Sie sind auf Seite 1von 14

SPE

Society of Petroleum Engineers

SPE 19711

An Improved Method for Gas Lift Allocation Optimization


N. Nishikiori, * R.A. Redner, D.R. Doty, and Z. Schmidt,* U. of Tulsa
* SPE Members
II
Copyright 1989, Society of Petroleum Engineers, Inc.
This paper was prepared for presentation at the 64th Annual Technical Conference and Exhibition of the Society of Petroleum Engineers held in San Antonio, TX, October 8-11 • 1989.

This paper was selected for presentation by an SPE Program Committee following review .of informatio~ contained in an abstr~~t~~~~~~e~sbyr~~~~~~o~~~~ ~~~t~~~~~:~~~ ~=~:~t
as presented, have not been reviewed by the Society of Petroleum Engineers and are subJect to correctlo~ by the :~~~:~f~~ Tublication r~vie:by Editorial Committees of the Society
any position of the Society of ~e~roleum En~ineer~, its officer~ or mte~be~s. Pa~~rsnp;~g~~~dds a~~~~~~a~en~~~~~ ~~~be dopied. ~he abstract should contain conspicuous acknowledgment
~~ ~e~~~~~~dE~~i~~~~ ~~!";;!~~rni~o ~~t:S~t~~~t~~ft~ t~~~li~a~~~~ ~a~a~~~eSP~. P.O. Box S33836, Richardson, TX 75083-3836. Telex, 730989 SPEDAL.

Fig.l is a schematic diagram of a typical gas lift system.


The most efficient gas lift system can be obtained by
ABSTRACT analyzing four main components:
A new method for finding the optimum gas injection 1. Flow through a porous media
rates for a group of continuous gas lift wells to 2. Flow through. a production string
maximize the total oil production rate is established. 3. Flow through a flowline and trunkline, and
The new method uses a quasi-Newton non-linear 4. Total injection gas volume available for the
optimization technique which is incorporated with the system.
gradient projection method. The method is capable of
accommodating restrictions to the gas injection rates. Usually, the total injection gas volume available for the
The only requirement for fast convergence is that a system is limited to a certain amount so that a proper
reasonable estimate of the gas injection rates must be allocation of injection gas is critical when designing a
supplied as an initial point to the optimization method. gas lift system.
A method of estimating the gas injection rates is
developed for that purpose. The objective of gas allocation optimization as it relates
to this study, is to maximize the total oil production rate
A computer program is developed capable of from the gas lift system under the amount of gas
implementing the new optimization method as well as volume available to support the system. This objective
generating the initial estimate of the gas injection can be considered as a problem of maximizing a non-
rates. This program is then successfully tested on field linear function which gives the total oil production
data under both unlimited and limited gas supply. The rate from the system. The variables or unknowns for
new optimization technique demonstrates superior this function are the gas injection · rates for each well
performance, faster convergence, and greater and these are subjected to physical restrictions which
application than the standard 11 equal slope allocation 11 are referred to as constraints. For example, the gas
procedure. injection rate for each well must not be negative, and
the sum of the gas injection rates should be less than or
equal to the total gas volume available for the system.
INTRODUCTION
In this study, a non-linear optimization technique is
Continuous flow gas lift is one of the most common introduced into the problem of finding the optimum gas
artificial lift methods employed throughout the world. injection rates. This paper concentrates on presenting
Ideally, sufficient gas should be injected into each a general description of the quasi-Newton optimization
individual well so that it would yield maximum technique along with the results of numerical
revenues. However, in most cases the total amount of experimentation and a comparison with the equal slope
gas available is insufficient to achieve this objective allocation method. A detailed analysis of the theory
simultaneously for every well in the field. Thus, it is surrounding this method can be found in Ref.l.
often necessary to allocate a restricted amount of gas to
each well in some optimal way to achieve maximum oil
production from the field. LITERATURE REVIEW
Historically, several studies have been conducted to
References and figures at end of paper. determine the optimum gas injection rates.

105
2 AN IMPROVED METHOD FOR GAS LIFT ALLOCATION OPTIMIZATION SPE 19711

May hi 112 analyzed the relationship between the gas


injection rate and the oil production rate, and called the . MATHEMATICAL FORMULATION
curve a "gas lift performance curve," which is OF THE PROBLEM
illustrated in Fig.2 He defined the most efficient gas
injection rate as the rate at which an incremental The total oil production (Q 0 ) from n wells, which is the
expense for gas injection is equal to some percentage of sum of the individual oil production rates (q 0 i) can be
the i~cremental revenue produced at that gas injection considered as a function of the individual gas injection
rate. rate.s (qgi). Mathematical formulation of the total oil
production is expressed as follows:
Radden et al. 3 presented an analytical procedure to
determine the most profitable distribution of gas to n
wells in a continuous flow gas lift system.
developed a computer program to perform calculations
They
Oo= Lqoi = f(Qg)
for the gas allocation and it was successfully applied to i=l
a yenezuela field for a group of thirty wells.
= f ( qgl • qg2 • "· • qgn ) ................................... ( 1)

Gomez 4 proposed a procedure to generate the gas lift


performance curve and also developed a computer The gas injection rates (Q g) is represented by an n-
program which fitted the second degree polynomial, q0 dimensional column vector:
= C1 + c2 X qg + C3 X qg2 to each gas lift performance
curve by the least square method. This polynomial was . Qg= ( qgl• qg2• ... • qgn)T, ................................................... (2)
then used to determine which well would produce the
largest amount of oil when equal amounts of where the superscript T denotes transposition. The
incremental gas were injected into each well. This well problem of finding the optimum gas injection rates to
would then be allocated this incremental amount of gas, maximize total oil production is then expressed as:
and this procedure would be continued until all the
available injection gas volume is used.
Max Oo = M a x f ( Qg ) , ................................................ (3)
Qg
Hong 5 investigated the effects of several variables on
continuous flow gas lift systems. He used a system
consisting of six gas lift wells and surface flowlines, subject to the following constraints:
and optimized the system under the condition of
n
variable well head pressures. The pro·cedure for
finding the optimum gas injection rate is basically the
same as that proposed by Gomez. However Hong
L qg i ~ 0g10T ............................................................... <4>
i= 1
employed a cubic spline interpolation technique for
the estimation of the gas lift performance curves. and

Kami et al. 6 established the method of equal slope · qg i ~ 0 for i = 1, 2, ... ,n................................................ (5)
allocation method under both unlimited and limited gas
supply. They presented the formulation of the The constraint defined by Inequality(4) indicates that
economic slope and the use of this slope to allocate a the sum of the individual gas injection rates should be
total amount of gas at the optimal economic point for a less than or equal to the total injection gas volume
group of wells in a step by step procedure. available for the system (QgTOT) and the constraint
defined by Inequality(5) indicates that each gas
In summary, historical development of gas allocation is injection rate must not be negative. Therefore, during
categorized into two different approaches: computation, the gas injection rates should always
satisfy the set of constraints defined by Inequalities(4)
1. The first one is based on an economic and(5 ).
optimization. In this approach, the optimum gas
injection rate is the point where the cost of an
incremental amount of injection. gas is equal to NON-LINEAR OPTIMIZATION THEORY
the profit from an incremental increase of oil
production. The methods developed by Mayhill2, Non-Linear Optimization Methods
Radden et al. 3 , and Kanu et al. 6 are based on this
approach Non-linear iterative methods are usually employed to
solve Eq.(3). These techniques require an initial
2. The second one is a procedure to obtain the gas estimate of the gas injection rates, Q gO, and a search
injection rate which gives the maximum oil direction d k for each iteration, where the superscript k
production obtainable from a group of wells, denotes the iteration number, and the search direction
based on the amount of injection gas volume d k is a n-dimensional column vector:
available for the system. The gas allocation
procedures developed by Gomez4 and HongS
follow this approach. d k = ( d1,
k k
d2, · · ·, dnk )T ...................................................... (6)

The gas allocation optimization which is discussed in The basic procedure used to update the gas injection
the following sections is based on the second approach. rates is described as follows:

106
SPE 19711 N. NISHIKIORI, R. A. REDNER, D. R. DOTY, Z. SCHMIDT 3

Newton's method is simply taken to be Qgk+ak, where ak


Step a : Start at the initial estimate Q go. Set k=O. maximizes the right hand side of Eq.(11). Therefore,
Step b : If the current injection rates Q gk are setting the derivative with respect to a of Eq.(ll) to be
optimum, terminate computation. zero, the following formula can be obtained:
Otherwise, determine the search direction
d k for the current Q gk. a= -rF< ~ > r 1Vf< <Jgk> ................................................. 02>
Step c : Find the step length a k which maximizes
f(Qgk + adk) with respect to a where a>O. Using this formula, the iterative procedure of the pure
Step d : Update the gas injection rates by form of Newton's method can be expressed as:
Qgk+1 = Qgk + akdk, and set k=k+l. Go back
0
-.cg
k+1 = 0"<Kk - F k
[ ( 0-.cg ) ]
-1
Vf( ~k ) ...................................... ( 13)
to Step b.

Different methods have been proposed for calculating If [F (Q gk)] is negative definite, and the initial starting
the search direction d k based on the information point is close to the maximum point, Newton's method
accumulated up to the kth iteration. The procedure exhibits a quadratic rate of convergence 11,12. However
which is undertaken in Step c is called a line search. Eq.(13) requires both first and second derivatives, and
This technique is also an iterative procedure which
the inversion of the matrix [F(Qgk)] at Qgk. which are
generates a sequence of points {a k} until the optimum the main disadvantages of this method.
point along the search direction is located.
The idea underlying the quasi-Newton method is to use
One of the oldest and most widely known methods for a positive definite symmetric matrix approximation in
maximizing a function is the gradient method 11 • 12 (also place of the inverse Hessian:
called the steepest ascent method). This method can be
derived by approximating the objective function with a
first degree Taylor polynomial approximation: - [F(Qgk)J- 1 = Hk ............................................................... (14)

k _ k T k The matrix H is improved from the previous step by


f( ~ +a> = f( 0"<K >+a g , ................................................ <7 > using the change in the gradient and the change in the
injection rates. One of the more commonly used
where a = ad k and gk is the gradient of f at Q gk ' updating formulas was presented by Davidon, Fletcher
and Powell11 • 12,16- 19 (DFP updating formula). This
T method is self-scaling and is given by:
vf <<Jgk > = ( af <~> • af <Ogk> • .. . • af <Ogk> ) •
aqg 1 aqg 2 aqg n
- gk ................................................................. (8)

In order to increase the total oil production rate for


every iteration, the following condition should always where
be satisfied:
T

d
kT k
g >0........................................................................... (9) I =- a;. Yk , ............................................................. <16 >
yk Hkyk
The property given by Inequality(9) is called the
ascent condition. For the gradient method, the search k k+1 k
y =g -g ' .................................................................. (17)
direction for all k is specified to be:
and
dk = gk .............................................................................. (10)

That is to say, the gradient method searches in the ak= Ogk+1_ <Jgk................................................................... <18 >
steepest ascent direction and the ~earch direction
guarantees that the objective function can be increased The search direction, d k is given by:
for some positive scalar a. Further analysis 11 • 12 proves
that in general, the gradient method converges
dk-Hk
- g k......................................................................... (19)
linearly to the optimum point. Therefore, in practice,
this method frequently converges slowly, which is its
main disadvantage. The rate of convergence of the quasi-Newton method
described by Eqs.(13)-(19) is superlinear, which is
A faster rate of convergence can be obtained by faster than the linear rate of the convergence of the
Newton's method. Newton's method can be derived from gradient method. On the other hand, this method is
the second order Taylor polynomial approximation: slower than the pure form of Newton's method which
has a quadratic rate of convergence. However, the
quasi-Newton method has several advantages over
Newton's method:

where F ( Q g k) is the Hessian matrix of second 1. Only first derivatives are required; and
derivatives. The updated gas injection rates Q gk+ 1 in 2. Computation of the Hessian matrix and its
inversion are not required.

107
4 AN IMPROVED METHOD FOR GAS LIFT ALLOCATION OPTIMIZATION SPE 19711
Constrained Optimization
ESTIMATION OF GAS INJECTION RATES
The two sets of constraints defined by Inequalities(4),
and (5) make a total of n+ 1 linear inequalities of the Procedure for Estimating Initial Gas Injection Rates
following form:
The gas allocation optimization algorithm requires an
T initial guess for the gas injection rates, Q g 0, which
ni Og - bi ~ 0 for i = 0, 1,- · · ,n , .................................... (20)
satisfies all the constraints defined by Inequalities (4)
where the n i are n-dimensional column vectors and (5). Since these constraints are relatively simple,
initial gas injection rates which satisfies all constraints
can be easily found. However, if the initial gas
·- ( n·11• n·12• ··· ,n1n
n .- . )T................................................. (21) injection rates, Q gO, are far from the optimum gas
injection rates, most of the computation will be spent
It is assumed that the n i are linearly independent unit on the process of moving towards the optimum gas
vectors which satisfy: injection rates. This results in excessive computation
time or function evaluations when compared with the
n computation which starts from a good guess for the
I nil 2
=n t ni = L:<nij )2 = 1, for i =0, 1, ···,n ....... (22) initial gas injection rates. Therefore, a good estimate of
the gas injection rates is essential to produce an
j=1
efficient gas allocation algorithm.
A constraint is active when strict equality in (20) holds. A number of initialization procedures were considered,
If a total of q constraints are active, then define the
nxq matrix N q by: however the following three procedures proved to be
the most successful.
Nq = [01, "2' .. ·"q ], ..................................................... (23) Equal Gas Allocation : One possible initial starting point
is to inject equal rates of gas into each well, i.e.,
and the q-dimensional column vector bq by:

0 0gTOT
bq = ( b 1' b2, ... 'bq )T....................................................... (24) qgi = - n - for i = 1, 2, ···,n ........................................ (29)

The system of equations for q constraints defined by so that:


inequality (20) can be expressed as:
0 0 0 0 T
~ = (qgl•qg2····,qgn) ............................................... (30)
N:Qg- bq ~ 0 ............................................................... (25)

The constraints for the gas injection rates formulated Productiyity Index Allocation Another possible way is
by Inequality(25) can be incorporated into the to allocate gas in proportion to the productivity index
updating formula for Q g given by Eq.(13) by defining a (PI) for each well, i.e.,
projection matrix, P q. When q constraints are active,
P q is defined by:
o Pli
q g i = Qg TOT x -n-- for i = 1, 2, · · ·, n. ....................... (31)
1
P q =I -Nq( N:Nq )- N:. ................................................ (26)
I,Pii
The algorithm established in this study for gas i= 1
allocation optimization uses the quasi-Newton updating
formulas given in Eqs.(13)-(19) in addition to Maximum Liguid Production Allocation : A procedure
incorporating the projection matrix. When the established in this study is to allocate gas in proportion
number of active constraints remains unchanged, the to the maximum liquid production rate possible for each
updating formula for H q k follows Eq.(15). However, well. First, consider the case where the total amount of
when the jth constraint n j is added, increasing the gas is limited, then the initial gas allocation is given by:
number of active constraints to q+ 1, the updating
formula is replaced by: o Ou
qgi= QgTOT x -n-- for i = 1, 2, ···,n ........................ (32)
H qk nj n jTHqk
k+l Hk
H q+l = q- T k .•.....•..............•...•..•••...••....•..•.. (27) LC4.i
i= 1
n j Hq nj

Conversely, when the qth constraint is removed, H q -1 k The maximum production rate (QLi) can usually be
obtained by a complete production system analysis
can be obtained from H qk as follows:
which takes into account all components of the
production system and utilizes two-phase flow
Hk k Pq-lnqn:Pq-1 correlations. However, such calculations are
q-1 = Hq + T •••••..•••••••••••••••.•••.••••••.•••••• (28) complicated and expensive. Therefore approximations
n q Pq-l nq and simplifications are introduced to obtain a quick
estimation of the maximum liquid production rate from
each well without losing significant accuracy. The
Detailed discussion including the computational approximations and simplifications utilized in this
algorithm can be found in Ref.l. study are as follows:

108
SPE 19711 N. NISHIKIORI. R.A. REDNER. D.R. DOTY Z. SCHMIDT 5

1. Flowing pressure curves in the vertical tubing the points fall between ±. 20% of the optimum gas
string above a gas injecti~n point is appro~i~ated injection rate, which signifies the accuracy of the
by a straight line wh1ch has the mm1mum initial gas estimation method given by Eqs.(32)-(35).
pressure gradient. The minimum pressure
gradient is read from the Hagedorn and Brown
pressure gradient curves9. The flowing pressure DISCUSSION OF COMPUTATIONAL RESULTS
gradient below the gas in~ection. point is al~o
approximated by a straight hne wh1ch . has a stat1c A computer program for the gas allocation optimization
pressure gradient. algorithm established in this study (Eqs.13-28, 32-35)
was written and tested on actual field data.
2. The injection gas pressure at a certain depth is Computational performance of the algorithm was
calcula,ted by adding the static gas column weight judged on both accuracy and the rate of convergence.
to the surface operating gas pressure. The gas
injection depth (L) is determined when the Criteria for Accuracy and Termination
difference of injection gas pressure and the
flowing tubing pressure is 150 psi. The accuracy can be represented by the norm of the
projected gradient which is expressed as:
3. The flowing pressure gradient in the horizontal
flowline is approximated by a quadratic function 2
which is generated from the Beggs and Brill
pressure gradient correlation9. i(iPqij'gjJ , ........................... (36)
i=l j=l
For the case when the total gas supply is unlimited or
very large, the above procedure must be modified. In and ideally liP q gil = 0 at the optimum point. The rate of
order to estimate the gas injection rates for an convergence is measured by the relationship between
unlimited gas supply, the gas-liquid rat~o at the
the liP q g II and the number of function evaluations
maximum production is estimated by assummg that a
gas-liquid ratio of 200 SCF/STB should be necessary for during the process of locating the optimum gas
injection rates. The number of function evaluations is
each 1000 ft of liftS· For example, if a well is producing increased by one when the oil production rate for each
Q Li STB/D of liquid with a gas injecti.on depth of .Li .ft well is calculated for the desired gas injection rate. The
from the surface, the estimated max1mum gas hqmd number of iterations is also counted during the
ratio and the total injection gas volume required for n computation. One iteration is completed when the
wells can be calculated from: matrix H qk is updated by Eq.(15) or Eq.(27).

GLRMax i = (
• 1000
2:l_)
x200 , for i =1, 2, ... ,n , .............. (33) Eventually, the computation must be terminated when
the current result is close to the optimum. This is
accomplished by specifying reasonable termination
and criteria for the computational algorithm. The
termination criteria used in this study are summarized
n
in Table 1, and if one of these criteria is satisfied during
QgMax = ~·L(GLRMax,i X Q,i) ................................. (34) the computation, the gas allocation algorithm
10 i=l terminates immediately.

If the total gas volume available for the system is more Computational Procedure
than the value calculated by Eq.(34}, the initial gas
injection rate for each well is given by the following In view of the possibilities for choosing the numerical
equation rather than using Eq.(32): parameters, a sensitivity analysis is first performed 1
and the following numerical scheme is employed:
q q = - 1- ·GLRMax i x Q i, for i = 1, 2, ... , n ............. (35)
g1 106 • .
1. Oil production rates are calculated by a production
system analysis at specified gas injection rates.
Effect of Initial Gas Injection Rate Estimation For other gas injection rates, the oil production
rates are calculated by interpolation using the
A good initial estimate of the optimum gas injection nearest four points to the desired gas injection
rates is extremely important for fast convergence of rate.
the algorithm. Computational results confirm that the
algorithm converges to the same optimum gas injection 2. Central difference approximations are employed
rates regardless of the initial gas injection rates used. for the computation of first derivatives. The step
However, when the gas allocation algorithm starts . from size of the finite difference is calculated by
the gas injection rates determined by the Max1m~m taking 5 % of the current gas injection rate.
Liquid Production Allocation procedure, fewer funcuon
evaluations are required when compared to the other 3. Initially, the matrix H is set to the Identity matrix
two initialization procedures. This observation has been I, and a self-scaling factor, y, is multiplied by H q
confirmed for a number of different gas lift systems. only at the initial updating stage.
The reduction is especially significant when a large
amount of total injection gas volume is available. Fig.3
shows the estimated gas injection . rates generated by 4. For the base case, the initial gas injection rate for
the Maximum Liquid Production Allocation procedure each well is given by Eq.(29), with an upper limit
plotted against the optimum gas injection rates. Most of of 2.0 MMSCF/D.

109
AN IMPROVED METHOD FOR GAS LIFT ALLOCATION OPTIMIZATION SPE 19711
Computational Results which is illustrated in Fig.1. In all cases tested, the
computational results were similar to the example
In order to demonstrate the speed and accuracy of the presented.
gas allocation algorithm, it is applied to a gas lift system
consisting of 13 wells which are directionally drilled Comparison a&ainst the Equal Slope Allocation Method
from an offshore platform. These wells are directly
connected to the separator so that all the wells have a Because the equal slope allocation method is in common
constant well head flowing pressure. The general us.e, it is desirable to compare the method developed in
descriptions of the wells are tabulated in Table 2 and this study to that method. However, such a comparison
illustrated in Fig.4. This case is cited from Ref.4, in is difficult because the internal structure of both
which detailed well completion data can be found. The methods are completely different. For example, the
gas injection rates tabulated in Table 3 are used to equal slope method requires the generation of the
calculate the oil production rate. complete gas lift performance curve for each well
before starting the gas allocation procedure. Thus, the
The step by step progress of the computation is total number of functional evaluations required
presented in Fig.S through Fig. 7. The total injection gas depends in part on how closely the points are spaced in
volumes considered are; unlimited, 30 MMSCF/D, and 10 order to generate these graphs. On the other hand, the
MMSCF/D, respectively. In these figures, the total oil number of functional evaluations for the quasi-Newton
production rate from the system, and the value of method depends strongly upon the accuracy. of the
liP q g II are plotted against the iteration number, which initial guess along with the stopping criteria.
is represented by ITER. It can be observed that the
algorithm converges to the optimum point quickly. However, it still is possible to compare the
When the computation starts, computed total oil computational efficiency of the two methods if
production increases rapidly during the first several reasonable choices are made for each of the two
iterations, then the rate of increase becomes much methods. Table 5 contains the total number of function
slower. The computed gas injection rates also move evaluation required to maximize the production for the
close to the optimum allocation rates within the first gas lift system summarized in Table 4 for different
several iterations, after which the relative movement amounts of total injection gas. Observe that the equal
between iterations is small. This behavior can be seen slope allocation method requires the same number of
from Fig.8, where the movement of the gas injection function evaluations for all cases because the method
rates for three wells are plotted against the iteration requires the complete generation of all gas lift
number for the case when the total injection gas performance curves. However, the quasi-Newton
volume is 30 MMSCF/D. Starting from the initial gas allocation procedure does not require the generation of
. injection rate of 2.0 MMSCF/D, each gas injection rate the entire gas lift performance curve for each welL
moves close to the optimum after three iterations. Therefore, this new method typically reduces the
required number of function evaluations when
Table 4 summarizes the calculated optimum gas compared with the equal slope allocation method.
injection rates for different total injection gas volumes.
When an unlimited amount of gas is available to the Flexiability of the New Gas Allocation Method
system, the gas allocation algorithm converges quickly
to the optimum gas injection rates corresponding to the This study has verified that the new gas allocation
maximum possible total oil production for this system. method is both fast and efficient when used to maximize
For unlimited gas, there are no active constraints production from a field of gas lift wells. However, there
throughout the computation. However, if the total are other existing procedures, namely the equal slope
injection gas volume is limited to 30 MMSCF/D, the allocation method, that can be used to solve the same
constraint defined by Inequality(4) becomes active and problem. Most of these procedures are constructed in
the optimum gas injection rates are below the values such a manner that it is difficult if not impossible to
which would yield maximum oil production for each adapt them to new and different applications. The gas
well. Further, if only a small amount of injection gas allocation optimization method developed in this study
volume is available, such as 10 MMSCF/D, the has flexibility as one of its most desirable attributes.
constraints defined by Inequality(S) become active for Be~a~se the optimization algorithm is designed to
several wells with the result that no gas will be optimize almost any quantity, it is a simple task to
allocated to some wells (i.e., Wells 2, 7, and 10). modify . the existing procedure to optimize more complex
production systems. In particular, the objective can be
Fig.9 compares the rate of' convergence of the quasi- changed from maximum production to maximum
Newton method and the gradient projection method in economic profitability, or the constraints can be
which the search direction for all k is simply defined changed from limited total injection gas to constraints
.by: specific to an individual field. The optimization
algorithm does not limit its range of application.
k k
d=Pqg ......................................................................... (37)
CONCLUSIONS
Fig.9 illustrates that when there is no active constraint,
the quasi-Newton method converges much faster than
1. A new and improved method of gas allocation
the gradient projection method. The superlinear rate optimization is established. The algorithm follows
of convergence of the quasi-Newton method can be the quasi-Newton matrix updating scheme in
seen from this figure. Such superior performance of conjunction with the gradient projection method.
the quasi-Newton method was also confirmed by two
other two gas lift systems; a system consisting of five , 2. Inequality constraints can be incorporated
wells directly connected to the separator which is efficiently into the gas allocation optimization by
similar to the first system discussed above, and a system
the use of a projection matrix, and it enables the
consisting of five wells with flowlines and a trunkline

110
SPE 19711 N. NISHIKIORI, R.A. REDNER, D.R. DOTY, Z. SCHMIDT 7

method to handle both unlimited and limited qg gas injection rate for each well, MMSCF/D
amounts of total injection gas volume. [std m3/d]
3. A new method of estimating the initial gas
injection rates is also established in this study. In Qgror = total injection gas volume, MMSCF/D [std
this method, a total injection gas volume is m3/d]
allocated to each well in proportion to the
maximum liquid production rate possible for the maximum liquid production rate for a well,
well, which is calculated by a simplified STB/D [stock-tank m3 /d]
production system analysis. Numerical
experiments indicate that most of the estimated total oil production rate from the system,
gas injection rates lie between ±. 20% of the STB/D[stock-tank m3 /d)
optimum values. Numerical experimentation
confirmed the improved performance of the gas qo oil production rate from a well, STB/D
allocation optimization method for both limited [stock-tank m3/d]
and unlimited gas supply when used in
conjunction with the initial estimation method.
I
correction to g: yk = gk+ 1 - gk
y
4. A computer program was developed based on the
algorithm and tested on field data. Greek Letters

5. Numerical experimentation indicated that the a = step length


new method requires fewer function evaluations
to locate the optimum gas injection rate when y scale factor defined by Eq.(16)
compared with the equal slope allocation
procedure. a correction to Qg: ak = Qgk+ 1 - Qgk

Subscripts
6. Since the only requirements for the new gas
allocation optimization method are specifying the i,j indices for components
initial estimate for the gas injection rates along
with the objective function which is to be q index for the number of active constraints
optimized, this method can be easily adapted to
more complex production systems as well as MAX maximum value
modified to perform economic profitability
analysis. Superscripts
-1 inverse of the matrix
NOMENCLATURE
0 initial
d vector of search direction
k kth iteration
di component of d, MMSCF/D [std m3/d]
Operators
F a symmetric negative definite matrix
v partial derivative operator
f(Qg) objective function
II· II norm of the vector
g vector of gradient

= component of g, STB/MSCF [stock-tank


ACKNOWLEDGEMENT
m3/std m3]

H = approximation to the Hessian matrix The authors wish to express their sincere thanks to Dr.
Bashir Agena of Computer Enhanced Artificial Lift
I Identity matrix Consulting, Inc. for providing the computation results I

performed by the equal slope allocation method.


L gas injection depth, ft [m]

N = matrix representing constraints REFERENCES


ni = unit vector representing a constraint
1. Nishikiori, N. : " Gas Allocation Optimization for
nij component of Di Continuous Flow Gas Lift Systems," M.S. Thesis, U.
of Tulsa (1989).
Pq projection matrix defined by Eq.(26)
2. Mayhill, T. D. : "Simplified Method for Gas-Lift
P q ij = component of P q Well Problem Identification and Diagnosis,"
Paper SPE 5151; presented at the SPE 49th
Qg = vector of gas injection rates Annual Fall Meeting, Houston, TX, Oct. 6-9, 1974.

111
8 AN IMPROVED METHOD FOR GAS LIFT ALLOCATION OPTIMIZATION SPE 19711
3. Radden, J. D., Sherman, T. A. G., and Blann, J. R. : 19. Shanno, D. F. : "Conjugate Gradient Methods
"Optimizing Gas-Lift Systems," Paper SPE 5150 ; with Inexact Searches," Mathematics of
presented at the SPE 49th Annual Fall Meeting, Operation Research ( August, 1978 ), pp. 244-
Houston, TX, Oct. 6-9, 1974. 256.
4. Gomez, V. : "Optimization of Continuous Flow 20. Goldstein, A. A. : "On Steepest Descent," J SIAM
Gas Lift Systems," M.S. Thesis, U. of Tulsa, 1974. Control ( 1965 ), pp. 147-151.
5. Hong, H.T. :"Effect of the Variables on 21. Powell, M. J. D. : "Restart Procedures for The
Optimization of Continuous Gas Lift System," M.S. Conjugate Gradient Method," Math. Programming
Thesis, U. of Tulsa, 1975. ( 1977 ), pp. 241-254.
6. Kanu, E. P., Mach, J., and Brown K. E. : 22. Powell, M. J. D. : "A Survey of Numerical
"Economic Approach to Oil Production and Gas Methods for Unconstrained Optimization," S I AM ·
Allocation in Continuous Gas Lift," J Pet Tech , ( Reyiew ( January, 1970 ), pp. 79-97.
October 1981 ) pp. 1887-1892.
23. Lenard, M. L. : "A Computational Study of Active
7. Clegg, J. D. : "Discussion of Economic Approach
Set Strategies in Nonlinear Programming with
to Oil Production and Gas Allocation in Linear Constraints," Math Programming ( 1979 ),
Continuous Gas Lift," J Pet Tech., ( February
pp. 81-97.
1982 ) pp. 301-302.

8. Brown, K. E. : The Technology of Artificial Lift


Methods, Vol. 2a, PenWell Publishing Co., Tulsa SI METRIC CONVERSION FACTORS
OK ( 1980 ), pp. 186-202.

9. Brown, K. E. : The Technology of Artificial Lift


0
API 141.5/( 131.5+ 0 API) = g/cm3
bbl X 1.589 873 E-01 = m3
Methods, Vol. 3a, PenWell Publishing Co., Tulsa
OK ( 1980 ).
ft X 3.048* E-01 = m
ft3 X 2.831 685 E-02 = m3
10. Brown, K. E. : The Technology of Artificial Lift
in. X 2.54* E+OO = em
Methods, Vol. 4, PenWell Publishing Co., Tulsa
psi X 6.894 757 E+OO = kPa
OK ( 1984 ), pp. 169-184.
scf/bbl X 1.801 175 E-01 = std m3tm3

11. Luenberger, D. G. : Linear and Nonlinear *Conversion factor is exact


Programming, Second Edition, Addison-Wesley
Publishing Co., Reading MA ( 1984 ).
TABLE 1
12. Fletcher, R. : ~P~ra~c:.::.ti~c~al~~M~e:.::.th~o~d~s"--~o~f____;O~pt~im~iz~a~ti>..>::o=n,
Second Edition, John Wiley & Sons Ltd., New TERMINATION CRITERIA FOR GAS ALLOCATION
York NY ( 1987 ). ALGORITHM
13. Crowder, H., and Wolfe P. : "Linear Convergence Criterion( I) Projected gradient becomes small:
of the Conjugate Gradient Method," IBM J Res
Deyelop ( July, 1972 ), pp. 431-433.
II P q gk II < 1 x 10-3 ( STB/MSCF).
14. Shanno, D. F., : "Matrix Conditioning and
Nonlinear Optimization," Math Programming ( Criterion(2) Each element of the search
1978 ), pp. 149-160. direction becomes small:

15. Rosen, J. B. : "The Gradient Method for Nonlinear dik ~ 1 x 10-3 ( MMSCF/D ) for i = 1, 2, ... , n .
Programming. Part I. Linear Constraints," L
Soc lndust Appl Math ( March, 1960 ), pp. 181- Criterion(3) Negligible increase in total oil
217. production:

16. Goldfarb, D. : "Extension of Davidson's Variable


Metric Method to Maximization Under Linear
Inequality and Equality Constraints," SIAM J.
Appl Math ( July, 1969. ), pp. 739-764.
Criterion(4) Negligible change of the gradient
17. Goldfarb, D., and Lapidus, L. : "Conjugate vector:
Gradient Method for Nonlinear Programming
Problems with Linear Constraints," Industrial
and Engineering Chemistry Fundamentals
( February, 1968 ), pp. 142-151.

18. Gill, P. E., Murray, W., and Wright, M. H. : Criterion(4) Slope at Qgk is small:
Numerical Methods for Constrained Optimization,
Academic Press, London ( 1974 ), Chapter I, II,
III.

112
SPE 1 9 7l 1

TABLE2 TABLE3

GENERAL DESCRIPTIONS OF THE WELLS GAS INJECTION RATE ASSIGNED FOR THE
CALCULATION OF OIL PRODUCTION RATE
Minimum Maximum
Inclined depth (ft) 9478 15322 Data qg
Vertical depth (ft) 8654 9470 .NQ.. CMSCF/D)
Tubing outside diameter (in.) 3 1/2 4 1/2
Formation GOR (SCF/STB) 170 1435 1 0
Average reservoir pressure (psi a) 2964.7 3664.7 2 50
Productivity Index (STB/D/psi) 0.73 5.68 3 100
Water cut (%) 0.20 57.5 4 200
API oil gravity ( 0 API) 34 34 5 400
6 600
7 1000
8 2000
9 3000
10 4000
11 5000
12 7000

TABLE4

CALCULATED QPfiMUM GAS INJECTION RATES AND


OIL PRODUCTION RATES FOR THE SYSTEM

QgTOf Unlimited 30 MMSCF/D 10 MMSCF/D TABLES


WELL qg qo qg q0 qg q0
No MMSCF/Q STB/D MMSCFIP STB/D MMSCF/D STB/D NUMBER OF FUNCTION EvALUATIONS REOUIRED TO
WELL 1 5.383 6942 3.962 6906 1.534 6523 LOCAlE THE OPTIMUM GAS INJECTION RATES
WELL 2 5.309 1060 1.012 1004 0 857
WELL 3 5.212 4121 3.294 4080 1.334 3777 Qgror This study Equal slope
WELL 4 4.518 4689 2.807 4649 0.548 4301 (MMSCF/D) (qgOby Eq.32) allocation
WELL 5 5.979 1104 1.590 1050 0.435 885
WELL 6 5.463 2812 2.706 2759 1.114 2523
WELL 7 4.736 1143 0.577 1093 0 1038 Unlimited 70 138
WELL 8 5.194 2570 2.190 2519 0.817 2318 50 67 138
WELL 9 4.887 4610 3.164 4570 1.026 4245 30 73 138
WELL10 4.449 1321 0.391 1271 0 1236 20 68 138
WELL11 6.300 3492 4.050 3443 1. 743 3080 10 71 138
WELL12 4.692 3268 2.439 3221 0.836 2976 5 71 138
WELL13 6.263 989 1.818 936 0.613 766
Total 68.385 38121 30.000 37501 10.000 34525 Average 70 138

113
SPE 1 9 711

Gas Distribution Line


High Pressure Gas •'.8--------
Low Pressure Gas

Compressor
To Sales

WELL2
Trunk line
WELL 3

Separator

Reservoir

Fig.1 - Schematic Diagram of A Gas Lift


System

qoMax Slope= 0

~z
0
5
::::>
I
I
I
0 I
0 I
~ I
I

d I
0 I I
I I
Llq I

-
I ~ tJ. gMax
GAS INJECTION RATE ( SCF/D)

Fig.2 - A Typical Gas Lift Performance Curve

114
SPE 1 9 71 1
+20%
10.0

~
C,)
Cll
~ 8.0 -20%
~

~
1!1
<
~
z 6.0
0
5 1!11!1

-2
Cll
<
0
4.0

~
:::::>
~
2.0
t
0

0.0 +--~--+---.---+-----,1"""---f----.---~----,~-~
0.0 2.0 4.0 6.0 8.0 10.0

ESTIMATED GAS INJECfiON RATE ( MMSCF/D)

Fig.3 - Comparison of the Optimum Gas


Injection Rates and Estimated Gas
Injection Rates

Pwh
Gas WELL 1
WELL2
WELL3
WELL4
WELLS
WELL6
Liquid Psep
WELL7
WELL 8
WELL9
WELL 10
Separator WELL 11
WELL 12
WELL 13

Pwh = Psep = Constant for each well

Fig.4 - Illustration of the Gas Lift System

115
SPE 1 9 711'

39ooo~------~--------~--------~1o 0

38000 10- 1

-
~
~
Qo
-
c.o
~
~
'-'
• IIPqgll
-
0
Cl 37000 10- 2

36000 ......."'P"-'~......-....---4~--..___,..___,..--+--.-...........-.........-+- 10 - 3
0 5 10 15
ITER

Fig.S Progress of the Computation


OgTOT = Unlimited

37600 ~-----.,..------,-------...----~ 10 °
37400

10- 1

-~ 37200


Ql
IIPqgll
-c.o
~
~
'-' 37000
-
0
Cl 10- 2

36800

36600 -+-----+----+----+-----+- 10 -. 3
0 2 4 6 8
ITER

Fig.6 - Progress of the Computation


OgTOT = 30 MMSCF/D

116
seE 1 9 71 1
' .
34600 -------.----.........-------...- 10 °

10- 1
34200

-e.
e:Q
El


Qo
IIPqgll
-b.O
z
~
Cl) -
'-"

& 33800 10- 2

33~o+--~~-~-+-~~-+-~-+1o- 3

0 2 4 6 8
ITER

Fig. 7 Progress of the Computation


OgTOT = 10 MMSCF/0

3.0 -r---------r-----r-----,..------.

-@;
{.)
v.l
~ 2.0
~ WELL2
'-"
tlJ
WELL7
~
z WELL 12
0
6
-2
1.0

C l)
<
0

0.0 -t----r----+--.,..----+-----.---;1----.-----1
0 2 4 6 8

ITER
Fig.S - Movement of Gas Injection Rate for
Three Wells
117
SPE 19 711

10° ~------~----~------~--------------

m Gradient

• Quasi-Newton

10· 1 ~----~~-----+------~------+-----~

10· 3 ~--~--~~--~---?--~--~--~--~~
0 4 8 12 16 20

ITER

Fig. 9 - Comparison of the Rate of Convergence

118

Das könnte Ihnen auch gefallen