Académique Documents
Professionnel Documents
Culture Documents
=
=
Q
x
Q
q
q
x
k
k
j
k
l l
n
l
j
c
1
, oil, gas, water;; j n = 1, , , . . . . . (5)
where n
c
is the number of fluid components; q
l
is the mass flow rate
of Component l, which is one of the primary unknowns in network
equation system; x
j
is a decision variable; and n is the number of
decision variables.
Q
q
k
l
is evaluated through flash calculations. To
obtain
q
x
l
j
, we perturb the right-hand side of the connection equa-
tion at the gas lift connection corresponding to x
j
in the linearized
network equation system and solve the system. The matrix of the
linearized equations has already been factored in the full-network
solve, the solution of the perturbed system requires only a for-
ward and backward substitution, which is relatively inexpensive.
However, because a large number of function evaluations may be
needed during optimization iterations and the computational cost
of function evaluations can be very high for a relatively large net-
work, this approach can be prohibitively expensive.
Proxy Functions Based on Partial-Network Solution. The sec-
ond method constructs a set of tabular proxy functions that
approximate the objective function and constraints as functions of
gas lift rates. This approach is based on the assumption that the
gathering nodes where the gas lifted wells or risers join have stable
pressures. Therefore, these wells/risers can be decoupled from each
other at the gathering nodes, and the interaction between wells is
ignored. This assumption is generally reasonable because wells are
normally choked at the wellhead and the ow through the choke
can be critical (at sonic velocity), or multiple wells are gathered
at a separator under a pressure control, or the pressure drop across
the well tubing is dominant in the pipeline network. Numerical
experiments also showed that the gathering-node pressures tend to
be stable after a short period of time (a few timesteps) once the gas
lift injection starts. In addition, the gathering-node pressures get
updated at every global Newton step, so the interactions between
112 February 2012 SPE Reservoir Evaluation & Engineering
wells have been taken into account in an explicit manner. Another
assumption of the method is that each gas lifted well or riser has
only one gas lift injection point, while the method based on full-
network solution does not have this restriction.
For every gas lifted well/riser, a partial network consisting of
this well/riser up to the first gathering node (or the sink node if
the well/riser connects to the sink directly) is set up. The terminal
node at the top of the partial network takes the pressure from the
previous reservoir/network global Newton step as the minimum
pressure constraint. Fig. 2 is an illustration of how a full network
is decoupled into two partial networks at the gathering node (Node
2). Because the flow rates of the partial networks are determined
by the boundary condition at Node 2 (node pressure is P
2
), Node
1 and Conn 1 become irrelevant and are excluded from the partial
networks. For a gas lifted riser, the partial network includes all
wells flowing to the riser. The gas lift rate of a partial network is
a decision variable. The partial-network model is solved over a
range of prespecified gas lift rates to give the tabular sets neces-
sary to furnish the proxy models of the objective and constraint
functions. The proxy value of the objective function of a partial
network can be evaluated using Eq. 4 on the basis of solution of
the partial network at a given gas lift rate. The same approach is
used for the proxy values of constraint functions. Fig. 3 shows a
sample proxy function of the objective function, and Fig. 4 shows
a sample proxy function of an oil-phase rate corresponding to an
oil-phase target, both as functions of a decision variable (gas lift
rate of a gas lifted well). In this example, the well cannot flow until
the gas lift rate is greater than approximately 600 Mscf/D.
The overall objective function (defined in Eq. 4) can be defined
as the sum of the proxy functions of the objective function of all
gas lifted wells:
F x F x
pj j
j
n
( ) ( ) =
=
1
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)
where F
pj
(x
j
) is the proxy function of the objective function of Well
j, which is associated with Decision Variable j.
Optimization constraint function in Eq. 2 becomes
g x g x i m
i pi j
j
n
j
( ) ( ) = =
=
,
, , ,
1
1 , . . . . . . . . . . . . . . . . . . . . . (7)
where g
pi,j
is the proxy function of Well j, corresponding to Opti-
mization Constraint i.
Once the tabular proxy functions for each gas lifted well have
been constructed at selected reservoir/network global Newton
steps, the gas lift optimization can be performed using the values
of the overall objective function and constraint functions calculated
from Eqs. 6 and 7. Either piecewise linear interpolation or piece-
wise polynomial interpolation of the tabular proxy functions can
be used at a given set of decision-variable values (gas lift rates).
The derivatives of these functions with respect to a decision vari-
able are simply the slopes of the proxy functions of this decision
variable, or
= =
F x
x
F x
x
j n
j
pj j
j
( )
, , ,
d ( )
d
1 . . . . . . . . . . . . . . . . . . . . . . (8)
Gas lift optimization with GRG
Rate allocation for ranking-type
targets
Scale back gas lift rates
Reservoir/network global Newton
Next global
Newton step
No
Next timestep
Yes
Converged?
Rate/pressure constraints
at individual connections and
nodes, and scale-type targets.
Satisfy:
Satisfy:
Ranking-type targets.
Once
Start of timestep
Fig. 1Flow chart of gas lift optimization procedure.
TABLE 1TREATMENT OF VARIOUS CONSTRAINTS
Constraints Satisfied At
Rate constraints at individual connections
and pressure constraints at nodes
Potential network solution for function
evaluations
Scale-type targets Optimization constraints
Ranking-type targets Post-optimization scaling back procedure
February 2012 SPE Reservoir Evaluation & Engineering 113
and
= = =
g x
x
g x
x
i m j n
i
j
pi j j
j
( ) d ( )
d
,
, , , ; , , 1 1 . . . . . . . . . . (9)
The method using proxy functions is much faster than the
method using the full-network solution for the following reasons:
First, the number of gas lift rates used to construct proxy functions
is relatively small (approximately 10) compared to the number of
full-network solutions needed in the first method. Second, solving
the partial network is much faster than solving the full network
because the former is a much smaller linearized equation system.
Third, the solving of partial networks can be easily parallelized
because they are decoupled.
Proxy-Function Scaling
When the network configuration and constraints do not change with
time, and the pressure at the decoupling node of the partial network
does not change much, the shapes of proxy functions may change
relatively little. Fig. 5 shows an example of the proxy functions of
the objective function at two timesteps, t1 and t2 with t1 < t2, that
are close in time. In this case, instead of calculating all actual points
of t2, we could first evaluate the function values at three gas lift rates
at time t2 using partial-network solutions and compare them with
the values at time t1. If they change only slightly and the changes
are in the same trend, we could safely estimate the values at other
gas lift rates at time t2 by proportionally shifting the curve at time
t1 to time t2, on the basis of the three points computed at time t2.
In other words, the other points in curve t2 could be calculated
by interpolating the three actual points. This way, we reduce the
number of partial-network solutions required to construct the proxy
functions at time t2. The three checking points can be computed at,
respectively, a gas lift rate close to zero, the gas lift rate that gives
the maximum objective-function value at time t1, and the gas lift
rate at which the slope of objective function at t1 is largest when
the gas lift rate is in logarithmic scale. If the third point is too close
to the first or the second point, we instead compute the point at the
average gas lift rate of the first and the second points.
Sink
Sink Sink
1 k r o w t e N l a i t r a P k r o w t e N l l u F 2 k r o w t e N l a i t r a P
Perf 1 Perf 2 Perf 1 Perf 2
conn 1
conn 2
conn 3 conn 6
conn 4 conn 7
node 1
node 2
node 3 node 6
node 4 node 7
P
min
= 14.7 psi
P P
min
= P
n i m 2
= P
2
conn 2
conn 3
conn 4
conn 2
conn 6
conn 7
node 2 node 2
node 3
node 5
node 6
node 8
Decoupling
Gas lift 1 Gas lift 2 Gas lift 1 Gas lift 2
node 5 node 8
conn 5
node 4 node 7
conn 8 n n o c 5 n n o c 8
Fig. 2Decoupling of a full network into partial networks.
-2.00E+04
0.00E+00
2.00E+04
4.00E+04
6.00E+04
8.00E+04
1.00E+05
1.20E+05
10 100 1000 10000 100000
Gas Lift Rate (Mscf/D)
P
r
o
x
y
F
u
n
c
t
i
o
n
o
f
O
b
j
e
c
t
i
v
e
F
u
n
c
t
i
o
n
(
D
o
l
l
a
r
s
/
d
a
y
)
Fig. 3Proxy function of the objective function associated
with a decision variable.
0.00E+00
5.00E+03
1.00E+04
1.50E+04
2.00E+04
2.50E+04
3.00E+04
10 100 1000 10000 100000
Gas Lift Rate (Mscf/D)
P
r
o
x
y
F
u
n
c
t
i
o
n
o
f
O
i
l
-
P
h
a
s
e
R
a
t
e
C
o
n
s
t
r
a
i
n
t
(
S
T
B
/
D
)
Fig. 4Proxy function of a constraint function associated with
a decision variable.
114 February 2012 SPE Reservoir Evaluation & Engineering
Gas Lift Rate Reduction
As mentioned in Step 3 of the gas lift optimization procedure,
certain connection rates may need to be reduced to satisfy rank-
ing-type targets. A target allocation procedure determines if there
are any ranking-type targets getting violated after the gas lift
optimization. If any such target is violated, the flow rates of the
controlled connections of the target are reduced on the basis of
the ranking method (e.g., the flow rate of the connection with the
highest water cut will be reduced to meet the target if the ranking
method is water-cut based). The reduction factor is calculated and
is propagated to all downstream and upstream connections. The
gas lift rate can be reduced according to how much the flow rate
of the associated well tubing connection has been reduced. When
the proxy function method is used, we can interpolate a phase-flow
rate (e.g., oil-phase-flow rate) in the corresponding proxy function
to obtain the reduced gas lift rate.
When the full-network solution method is used, we may simply
reduce the gas lift rate using the same reduction factor as is being
applied to the flow rate at the gas lifted well tubing connection. This
approach assumes a linear relationship between the well-tubing-
connection flow rate and the gas lift flow rate, and this assumption
can be inaccurate. To improve the solution, we construct a local
proxy function (a phase-flow rate vs. gas lift rate, such as in Fig. 4)
for the gas lifted well through a partial-network solution using the
latest gathering-node pressure as the minimum pressure constraint
at the terminal node of the partial network. The proxy function can
then be used to estimate the reduced gas lift rate. Note that in this
approach, the proxy function is used just for gas lift rate reduction
estimates. It is not involved in gas lift optimization calculations.
Examples and Comparisons
For convenience, the simulator in which the proposed gas lift
optimization methods are implemented is referred to as Simula-
tor N. We compare our implementation with another commercial
reservoir simulator, Simulator V (Litvak and Darlow 1995), in one
of our examples. A scheme of automatic allocation of gas lift using
a performance curve has been implemented in Simulator V. In this
scheme, nodal analysis is used to construct the gas lift performance
curve (produced-oil-phase rate vs. gas lift rate) at selected timesteps
for each gas lifted well, assuming fixed minimum tubinghead pres-
sures. The gas lift rate is calculated (it is basically a table-lookup
procedure) on the basis of the minimum lift efficiencies of the
wells specified by the user, where the lift efficiency is defined as
the incremental oil produced per increment of gas lift gas, which is
effectively the slope of the gas lift performance curve. In Simulator
V, the reservoir-grid domain and the surface-network domain have
partial implicit coupling because they are solved with an alternative
Schwarz procedure; that is, at selected timesteps, the network is
solved with fixed reservoir grid conditions, then the reservoir grid
is solved with updated well bottomhole or well tubinghead condi-
tions provided by the latest network solution. These network and
reservoir grid solutions are repeated iteratively until the boundary
values (well flow rates and bottomhole pressures) between them
are converged. For comparison, we refer to the automatic gas lift
allocation method in Simulator V as the fixed-slope scheme,
the method using proxy functions in our implementation as the
proxy scheme, and the method using full-network solution in
our implementation as the full-network scheme.
Case 1. Case 1 is a black-oil model simulated for 900 days. It has
four gas lifted production wells with xed minimum tubinghead
pressures, one gas-injection well, and one water-injection well. All
wells are connected directly to a sink or a source. To match the
user-specied minimum lift efciency (0.01 STB/Mscf) used by the
xed-slope scheme, we set R
oil
= USD 30/STB, R
gas
= 0, R
wat
=
0, and R
glift
= USD 0.3/Mscf in the proxy and the full-network
schemes so that R
glift
/R
oil
= 0.01 STB/Mscf. There is an 8,000 STB/D
maximum oil-phase rate constraint at every well but they are inac-
tive during the simulation. The maximum gas lift rate allowed for
each well is 10,000 Mscf/D. Fig. 6 compares the oil-production
40000
50000
60000
70000
80000
90000
100000
110000
0.1 10 1000 100000 10000000
Gas Lift Rate (Mscf/D)
B
e
n
e
f
i
t
F
u
n
c
t
i
o
n
(
D
o
l
l
a
r
s
/
d
a
y
)
t1
t2
Fig. 5Proxy functions at two timesteps.
0 100 200 300 400 500 600 700 800 900 1000
Time (days)
0.2
0.4
0.6
0.8
1.0
N
o
r
m
a
l
i
z
e
d
Q
O
P
O
i
l
-
P
r
o
d
u
c
t
i
o
n
R
a
t
e
0
0.2
0.4
0.6
0.8
1.0
N
o
r
m
a
l
i
z
e
d
Q
G
L
G
G
a
s
L
i
f
t
I
n
j
e
c
t
i
o
n
R
a
t
e
QOP : Fixed-slope QOP : Full-network QOP : Proxy
QGLG : Fixed-slope QGLG : Full-network QGLG : Proxy
Fig. 6Oil-production rates and gas lift rates of Case 1.
February 2012 SPE Reservoir Evaluation & Engineering 115
rate and gas lift injection rate of the simulation results. Before 130
days, there is an active water-production target (8,000 STB/D) that
prohibits any gas lift injection. Afterward, as the reservoir depletes,
Simulators V and N manage to inject gas lift to improve oil produc-
tion without violating the water-production target. The total lift gas
available is 10,000 Mscf/D. The proxy scheme predicts a higher
oil-production rate after approximately 200 days by injecting more
lift gas. Fig. 7 shows that the proxy scheme also manages to reduce
water production compared to the xed-slope scheme. A closer
investigation of the results shows that the proxy scheme injects
more lift gas in three production wells with high oil-production
rates and low water cuts as well as less lift gas in a well with low
oil-production rate and high water cut compared to the xed-slope
scheme. This is probably because the proxy scheme performs
an optimization, and handles constraints more strictly than the
xed-slope method. Fig. 8 shows that the proxy scheme increases
the cumulative benet by approximately 3% over the xed-slope
scheme. The solution of the full-network scheme is virtually the
same as that of the proxy scheme in Figs. 6 and 7, which provides
a conrmation of the proxy-function implementation.
The simulation CPU times (in a Windows XP machine with
Intel Xeon 2.33 GHz CPU and 3 GB of RAM) using the fixed-
slope, proxy, and full-network schemes are 119, 124, and 316
seconds, respectively. As a result, the simulation speed of the proxy
scheme in Simulator N is comparable with the fixed-slope scheme
in Simulator V and is approximately 2.5 times as fast as the full-
network scheme. This example demonstrates that the optimization
method using proxy functions is much more efficient than the
method using the full-network solution and can achieve a more
optimal solution than the fixed lift efficiency scheme.
Case 2. Case 2 is used to investigate the effects of the interactions
between wells. The properties of the reservoir domain are based on
the SPE9 test case (Killough 1995), but the values of solution-gas/oil
ratio (R
s
) have been reduced by approximately three times (Table 2)
to make the effect of gas lift more obvious. It is a black-oil model
with 30 production wells, 20 of which are gas lifted starting at time
of 50 days. There are no injection wells. All the wells are connected
through a network that has three levels of gathering centers. The
highest level connects to a sink (Fig. 9). All network constraints are
listed in Table 3. We set R
oil
= USD 30/STB, R
gas
= USD 3/Mscf,
R
wat
= USD 2/STB, and R
glift
= USD 4.5/Mscf. Most of the oil-rate
constraints at wells are inactive during the simulation, and there is
no active pressure constraint except for a minimum pressure con-
straint (100 psia) at the sink node. Therefore, the well ow rates are
mostly determined by the hydraulics in the tubings and owlines,
and the interactions between wells are strong, especially when the
0 100 200 300 400 500 600 700 800 900 1000
Time (days)
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
N
o
r
m
a
l
i
z
e
d
Q
W
P
W
a
t
e
r
-
P
r
o
d
u
c
t
i
o
n
R
a
t
e
QWP : Fixed-slope QWP : Full-network QWP : Proxy
Fig. 7Water-production rates of Case 1.
0
0.2
0.4
0.6
0.8
1
1.2
0 200 400 600 800 1000
Time (days)
N
o
r
m
a
l
i
z
e
d
C
u
m
u
l
a
t
i
v
e
B
e
n
e
f
i
tFixed-slope
Proxy
Fig. 8Cumulative benefit of Case 1.
TABLE 2SOLUTION-GAS/OIL RATIO IN CASE 2
Pressure (psia) Rs (Mscf/STB)
0 0 0 5 . 0 0 . 0 0 0 4
2 3 6 4 . 0 0 . 0 0 6 3
1 3 2 4 . 0 0 . 0 0 2 3
4 6 7 3 . 0 0 . 0 0 8 2
9 7 2 3 . 0 0 . 0 0 4 2
5 5 7 2 . 0 0 . 0 0 0 2
2 1 2 2 . 0 0 . 0 0 6 1
1 6 6 1 . 0 0 . 0 0 2 1
9 0 1 1 . 0 0 . 0 0 8
9 3 5 0 . 0 0 . 0 0 4
0 . 0 7 . 4 1
116 February 2012 SPE Reservoir Evaluation & Engineering
gas lift rst starts. The pressure drops after the well tubing heads are
approximately at 400 psi. Fig. 10 shows that the oil-phase-produc-
tion rates of the eld predicted by full-network and proxy methods
are noticeably different at the early stage of the simulation after gas
lift starts. We now apply these two methods in sequence: Before
time = 60 days, the full-network method is used to handle the strong
coupling of wells shortly after the gas lift starts; after that time, the
proxy method is used for faster simulation. We refer to this scheme
as the mixed method. The switch time is chosen on the basis of the
observation that the ow is about to be stable using the full-network
Sink
Pmin = 100 psia
Gas lift
Well-17
Fig. 9Surface-network structure of Case 2. The arrows at the bottom of the plot denote gas lift injection to the bottomhole
nodes of the wells.
TABLE 3NETWORK CONSTRAINTS IN CASE 2
Maximum oil-phase rate constraint at each well
15,000 STB/D, time < 50 days
5,000 STB/D, time 50 days
Maximum gas lift rate constraint at each well
0 Mscf/D, time < 50 days
20,000 Mscf/D, time 50 days
Mscf/D 0 0 0 , 0 2 t f i l s a g l a t o t f o e t a r t e g r a T
a i s p 0 0 1 e d o n k n i s t a t n i a r t s n o c e r u s s e r p m u m i n i M
0 50 100 150 200 250
Time (days)
20000
30000
40000
50000
60000
Q
O
P
O
i
l
-
P
r
o
d
u
c
t
i
o
n
R
a
t
e
(
S
T
B
/
D
)
0
5000
10000
15000
20000
25000
Q
G
L
G
G
a
s
L
i
f
t
I
n
j
e
c
t
i
o
n
R
a
t
e
(
M
s
c
f
/
D
)
QOP : Full-network QOP : Proxy QOP : Mixed
QGLG : Full-network QGLG : Proxy QGLG : Mixed
Fig. 10Case 2: Effects of interactions between wells.
February 2012 SPE Reservoir Evaluation & Engineering 117
method approximately 10 days after the gas lift injection starts.
A general criterion to determine when the switch should happen
is still under investigation. Fig. 10 shows that the mixed method
achieved production rates very close to the full-network method. The
total gas lift rates computed by the three methods are overlapped.
Fig. 11 shows the oil-production rates and gas lift rates of Well-17.
For a short period of time after the gas lift starts, the rates computed
by the proxy method show some oscillations because unstable well-
head pressures, as shown in Fig. 11. The mixed method achieves
results very close to those of the full-network method. The CPU
time of the network calculations of the three methods (full-network,
proxy, and mixed) are 126, 50, and 71 seconds, respectively.
Case 3. Case 3 is a multiple-reservoir compositional model. More
than 10 reservoirs are connected through a common surface net-
work. The uids of these reservoirs have up to 19 components.
There are 100 production wells and a number of gas- and water-
injection wells. 60 of the production wells are assisted with gas
lift. The network consists of approximately 500 connections, with
a large number of rate and pressure constraints, and approximately
50 production and injection targets of either scale type or ranking
type. There are multiple levels of gathering centers and separators
in the network. The network eventually ows to a sink with a
minimum pressure constraint of 14.7 psia. Complicated procedures
are dened to dynamically change the constraints and network
congurations during the simulation. We set R
oil
= USD 60/STB,
R
gas
= 0, R
wat
= 0, and R
glift
= USD 0.6 /Mscf in this case. The oil-
phase-production rates computed by the full-network method agree
well with those computed by the proxy method, which is shown in
Fig. 12. The gas lift rates of the two methods follow approximately
the same trend, but the former gives higher values than the later, as
shown in Fig. 12. The case was run in parallel on four processors.
0 50 100 150 200 250
Time (days)
0
1000
2000
3000
4000
5000
6000
0
500
1000
1500
2000
2500
3000
QOP W17 : Full-network QOP W17 : Proxy
QOP W17 : Mixed QGLG W17 : Full-network
QGLG W17 : Proxy QGLG W17 : Mixed
Q
O
P
O
i
l
-
P
r
o
d
u
c
t
i
o
n
R
a
t
e
(
S
T
B
/
D
)
Q
G
L
G
G
a
s
L
i
f
t
I
n
j
e
c
t
i
o
n
R
a
t
e
(
M
s
c
f
/
D
)
Fig. 11Oil-production rates and gas lift rates of Well-17 in Case 2.
0 2000 4000 6000 8000 10000 12000 14000
Time (days)
0
0.2
0.4
0.6
0.8
1.0
1.2
N
o
r
m
a
l
i
z
e
d
Q
O
P
O
i
l
-
P
r
o
d
u
c
t
i
o
n
R
a
t
e
0
0.2
0.4
0.6
0.8
1.0
1.2
N
o
r
m
a
l
i
z
e
d
Q
G
L
G
G
a
s
L
i
f
t
I
n
j
e
c
t
i
o
n
R
a
t
e
QOP : Full-network QOP : Proxy
QGLG : Full-network QGLG : Proxy
Fig. 12Oil-production rates and gas lift rates of Case 3.
118 February 2012 SPE Reservoir Evaluation & Engineering
The proxy method is much faster than the full-network method.
The network solution CPU times of the two methods are 4.4 hours
and 9.9 hours, respectively.
From our general observations of a number of test cases (not all
shown in this paper), we have found that although the total produc-
tion rates predicted by the different optimization methods normally
match well, the gas lift rate allocations to individual wells can be
quite different because the optimization can be trapped by different
local optimal points. In addition, the network solution is typically
sensitive by nature; a small difference in the solutions at a timestep
may create very different flow patterns (different wells flow or are
shut-in) at a later timestep. The sensitivity of solutions can be allevi-
ated by adding rate or pressure constraints at individual wells.
Conclusions
1. We have implemented an efficient gas lift optimization scheme
in a general-purpose black-oil and compositional reservoir
simulator with fully coupled reservoir grid domain and sur-
face-network domain. The nonlinearly constrained optimization
problem is solved by the GRG method.
2. Network constraints are satisfied strategically in the optimiza-
tion scheme: All surface-network constraints are satisfied at
individual connections or nodes; all scale-type targets are satis-
fied during optimization; and all ranking-type targets are satis-
fied through a gas lift reduction procedure after optimization.
3. The function evaluations for optimization can be performed
by two methods. The first method repeatedly solves the full-
network equations using Newton iteration. The second method
constructs a set of proxy functions using partial-network solu-
tions.
4. The full-network method is more rigorous because it takes into
account the flow interactions among wells; the proxy-function
method neglects these interactions, assuming a fixed gather-
ing-node pressure, so the solution can be suboptimal when the
interactions are strong, especially during a short period of time
after the gas lift first starts. However, the proxy-function method
has the following advantages over the full-network method:
a. The computation speed of the proxy-function method is sig-
nificantly faster because once the proxy functions have been
constructed, they can be used to inexpensively approximate the
objective function and constraints as functions of gas lift rates
during optimization iterations. On the contrary, the full-network
method has to resolve the whole network at every optimization
iteration, which can be very expensive.
b. The number of partial-network solutions for proxy-function
construction can be reduced through a simple interpolation
scheme if the shapes of the proxy functions do not change much
between two optimizations.
c. The calculations of partial networks in proxy-function construc-
tion have been parallelized for faster simulation.
5. According to the test cases presented in the paper, the production
values predicted by the two methods are in good agreement. We
also observe that while the total gas lift rates given by the two
methods generally follow the same trend, the gas lift allocations
at individual wells can be different because optimization can be
trapped by local optimal points using derivative-based optimizer,
and solutions of the surface network are sensitive by nature. The
two methods can be used in tandem during the simulation.
Nomenclature
F = overall objective function, USD/D
F
pj
= proxy function of objective function of decision variable
j, USD/D
g
i
= constraint function i
g
pi,j
= proxy function of constraint function i of decision vari-
able j
H
j
= the upper bound of decision variable j
i = subscript denoting constraint function
j = subscript denoting decision variable
k = subscript denoting uid phase
L
i
= the lower bound of constraint function i
m = number of constraint functions
n = number of decision variables
n
c
= number of uid components
p = node pressure, psi
p
min
= minimum allowed pressure, psi
q
l
= mass ow rate of component l
Q
k
= total volumetric ow rate of phase k
Q
oil
= total volumetric ow rate of oil-phase production, STB/D
QOP = volumetric ow rate of oil-phase production, STB/D
Q
gas
= total volumetric ow rate of gas-phase production, Mscf/D
Q
glift
= total volumetric ow rate of lift gas, Mscf/D
Q
water
= total volumetric ow rate of water-phase production, STB/D
QWP = volumetric ow rate of water-phase production, STB/D
QGLG = volumetric ow rate of lift gas, Mscf/D
R
n
= n-dimensional real coordinate space
R
oil
= unit value of oil phase, USD/STB
R
gas
= unit value of gas phase, USD/Mscf
R
glift
= unit cost of gas lift operation, USD/Mscf
R
wat
= unit cost of water-phase treatment, USD/STB
S
j
= the lower bound of decision variable j
U
i
= the upper bound of constraint function i
x = vector of decision variables
x
j
= decision variable j
Acknowledgments
Bill Watts provided helpful comments during revision of this paper.
The authors wish to thank Halliburton for permission to present
this paper.
References
Coats, B.K., Fleming, G.C., Watts, J.W., Rame, M., and Shiralkar, G.S.
2003. A Generalized Wellbore and Surface Facility Model, Fully
Coupled to a Reservoir Simulator. Paper SPE 79704 presented at the
SPE Reservoir Simulation Symposium, Houston, 35 February. http://
dx.doi.org/10.2118/79704-MS.
Dutta-Roy, K. and Kattapuram, J. 1997. A New Approach to Gas-Lift
Allocation Optimization. Paper SPE 38333 presented at the SPE
Western Regional Meeting, Long Beach, California, USA, 2527 June.
http://dx.doi.org/10.2118/38333-MS.
Fang, W.Y. and Lo, K.K. 1996. A Generalized Well-Management Scheme
for Reservoir Simulation. SPE Res Eng 11 (2): 116120. SPE-29124-
PA. http://dx.doi.org/10.2118/29124-PA.
Fletcher, R. 1987. Practical Methods of Optimization, second edition,
317322. New York: John Wiley & Sons.
Gutierrez, F.A., Hallquist, A.E., Shippen, M.E., and Rashid, K. 2007.
A New Approach to Gas Lift Optimization Using an Integrated
Asset Model. Paper IPTC 11594 presented at the International Petro-
leum Technology Conference, Dubai, 46 December. http://dx.doi.
org/10.2523/11594-MS.
Hepguler, G., Barua, S., and Bard, W. 1997. Integration of a Field Surface
and Production Network With a Reservoir Simulator. SPE Comp App
9 (3): 8892. SPE-38937-PA. http://dx.doi.org/10.2118/38937-PA.
Kanu, E.P., Mach, J., and Brown, K.E. 1981. Economic Approach to Oil
Production and Gas Allocation in Continuous Gas Lift. J Pet Technol
33 (10): 18871892. SPE-9084-PA. http://dx.doi.org/10.2118/9084-PA.
Killough, J.E. 1995. Ninth SPE Comparative Solution Project: A Reexami-
nation of Black-Oil Simulation. Paper SPE 29110 presented at the SPE
Reservoir Simulation Symposium, San Antonio, Texas, USA, 1215
February. http://dx.doi.org/10.2118/29110-MS.
Lasdon, L.S. and Waren, A.D. 1997. GRG2 Users Guide. Technical docu-
ment, University of Texas at Austin, Austin, Texas (2 October 1997).
Litvak, M.L. and Darlow, B.L. 1995. Surface Network and Well Tubinghead
Pressure Constraints in Compositional Simulation. Paper SPE 29125
presented at the SPE Reservoir Simulation Symposium, San Antonio,
Texas, USA, 1215 February. http://dx.doi.org/10.2118/29125-MS.
Rashid, K. 2010. Optimal Allocation Procedure for Gas-Lift Optimization.
Ind. Eng. Chem. Res. 49 (5): 22862294. http://dx.doi.org/10.1021/
ie900867r.
February 2012 SPE Reservoir Evaluation & Engineering 119
Rashid, K., Demirel, S., and Couet, B. 2011. Gas-Lift Optimization with
Choke Control using a Mixed-Integer Nonlinear Formulation. Ind. Eng.
Chem. Res. 50 (5): 29712980. http://dx.doi.org/10.1021/ie101205x.
Shiralkar, G.S. and Watts, J.W. 2005. An Efficient Formulation for Simultaneous
Solution of the Surface Network Equations. Paper SPE 93073 presented at
the SPE Reservoir Simulation Symposium, The Woodlands, Texas, USA,
31 January2 Feburary. http://dx.doi.org/10.2118/93073-MS.
Stoisits, R.F., Crawford, K.D., MacAllister, D.J., McCormack, M.D.,
Lawal, A.S., and Ogbe, D.O. 1999. Production Optimization at the
Kuparuk River Field Utilizing Neural Networks and Genetic Algo-
rithms. Paper SPE 52177 presented at the SPE Mid-Continent Opera-
tions Symposium, Oklahoma City, Oklahoma, USA, 2831 March.
http://dx.doi.org/10.2118/52177-MS.
Wang, P. and Litvak, M. 2008. Gas Lift Optimization for Long-Term
Reservoir Simulations. SPE Res Eval & Eng 11 (1): 147153. SPE-
90506-PA. http://dx.doi.org/10.2118/90506-PA.
Watts, J.W., Fleming, G.C., and Lu, Q. 2009. Determination of Active
Constraints in a Network. Paper SPE 118877 presented at the SPE
Reservoir Simulation Symposium, The Woodlands, Texas, USA, 24
February. http://dx.doi.org/10.2118/118877-MS.
Qin Lu is a technical advisor for Landmark Software and
Services at Halliburton where he works on the development of
Nexus reservoir simulator. He holds a BS and a MS degrees, both
in mechanical engineering, from the University of Science and
Technology of China, respectively, and a PhD degree in petro-
leum engineering from the University of Texas at Austin.
Graham Fleming is a chief technical advisor for Landmark
Software and Services at Halliburton where he is the devel-
opment lead for the Nexus reservoir simulator. He has a PhD
degree from the California Institute of Technology and a BE
degree from the University of Auckland.