas-lift and inflow
performance curves; once the performance curves are
established, it takes relatively little extra time to formulate and
solve Problem 7 and 8.
Field Examples
The GLINC gas-lift optimization method and the lift gas
damping method were successfully applied to the long-term
development studies of two North Sea oil fields, respectively.
These application examples are presented below to
demonstrate the advantages and shortcomings of the
developed methods.
Field Example 1. A full field model was developed to
study the long-term development plan of a North Sea oil field.
The reservoir model contains about 20 production wells. All
but one production well in the model are on automatic lift gas
allocation. All production wells are tied to a processing center
through a surface pipeline network system. The production
system is organized in such a way that the production wells
can be classified into two groups, and the wells within each
group share at least one common flow line. It is observed that
the production rate of some wells interfere each other through
the common flow lines.
One objective of this reservoir model is to investigate the
appropriate surface facility capacities of the field. As a
consequence, a total lift gas injection rate as well as total oil,
gas, water, and liquid flow rate constraints are specified in the
model, and the optimal allocation of lift gas rates and
production rates during the simulation is crucial to identify the
right facility capacities.
The full field model was first run with the separable
programming rate allocation method. The minimum gas-lift
efficiency was specified as 50 STB/MSCF. One concern with
this run was whether the SP method is suitable for this model.

1
0.9
0.8
Normalized Oil Rate

( )

The major assumption of the SP method is that for a given lift

gas rate, the oil rate of a well in the entire production system
follows the gas-lift performance curve built from the isolated
single-well system (recall that a well is isolated from the
surface pipeline network by fixing its well head pressure).
Since some wells have noticeable interference through
common flow lines, this assumption does not hold for this
model. This was demonstrated in Fig. 1a, which shows that the
oil rate of well A1 after the first optimization is far away from
the gas-lift performance curve built for that optimization. It
was observed that in subsequent Newton iterations, the postoptimization oil rate of a well converged to the gas-lift
performance curve built for the SP gas-lift optimization in
those iterations (Fig. 1b-1c). However, the concern that
whether the lift gas rates allocated from the SP method are
suitable for this long-term reservoir development study
remained.

0.7
0.6
0.5

Gas lift performance curve

0.4

Allocated lift gas and oil rate

0.3
0.2
0.1
0
0.00

0.20

0.40

0.60

0.80

1.00

1.20

Normalized Gas Lift Rate

Fig. 1a - Gas-lift performance curve and the allocated lift gas and
oil rates for well A1 at the first Newton iteration of the first time
step. The oil rate and lift gas rate s are normalized.

1
0.9
0.8
Normalized Oil Rate

v
2. Find the optimum point x 2,* for the second objective
function f 2 subject to the original and an additional
constraint (Problem 8)
v
(8a)
Minimize f 2 (x )
v
(8b)
Subject to c i (x ) 0 , i = 1,..., m
1 v
1,* v 1,*
(8c)
f (x ) (1 ) f x
where [0,1] is called the damping factor. If equals
v
0, the solution of Problem 8 is x 1,* , and there is no lift gas
rate damping. If equals 1, the solution of Problem 8 will

0.7
0.6
0.5
0.4
0.3

Gas lift performance curve

0.2

Allocated lift gas and oil rate

0.1
0
0.00

0.20

0.40
0.60
0.80
Normalized Gas Lift Rate

1.00

1.20

Fig. 1b - Gas-lift performance curve and the allocated lift gas and
oil rates for well A1 at the second Newton iteration of the first
time step. The oil rate and lift gas rates are normalized.

SPE 90506

1
0.9

Normalized Oil Rate

0.8
0.7
0.6
0.5
Gas lift performance curve

0.4

Allocated lift gas and oil rate

0.3
0.2
0.1
0
0.00

0.20

0.40
0.60
0.80
Normalized Gas Lift Rate

1.00

1.20

Fig. 1c - Gas-lift performance curve and the allocated lift gas and
oil rates for well A1 at the third Newton iteration of the first time
step. The oil rate and lift gas rate s are normalized.

To assess the performance of the SP method, another run

was made with the GLINC method as the rate allocation
method. In this run, parameter emin was set to a small value of
10 STB/MSCF at the beginning of the run so that the
optimization will not stuck at a point far from the optimal lift
gas rates. Parameter emin was set to a bigger value of 50
STB/MSCF after the first time step so that the lift gas rates
change less frequently and the simulator runs f aster.
Although the GLINC method handles explicitly the flow
interactions among wells, it is a heuristic search method that
does not guarantee local optimum. To further verify the
performance of the SP and GLINC method, two conventional
optimization methods were applied to the constrained gas-lift
optimization problem.
The first method is a genetic algorithm12 (GA). In this
method, the lift gas rates for wells on automatic lift gas
allocation are selected as the decision variables and encoded
as a binary string (or genes). The method encodes multiple
sets of solutions (lift gas rate distributions) into populations,
evaluate the fitness of each population (i.e. evaluate the total
oil rate for a given set of lift gas rates subject to the flow rate
constraints), and evolve the populations through the means of
selection, crossover, and mutation. As in the GLINC method,
the flow rate constraints of surface facilities are handled in the
function evaluation procedure by using the linear
programming model9 to scale infeasible flow rates to the
feasible region.
The second method is a direct search algorithm that tries to
maintain a regularly-shaped simplex throughout the iterations.
The method used in this study was implemented by Powell13
and referred to in this paper as the COBYLA method. As in
the GA method, the lift gas rate for each well on automatic lift
gas allocation is selected as the decision variable and the flow
rate constraints of surface facilities are handled in the function
evaluation procedure using the linear programming model9.
The GA and COBYLA methods are time consuming. To
reduce the number of calls to the GA and COBYLA methods,
the runs with the GA method and the COBYLA method have
a maximum time step of 10 days while the runs with the SP
and GLINC methods have a maximum time step of 6 days.

Results from the four different rate allocation methods are

shown in Fig. 2 through Fig. 5. Fig. 2 shows that the field
cumulative oil productions from the four optimization
methods are close; while Fig. 3 shows that the daily field lift
gas injection volumes from the four methods are quite
different. The reason that the SP and GLINC method allocated
significantly less lift gas than the GA and COBYLA methods
is that the SP and GLINC methods have parameters to control
the balance of the injection cost and oil rate increase while the
current implementations of the GA method and COBYLA
method do not have. Fig. 4 and Fig. 5 compare the lift gas rate
and oil rate allocated by the four optimization methods for
well A2, respectively. It is observed that the lift gas rates
allocated by the four methods follow roughly the same trend.
Although the absolute lift gas rate differences between four
methods are significant, the oil rates are similar. This is
because that for a gas-lift well, the gas-lift efficiency
decreases to a small value as the lift gas rate increases beyond
certain value. In the actual field operations, the wells are
operated with lift gas rates similar to those obtained from the
GLINC method. Table 1 shows that the SP method and the
GLINC methods require significantly less CPU time on both
well management and overall simulation than the GA method
and the COBYLA method do.
In summary, for this full field model, both the SP method
and GLINC method produce good results and are much more
efficient than the GA and the COBYLA methods.
Field Example 2. A full field model was developed to
study the long-term development plan for another North Sea
oil field. This model contains about 20 production wells. All
wells are on automatic lift gas allocation when they are in
production.
Two case studies of this model are presented here. In Case
1, total oil, water, and liquid flow rate constraints were
specified. When the rate allocation problem in this case was
solved using the SP method without lift gas rate damping, it
was observed that some wells have significant lift gas rate
oscillations. To overcome this problem, a second run was
made with a lift gas rate damping factor of 0.02 specified. In
the second run, the lift gas rate for individual wells was
greatly smoothened (Fig. 6) while the field cumulative oil
production has discernable differences with that of the first run
only at the very end of the simulation (Fig. 7). Table 2 shows
that the CPU time required by the second run is slightly less
than the CPU time required by the first run.
Case 2 is a variation of case 1. The major difference is that
Case 2 contains a total gas flow rate constraint as well as the
total oil, water, and liquid flow rate constraints of Case 1.
Again, two runs were made to demonstrate the effect of lift
gas rate damping. The first run had no lift gas rate damping.
The second run specified a lift gas rate damping factor of 0.02.
It was observed that for this case, the run with lift gas rate
damping consumed only 32% of the CPU time required by
the run without lift gas rate damping (Table 3) while the lift
rates for individual wells were smoothened with various
degree (Fig. 8) and the cumulative oil production for the field
was only slightly impacted (Fig. 9).

SPE 90506

1.2

0.9

Normalized Oil Rate

Normalized Cumulative Oil Production

1.1

0.8
SP

0.7

GLINC
0.6

COBYLA
GA

0.5

SP
GLINC

0.8

COBYLA
GA

0.6

0.4

0.2

0.4
0.3

2000

4000

6000

8000

10000

2000

Time, Days

6000

8000

10000

Time, Days

Fig. 2 - Field Example 1: normalized field cumulative oil

production allocated by the four rate allocation methods.

Fig. 5 - Field Example 1: normalized oil rate for well A2 allocated

by the four rate allocation methods.
Table 1 - Field Example 1: performance statistics for simulation
runs with different rate allocation methods.

1
SP
0.9

Normalized Lift Gas Rate

4000

GLINC

Number of Time Steps *

Number of Iterations *
Time on Well
Management (min)
Total CPU Time (min)

COBYLA
0.8

GA

0.7
0.6

0.5

SP
1361
4703
196

GLINC
1369
4522
266

GA
960
5136
6793

COBYLA
904
4562
712

2401

2466

9750

3710

The runs with SP and GLINC methods have a maximum time step of
6 days while the runs with the GA and COBYLA methods have a
maximum time step of 10 days.

0.4
0.3

0.9

2000

4000

6000

8000

10000
0.8

Fig. 3 - Field Example 1: normalized field daily lift gas injection

volume allocated by the four rate allocation methods.

1.2

SP
GLINC

Normalized Lift Gas Rate

0.7
0.6
0.5
0.4
0.3
Without Damping

0.2

With Damping

COBYLA
GA

0.8

Normalized Lift Gas Rate

Time, Days

0.1
0
0

0.6

2000

4000

6000

8000

10000

Time, Days
0.4

Fig. 6 - Field Example 2 Case 1: normalized daily lift gas rate for
well A3 obtained from the SP method with damping and without
damping.

0.2

0
0

2000

4000

6000

8000

10000

Time, Days

Fig. 4 - Field Example 1: normalized lift gas rate for well A2

allocated by the four rate allocation methods.

SPE 90506

1
Normalized Cumulative Oil Production

Normalized Cumulative Oil Production

1.2
Without Damping
1

With Damping

0.8

0.6

0.4

0.2

0.9

Without Damping

0.8

With Damping

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

0
0

2000

4000

6000

8000

10000

2000

Fig. 7 - Field Example 2 Case 1: normalized field cumulative oil

production obtained from the SP method with damping and
without damping.
Table 2 - Field Example 2 Case 1: performance statistics for the
simulation run with damping and the simulation run
without damping.

Number of Time Steps

Number of Iterations
Time on Well
Management (min)
Total CPU Time (min)

Without
Damping
2722
10291
177

2716
10388
153

1186

1119

Normalized Lift Gas Rate

With Damping

0.8

0.6

0.4

0.2

0
2000

4000

6000

8000

10000

Table 3 - Field Example 2 Case 2: performance statistics for the

simulation run with damping and the simulation run
without damping.

Number of Time Steps

Number of Iterations
Time on Well
Management (min)
Total CPU Time (min)

Without Damping

6000

Fig. 9 - Field Example 2 Case 2: normalized field cumulative oil

production obtained from the SP method with damping and
without damping.

With Damping

1.2

4000

Time, Days

Time, Days

8000

10000

Time, Days

Fig. 8 - Field Example 2 Case 2: normalized daily lift gas rate for
well A4 obtained from the SP method with damping and without
damping.

Without
Damping
3465
48517
384

2223
9270
132

3674

1180

With Damping

Summary and Conclusion

The developed GLINC method is simple and easy to
implement. Though the method is a local search method
and handles the flow rate constraints with approximations,
it generates results of good quality for long-term
simulation studies, as verified by the separable
programming (SP) method, the GA method, and the
COBYLA method.
Although the SP method does not handle flow interactions
through common flow lines, its execution in consecutive
Newton iterations can mitigate this shortcoming.
The GLINC and SP methods have distinctive
characteristics. The GLINC method is more rigorous in
function evaluation; however it does not guarantee local
optima. The SP method uses significant simplifications in
its function evaluation; but it guarantees the global
optimum of the reformulated linear programming
optimization problem. These two methods can be used as
an alternative and verification method to each other.
It was verified by the GA method and the COBYLA
method that both the SP method and GLINC method are
efficient and capable of generating quality results for some
models with flow interactions among wells through
common flow lines.
For certain cases, the new lift gas rate damping method can
significantly mitigate lift gas rate oscillations of individual
wells and/or reduce the number of convergence failures of
a simulation run, thus decrease the total CPU time
requirement of the simulation studies.

SPE 90506

Nomenclature
v
=
ci (x )

i th constraint
function of decision
v

e
e min

=
=

f
m
nw
Qg

=
=
=
=

variable x
gas-lift efficiency
minimum gas-lift efficiency threshold
used in method GLINC
objective function
number of constraints
number of wells
total gas flow rate capacity

qg ,i
v
q lg
v
qo
v
qw
v
x

formation gas rate of well i , MSCF/d

well lift gas rates, MSCF/d

=
=
=

well oil rate, STB/d

well water rate, STB/d
decision variable of an optimization
problem

Symbol

q lg,i

damping factor for the lift gas rate

damping method
lift gas rate change for well i in the
GLINC gas-lift optimization method

Acknowledgement
The authors wish to acknowledge Peter Clifford and Chris
Macdonald for their support and valuable opinions. The
authors also thank the management of BP for granting
permission to publish this paper.
References
1. Fang, W.Y. and Lo, K.K., A Generalized WellManagement Scheme for Reservoir Simulation, SPE
Reservoir Engineering, May 1996, 116-120.
2. Wang, P., Litvak, M.L. and Aziz, K., Optimization of
Production from Mature Fields, paper presented at the
17th World Petroleum Congress, Rio de Janeiro, Brazil,
Sep 1-5, 2002.
3. Hepguler, G., Barua, S. and Bard, W., Integration of a
Field Surface and Production Network with a Reservoir
Simulator, SPE Computer Applications, Jun 1997, 88.
4. Davidson, J.E. and Beckner, B.L., Integrated
Optimization for Rate Allocation in Reservoir
Simulation, paper SPE 79701 presented at the SPE
Reservoir Simulation Symposium held in Houston, TX,
U.S.A., Feb 3-5, 2003.
5. Miettinen, K., Nonlinear Multiobjective Optimization,
Kluwer Academic Publishers, Boston, 1999.
6. Landmark, VIP-EXECUTIVE Technical Reference, 2003.
7. Litvak, M.L., and Darlow, B.L., Surface Network and
Well Tubinghead Pressure Constraints in Compositional
Simulation, paper SPE 29125 presented at the 13th SPE
Symposium on Reservoir Simulation held in San Antonio,
TX, Feb 12-15, 1995.
8. Wang, P., Development and Application of Production
Optimization Techniques for Petroleum Fields, Ph.D.
Dissertation, Stanford University, California, 2003.

9. Lo, K.K., Starley, G.P. and Holden, C. W., Application

of Linear Programming to Reservoir Development
Evaluations, SPE Reservoir Engineering, Feb 1995, 52.
10. Azarm, S., Class Notes for Multiobjective Optimization,
Retrieved
in
October
2002
from
http://www.glue.umd.edu/~azarm/optimum_notes/multi/
multi.html.
11. Bertsimas, D., and Tsitsiklis, J., Introduction to Linear
Optimization, Athena Scientific, Belmont, Massachusetts,
1997.
12. Holland, J.H., Adaptation in Natural and Artificial
Systems, The University of Michigan Press, Ann Harbor.
1975.
13. Powell, J.M.D., A direct search optimization method that
models the objective and constraint functions by linear
interpolation, DAMTP/NA5, Cambridge, England.