1 GMC 2004

Compressor Station Optimization Using

Simulation-Based Optimization


Kirby S. Chapman Mohammad Abbaspour Prakash Kirishnaswami
Professor and Director Associate Researcher Professor
[email protected] [email protected] [email protected]

National Gas Machinery Laboratory, Kansas State University
245 Levee Dr., Manhattan, KS 66502
Phone: 785-532-2319, Fax: 785-532-3744


ABSTRACT
Mathematical modeling is one of the most cost-effective tools
that can be used to aid in design, operation, and optimization
studies. The systems under consideration actually operate in an
unsteady nature, and although much effort has been and
continues to be spent on unsteady mathematical models, many
over-simplifications are introduced that bring into question the
simulation results.
One of the primary concerns in the operation of a compressor
station is minimization of fuel consumption while maintaining
the desired throughput of natural gas. In practice, the station
operator tries to achieve this by shutting down units or
controlling individual unit speeds based on experience. This is
generally a trial-and-error process without any guarantee of
optimality. In this paper we present a robust structured solution
process for tackling this problem using simulation-based
optimization.
The problem of optimizing the operation of a compressor
station is formulated as a nonlinear programming problem
(NLP) in which the design variables are the compressor unit
speeds and the objective function to be minimized is the fuel
consumption. A constraint was also placed on the minimum
mass flow rate through the station to ensure that adequate flow
is maintained while minimizing fuel consumption. This NLP
was then solved using a sequential unconstrained minimization
technique (SUMT) with a derivative-free grid search for
handling the unconstrained minimizations. The simulation
algorithm mentioned earlier is invoked whenever the
optimization needs to evaluate the system response at a
candidate operating point.
Several representative numerical examples have been solved
by the proposed approach. The results obtained indicate that the
method is very effective in finding operating points that are
optimal with respect to fuel consumption. The optimization can
be done at the level of a single unit, a single compressor station,
a set of compressor stations, or an entire network. It should also
be noted that the proposed solution approach is fully automated
and requires no user involvement in the solution process.
This work is a part of the DOE project Virtual Pipeline
System Testbed to Optimize the U.S. Natural Gas Transmission
System.

INTRODUCTION
Natural gas enters the pipeline from a supply source, and then
is transported to one or more delivery points. One of the most
important collections of components in this system is the
compressor station located about every 60 miles along the
pipeline. The compressor station overcomes the gas pressure
drop in the pipe. Consequently, detailed mathematical modeling
of compressor stations is critical for optimizing and
understanding the ability of the gas pipeline system to deliver
natural gas to the end-user.
Several investigators tried to simulate unsteady conditions
for pipeline systems and some of them focused in compressor
station modeling. Botros et al. [1,2] and Botros [3] presented a
dynamic compressor station simulation that consisted of
nonlinear partial differential equations describing the pipe flow
together with nonlinear algebraic equations describing the
quasi-steady flow through various valves, constrictions, and
compressors. This model included mathematical descriptions of
the control system, which consists of mixed algebraic and
ordinary differential equations with some controller limits.
Bryant [4] modeled compressor station control, which had
some advantages such as the ability to set individual unit swing
priority, the ability to try and meet multiple setpoints and the
ability for units to automatically come on-line and off-line. The
model used automatic linepack tuning instead of automatic
pipeline roughness tuning.
Stanley and Bohannan [5] discussed the application of
dynamic simulation to centrifugal compressor control system
2 GMC 2004
design. The simulation studies resulted in design
recommendations concerning the number and location of
recycles required, sizing of recycle control valves, and setpoint,
gain, and reset settings for control system instrumentation. This
paper solves equations in ordinary form for compressors
without considering pipe equations into the compressor station.
Turner and Simonson [6,7] developed a computer program
for compressor stations that is added to SIROGAS, which is a
program that solves a pipeline network for steady state and
transient mode.
Schultz [8] derived the real-gas equations of polytropic
analysis to show their application to centrifugal compressor
testing and design. Odom [9]

reviewed the theory of centrifugal
compressor performance, and also presented a set of
polynomial equations for centrifugal compressor maps, which
used constant coefficients for these equations for different
compressors.
Carter [10] presented a hybrid mixed-integer-nonlinear
programming method that is capable of efficiently computing
exact solutions to a restricted class of compressor models, and
attempted to place station optimization in context with regard to
simulation.
Letniowski [11] presented an overview of the design process
for a compressor station model that is part of a network model.
Jenicek and Kralik [12] developed optimized control of a
generalized compressor station. The model described an
algorithm for optimizing the operation of the compressor
station with fixed configuration.
Botros [13] presented a numerical study of gas recycling
during surge control, and furnished a basic understanding of the
thermodynamic point of view and showed the variation of gas
pressure, temperature and flow.
Boyd et al. [14] considered the fuel cost minimization
problem in the steady-state gas pipeline networks by using
mathematical model over compressor station.
Carter [15] developed a nonsequential Dynamic
Programming (DP) algorithm to handle looped networks when
the mass flow rate variables are fixed. The main advantages of
DP are that a global optimum is guaranteed to be found and that
nonlinearity can be easily handled.
Wu et al. [16] presented a two-model relaxation, one in the
compressor domain and another in the fuel cost function, and
derived a lower bounding scheme. The empirical evidence has
been presented that showed the effectiveness of the lower
bounding scheme.
Cobos-Zaleta and Rios-Mercado [17] used a MINLP model
for the problem of minimizing the fuel consumption in a
pipeline network. A computational experience was presented by
evaluating an outer approximation with equality relaxation and
augmented penalty method.
Siregar et al. [18] developed a mathematical model, which in
turn solved analytically and numerically the optimum pipeline
diameter and routing.
Edgar et al. [19] presented a computer simulation to optimize
the design of a gas transmission network, which considered the
number of compressor stations, the diameter and length of
pipeline segments, and the operating conditions of each
compressor station. Two solution methods were used.
Osiadacz [20] described the algorithm for optimal control of
a gas network based upon hierarchical control and
decomposition of the network. This work is concerned with the
minimization of operating costs for high-pressure gas networks
under transient conditions.
The work presented in this paper is an important advance
over current methods in the accurate simulation of transient
non-isothermal behavior in natural gas pipelines, and extends
the knowledge found in the literature by demonstrating the
impact of varying boundary conditions on compressor station
components. In addition, it also shows how this type of detailed
simulation can be used for optimizing the operation of a
compressor station to minimize fuel consumption while
maintaining desired throughput.


Nomenclature

A Cross- sectional area of pipe (m
2
)
1
b -
6
b Coefficients for centrifugal compressor map (-)

p
C Specific heat at constant pressure (J/kg. K)
D Pipe diameter (m)
f Friction factor (-)
g Gravitational acceleration (m/s
2
)
h Specific enthalpy (J/kg)
Head Isentropic head (kJ/kg)
LHV Low heating value (kJ/kg)
m& Mass flow rate (kg/s)
N Number of node (-)
r
N
Speed (rpm)
P Pressure of the gas (Pa)
Q Capacity (m
3
/hr)
R Specific gas constant (kJ/kg)
t Time (s)
T Temperature (K)
v Velocity of the gas directed along the axis of the
pipe (m/s)
w
V
Isentropic wave speed (m/s)
W Frictional force per unit length of pipe and per
unit time (N/m)
x Distance along the pipe (m)
Z Compressibility factor (-)
Efficiency (-)
g
Specific gravity (-)
Angle of inclination of pipe to the horizontal
(radian)
Isentropic exponent (-)
Density of the gas (kg/m
3
)
Heat flow (J/ms)

Subscripts
ac Actual condition
d Discharge
dr Driver
f Fuel
is Isentropic
mech Mechanical
3 GMC 2004
s Suction
Sc Standard condition

Governing equations

The non-isothermal compressible flow of natural gas in a
compressor station is governed by the time-dependent
continuity, momentum, and energy equations, and an equation
of state for homogeneous, geometrically one-dimensional pipe
flow. The compressors within the compressor station are
modeled using centrifugal compressor map-based polynomial
equations.

Pipe Equations
Chapman and Abbaspour [21,22,23]

developed the basic
equations for one-dimensional, unsteady, compressible flow
that include the effects of wall friction and heat transfer. The
continuity, momentum, and energy equations as functions of
pressure, temperature and mass flow rate are written as:

2
1 1
w
T
V P ZRT P Z P
t PA ZRT Z P x
m
1
_ _
+
1

, ,
]
&

2
1 1
1
w
P
V m T Z T
A x T Z T x
m
m
_ _ _
+ + +


, , ,
& &
&


2
2
1
w
P
V T Z ZRT
w
CpT Z T A PA
m

_ _
' ;
, ,
)

+ +

&
(1)

2
1
T
ZRT ZRT P Z
P
AP t PA x Z P
m m m
m
_ _ _
+
'

, , ,
& & &
&

1
P
P ZRT P P T Z
t PA x T Z T
m m _ _ _
+ + +

, ,
,
& &

ZRT P
P x
T ZRT T
t PA x
m
+

_
+
;

,)
&

sin
wZRT
g
PA
(2)
2
2
1
w
P
V T T Z ZRT
P
t Cp Z T AP x
m _ _
+ +



,
,
&

1
T
P Z P ZRT
Z P x PA
m
m
_ _ 1
+

1
, ]
,
&
&

2
2
1 1
w
P
V T Z T
CpT Z T x
_
_ _
+ +


,
,
,

2
1
w
T
V P Z
CpP Z P
_

' ;

,
)
2
mZRT
w
A PA
_
+

,
&
(3)

The wave speed
w
V is:

2
1 1
w
T P
ZRT
V
P Z P T Z
Z P CpT Z T

1 _ _
+
' ; 1
, ,
]
)
(4)

The parameter Z is the compressibility factor (Dranchuck

[24]):

2 3 3 5 6 2
1 4 3 3
1
A A A A
Z A r A r r
Tr Tr Tr Tr

_ _
+ + + + + +

, ,
(5)


To simulate the compressor station, the following
equations are used to describe the performance of a centrifugal
compressor. Compressor head is determined by:

0.28704 1
s s d
g s
T Z P
Head
P

_
_



,
,
(6)


and the relationship between the flow rate for standard
conditions and the actual mass flow rate is :


3
97.67 10
ac
sc
m R
Q

&
(7)


The power required by the compressor for these conditions is:


mech is
ac
Head m
Power

&
(8)

For the purpose of inputting centrifugal compressor
characteristics into a pipeline simulation model, it is suggested
that the entire head versus capacity map be digitized and stored
as a table. However, a simplified but still accurate
representation of the head versus capacity curve can be
obtained through the use of the normalized characteristics.
Figure 1 shows a sample compressor map.



Fig.1 Compressor Map [9]

4 GMC 2004

Three normalized parameters are necessary to describe a
compressor map,
2
r
Head N
,
ac r
Q N
,
is
. Using standard
polynomial curve-fit procedures for each centrifugal
compressor, the relationship between these parameters is:

2
2 1 2 3
ac ac
r r r
Q Q Head
b b b
N N N
+ +
_ _

, ,
(9)

and:
2
5 4 6
r r
ac ac
is
Q Q
b b b
N N
+
_ _
+

, ,
(10)

The coefficients
1
b ,
2
b ,
3
b ,
4
b ,
5
b ,
6
b that make Eqs. 9 and 10
fully characterize the specific centrifugal compressor map.
With the coefficients for Eqs. 9 and 10 stored in the computer,
knowing the isentropic head and inlet volumetric flow allow
computation of compressor speed and isentropic efficiency.
The fuel consumption for the compressor driver is currently
obtained by:
dr
f
Power
LHV
m

& (11)

The gas discharge temperature is obtained by:

1
100
s d
s d
s is
P T
T T
P

1
_
1 +

1 ,
]
(12)

and the mass balance for suction and discharge of the
compressor is:

, , ac s a c d
f
m m m + & & & (13)

The fully implicit method consists of transforming Eqs. 1, 2,
and 3 from partial differential equations to algebraic equations
by using finite difference approximations for the partial
derivatives. These equations are nonlinear and the Newton-
Raphson method is applied to solve these equations for the
compressible, non-isothermal transient flows through a pipe.
Quasi-steady flow can be assumed at each time step of the
numerical solution for the centrifugal and reciprocating
compressor equations.

Formulation of the Optimization Problem
In order to optimize the operation of a pipeline network, we
first formulate the problem at hand in the format of a standard
nonlinear programming problem (NLP). This standard form is
developed as:
Find the values of the design variables:

[b
1
,b
2
,.,b
r
]
T

to minimize an objective function:
f(b)

Subject to the constraints:
h
j
(b) = 0 , j = 1,, m
and g
j
(b) 0 , j = m+1,,n

The formulation of the network operation problem in the
standard NLP form must be done carefully, making sure that the
NLP formulation captures all the relevant aspects of the
associated network problem.

Let have the following assumptions:
N Number of compressor stations in the pipeline network
NC
j
Number of compressors in station j
n
ik
Speed of compressor k in station i.
nmin
ik
Minimum speed of compressor k in station i
nmax
ik
Maximum speed of compressor k in station i
m
fi
Fuel consumption rate of station i
m
i
Mass flow rate at station i and and let the specified
mmin
i
Minimum allowable mass flow rate at station

Then, the set of design variables is defined by
{n
ik
}, i = 1,,N; k = 1, NC
i

while the objective function is given by
f = (m
fi
) , i = 1,,N
and the constraints are
nmin
ik
n
ik
nmax
ik
, i = 1,,N; k = 1, NC
i

mmin m
i
, i = 1,,N.
Solution of the Optimization Problem
Once the network operation problem has been formulated as
an optimization problem as outlined above, it can be solved
using any of a variety of available methods. In this work, we
used the sequential unconstrained minimization technique
(SUMT) with an exterior penalty function. A directed grid
search method was used for the unconstrained minimization
that is required by the SUMT approach.
RESULT AND DISCUSSION
In this section we consider two parts, first a baseline
simulation and second optimization.

Simulation
Figure 2 shows the schematic of a compressor station,
including the boundary conditions and geometry of the
compressor station. Note that the numbers of compressors and
boundary conditions could be different for different stations.

A
C
1 2 3 4
D= 23.62 in / 0.6 m
L=62.14 mile / 100 km
D= 11.81 in / 0.3 m
L=328 ft/ 100 m
P
1
=896.91Psi / 6.184 Mpa
T
1
=599.67 R / 333.18 K
P
4
= 740 Psi / 5.11 Mpa
B
D= 23.62 in / 0.6 m
L=62.14 mile / 100 km

Fig. 2 Schematic of compressor station
5 GMC 2004
The compressor station is located between two long pipes
(100 km/ 62.14 mile). For this simulation 50 nodes were used
to dis cretize the compressor station for the inlet and outlet pipes
and five nodes were used for the internal pipes in the
compressor station. Three different types of centrifugal
compressors are used for this simulation, i.e., each compressor
has a different map and different parameters.
The panhandle equation is used to initialize the mass flow
rate and pressure drop in the pipe to start the simulations. At the
first step time, the mass flow rates at each node adjust to the
value that is calculated from the transient equation that is
different from the initial value. These values show the transient
behavior of flow inside the pipe.
A constant pressure boundary condition was applied at the
head of the inlet pipe that enters the compressor station, and t he
end of the outlet pipe that exit the compressor station, as shown
in Figure 2. Another boundary condition for this simulation is
constant speed for each compressor as follows:

13000
rA
N
rpm
12000
rB
N
rpm
11000
rC
N rpm

The formulation presented earlier was used to simulate the
transient performance of this station. Figure 3 shows the
variation of mass flow rate for inlet and outlet pipes at the
compressor station for different nodes. As shown in this figure,
between 50 and 100 minutes is required for the mass flow rate
to become uniform throughout the pipeline segment. At this
time the flow considered to be at steady- state.
Time (min)
M
a
s
s
f
lo
w
r
a
t
e
0 50 100 150 200
100
120
140
160
180
200
220
240
Node 50
Node 1
x 1 kg/s
x 3.4434 MMSCFD

Time (min)
M
a
s
s
f
l
o
w
r
a
t
e
0 50 100 150 200
100
120
140
160
180
200
220
240
Node 50
Node 1
x 1 kg/s
x 3.4434 MMSCFD

(a) (b)
Fig. 3 Variation of mass flow rate for the compressor station
inlet (a) and outlet (b) the pipes
In the same manner, we can explain the results for the outlet
pipe as shown in Figure 3-b. Because different types of
compressors were used, the mass flow rate through each
compressor will be different as show in Figure 4-a.

Time (min)
M
a
s
s
f
lo
w
r
a
t
e
0 50 100 150 200
50
55
60
65
70
75
80
Comp. No. A
Comp. No. B
Comp. No. C
x 1 kg/s
x 3.4434 MMSCFD

Time (min)
0 50 100 150 200
0.11
0.12
0.13
0.14
0.15
0.16
0.17
Comp. No. A
Comp. No. B
Comp. No. C
Fuel Consumption
x 1 kg/s
x 3.4434 MMSCFD

(a) (b)
Fig. 4 Variation of mass flow rate (a) and fuel consumption (b)
for different types of compressor in compressor station
The fuel consumption for this compressor station is found by
using the difference in the mass flow rate between the inlet and
outlet of the compressor station as shown in Figure 4-b.
Figure 5 shows the temperature variation with respect to time
in the inlet and outlet pipes of the compressor station for
different nodes. The temperature at the head of the inlet pipe is
a constant boundary condition. Because of the heat transfer
between the pipe and surroundings, the gas temperature will
decrease in the flow direction and after about 22 km the gas
temperature will be constant (surrounding temperature).
Time (min)
T
e
m
p
e
r
a
t
u
r
e
(
D
im
e
n
s
i
o
n
l
e
s
s
)
0 50 100 150 200
290
300
310
320
330
340
0 km
2 km
6 km
10 km
18 km
>22 km
x 1 K
x 1.8
o
R

Time (min)
T
e
m
p
e
r
a
t
u
r
e
(
D
im
e
n
s
i
o
n
l
e
s
s
)
0 50 100 150 200
290
300
310
320
330
340
350
0 km
2 km
6 km
10 km
18 km
>22 km
x 1 K
x 1.8
o
R

(a) (b)
Fig. 5 Temperature distribution for inlet (a) and outlet (b) pipe
at compressor station

Figure 6-a shows the change in power with respect to time
for the three centrifugal compressors at the compressor station.
As shown in this figure, after a while the power will reach
constant value, which is the steady state condition. Fig. 6-b
shows the variation of isentropic efficiency with respect to time
for each compressor.

Time (min)
0 50 100 150 200
2400
2600
2800
3000
3200
3400
3600
Comp. No. A
Comp. No. B
Comp. No. C
Power
x 1 KW
x 1.341 HP

Time (min)
0 50 100 150 200
68
70
72
74
76
78
80
Comp. No. A
Comp. No. B
Comp. No. C
Effeciency

(a) (b)
Fig. 6 Variation of power (a) and efficiency (b) for different
types of compressor in compressor station

Figure 7 shows the variation of head with respect to flow rate
for each compressor and we can see that all compressors reach
the same value for head because of maintaining the discharge
pressure, but the flow rate in each one becomes different.
Flowrate
H
e
a
d
0.8 0.9 1 1.1 1.2 1.3 1.4
25
30
35
40
45
Comp. No. A
Comp. No. B
Comp. No. C
x 1 m
3
/s
x 2118. 88 cfm
x 1 kj/kg
x 334. 553 ft.lbf/lbm


Fig. 7 Variation of head with respect to flow rate
As shown in this section, the model has an ability to accept
different boundary and initial conditions using different types
6 GMC 2004
and numbers of compressors that show the behavior of
operating parameters for different compressors.

Optimization
The optimization examples presented in this paper have
been carefully selected to illustrate specific points.
In first example, we consider a compressor station with three
dissimilar compressors; this is a common situation in practice
and the optimum operating condition is difficult to find by
other means.
The second example consists of a fourteen compressor station,
and is designed to illustrate the application to situations where
unit shutdowns have to be taken into account.

Example 1:
The system considered here is a single compressor station
with three dissimilar compressors as shown in Figure 2. In this
case, the three units are not identical and they each have
different compressor maps. The compressor speed limits for
this case are given in Table 1, and the goal of the optimization
is to minimize the total fuel consumption while maintaining a
station throughput of 170 kg/s (585.38 MMSCFD).

Table 1- The input data for optimization example 1
N
rA
N
rB
N
rC

Initial Value 13000 rpm 12000 rpm 11000 rpm
Max. Value 15000 rpm 15000 rpm 15000 rpm
Min. Value 10000 rpm 10000 rpm 10000 rpm
Minimum Mass
flow rate
170 kg/s 585.38 MMSCFD

The results obtained by optimization are shown in Table 2.
It is seen that the optimal solution in this case gives us three
different speeds for the three compressors (12650, 11650, and
10650 rpm), and the final mass flow rate is close to its
minimum allowable value at 170.07 kg/s (585.62 MMSCFD).
The optimization reduces the fuel consumption by 3.4910
-3

kg/s (11.9410
-3
MMSCFD), which is a savings of about
8.15%. At the optimum, none of the units is operating at a
limiting speed.

Table 2- Final result for speed, fuel consumption, temperature
and efficiency for optimization example1

Initial Final
N
rA
(rpm) 13000 12650
N
rB
(rpm) 12000 11650
N
rC
(rpm) 11000 10650
Fuel Consumption
(kg/s- MMSCFD) 10
3

42.86 147.59 39.39 - 135.65
Mass flow rate
(kg/s- MMSCFD)
173.12- 596.13 170.07- 585.62
isA
79.75 79.60
isB
79.35 79.35
isC
76.98 76.97
Outlet Temp. T
3
(K) 342.67 340

Example 2:
The system considered here is a single compressor station
with fourteen identical compressors as shown in Figure 8. The
maximum and minimum values of speed for each compressor
are 15000 and 10000 rpm respectably and the goal of the
optimization is to minimize the total fuel consumption while
maintaining a station throughput of 600 kg/s (2066.06
MMSCFD). In this case, we also consider the possibility that
the optimum operating condition for this station may require
the shutdown of one or more units.
2
3
4
5
6
9
10
11
12
13
A B C D
P
A
=6.183977MPa
T
A
= 330.15 K
P
D
=5.10212MPa
D= 35.43 in / 0.9 m
L=62.14 mile / 100 km
D= 39.37 in / 1 m
L=62.14 mile / 100 km
D= 11.81 in / 0.3 m
L=328 ft/ 100 m
D= 11.81 in / 0.3 m
L=328 ft / 100 m
7
8
1
14

Fig. 8 Compressor station using fourteen compressors

Accordingly, we run the optimization separately using 9,
10, 11, 12, 13, and 14 compressors as shown in Table 3. By
comparing the optima thus obtained, we can see that the best
solution is to operate twelve compressors, with five
compressors running at 13,725 rpm and the remaining seven
compressors running at 13,750 rpm.

















Fig. 9 Compressor station optimization map

7 GMC 2004

Table 3. Final result for speed, isothermal efficiency, outlet temperature, and fuel consumption of optimization example 2

14 Compressors on-line 13 Compressors on-line 12 Compressors on-line
Initial Final Initial Final Initial Final
N
r1
(rpm) 15000 13475 15000 13537.5 15000 13725
N
r2
(rpm) 15000 13450 15000 13575 15000 13725
N
r3
(rpm)

15000 13475 15000 13575 15000 13725
N
r4
(rpm) 15000 13475 15000 13575 15000 13725
N
r5
(rpm) 15000 13475 15000 13575 15000 13725
N
r6
(rpm) 15000 13475 15000 13575 15000 13750
N
r7
(rpm) 15000 13475 15000 13575 15000 13750
N
r8
(rpm) 15000 13475 15000 13575 15000 13750
N
r9
(rpm) 15000 13475 15000 13575 15000 13750
N
r10
(rpm) 15000 13475 15000 13575 15000 13750
N
r11
(rpm) 15000 13475 15000 13575 15000 13750
N
r12
(rpm) 15000 13475 15000 13575 15000 13750
N
r13
(rpm) 15000 13475 15000 13575 0 0
N
r14
(rpm) 15000 13475 0 0 0 0
Fuel Consumption
(kg/s- MMSCFD)
2.06 7.09 1.57 5.39 2.00 6.91 1.55 5.35 1.94 6.68 1.55 5.32
Mass flow rate
(kg/s- MMSCFD)
579.12
1994.15
550.51
1895.64
577.53
1988.68
550.51
1895.64
574.60
1978.58
550.51
1895.64
1 is
78.47 78.88 79.382 79.43 79.90 79.94
2 is

78.47 78.50 79.382 79.62 79.90 79.94
3 is
78.47 78.88 79.382 79.62 79.90 79.94
4 is
78.47 78.88 79.382 79.62 79.90 79.94
5 is
78.47 78.88 79.382 79.62 79.90 79.94
6 is

78.47 78.88 79.382 79.62 79.90 79.95
7 is
78.47 78.88 79.382 79.62 79.90 79.95
8 is
78.47 78.88 79.382 79.62 79.90 79.95
9 is
78.47 78.88 79.382 79.62 79.90 79.95
10 i s
78.47 78.88 79.382 79.62 79.90 79.95
11 is
78.47 78.88 79.382 79.62 79.90 79.95
12 is

78.47 78.88 79.382 79.62 79.90 79.95
13 i s
78.47 78.88 79.382 79.62 0 0
14 is
78.47 78.88 0 0 0 0
Discharge Temp.
T
c
(K)
358.94 347.39 357.75 347.09 356.33 347.04


This example shows by using numerical optimization we
can find solutions that may be more fuel efficient than other
solutions. It can also be seen from Table 3 that at the optimum
(i.e. with only 12 compressors running at their optimized
speeds), the efficiency of each unit is about 79.95%. It is also
seen from Table 3 that in this case, the outlet temperature
dropped from 356.33

K to 347.04 K. Most importantly, the
total fuel consumption is reduced from 1.94 kg/s (6.68
MMSCFD) at the initial speeds for the 12-compressor case to
1.55 kg/s (5.32 MMSCFD) at the optimum. It should also be
noted that the second best solution is that obtained using 11 or
13 compressors, followed by the solutions obtained using 10
or 14 compressors; the worst optimum is the one for the 9-
compressor case.
Figure 9 shows the comp ressor station optimization map
for the fourteen-compressor example. This map expresses the
number of compressors on the horizontal axis and the station-
level fuel consumption on the vertical axis. The families of
curves on the map correspond to (1) total gas throughput and
(2) compressor speed. Interesting, for this compressor station
configuration, the optimal number of operating compressors is
consistently 12. the vertical line extending from 12
8 GMC 2004
compressors highlight the optima. The intersection of the
vertical line and each gas throughput line identifies the
compressor speed for optimal operation.

Table 3- Continue

11 Compressors on-line 10 Compressors on-line 9 Compressors on-line
Initial Final Initial Final Initial Final
N
r1
(rpm) 15000 14000 15000 14400 15000 15000
N
r2
(rpm) 15000 14000 15000 14400 15000 15000
N
r3
(rpm)

15000 14000 15000 14400 15000 15000
N
r4
(rpm) 15000 14000 15000 14400 15000 15000
N
r5
(rpm) 15000 14000 15000 14400 15000 15000
N
r6
(rpm) 15000 14000 15000 14400 15000 15000
N
r7
(rpm) 15000 14000 15000 14400 15000 15000
N
r8
(rpm) 15000 14000 15000 14400 15000 15000
N
r9
(rpm) 15000 14000 15000 14400 15000 15000
N
r10
(rpm) 15000 14025 15000 14425 0 0
N
r11
(rpm) 15000 14025 0 0 0 0
N
r12
(rpm) 0 0 0 0 0 0
N
r13
(rpm) 0 0 0 0 0 0
N
r14
(rpm) 0 0 0 0 0 0
Fuel Consumption
(kg/s- MMSCFD)
1.86 6.40 1.553 5.35 1.75 6.03 1.57 5.42 1.61 5.55 1.61 5.55
Mass flow rate
(kg/s- MMSCFD)
569.74
1961.84
550.51
1895.64
532.08
1935.49
550.51
1895.64
550.64
1895.64
550.51
1895.64
1 is
79.83 79.74
78.95 78.80 76.95 76.95
2 is
79.83 79.74
78.95 78.80 76.95 76.95
3 is
79.83 79.74
78.95 78.80 76.95 76.95
4 is
79.83 79.74
78.95 78.80 76.95 76.95
5 is
79.83 79.74
78.95 78.80 76.95 76.95
6 is
79.83 79.74
78.95 78.80 76.95 76.95
7 is
79.83 79.74
78.95 78.80 76.95 76.95
8 is
79.83 79.74
78.95 78.80 76.95 76.95
9 is
79.83 79.74
78.95 78.80 76.95 76.95
10 is
79.83 79.70
78.95 78.80 0 0
11 is
79.83 79.70
0 0 0 0
12 is
0 0
0 0 0 0
13 is
0 0
0 0 0 0
14 is
0 0
0 0 0 0
Discharge Temp.
T
c
(K)
354.59 347.37 352.42 348.17 349.6 349.6

For example, to transport 1835 MMSCFD, the best
operation condition is 12 compressors with 13000 rpm speed,
and for this condition fuel consumption becomes 4.5
MMSCFD.
While this particular compressor station map is specifies to
this compressor station, the methodology suggests that one can
create a compressor station map to any compressor station
configuration.
CONCLUSION
This study used a fully implicit finite difference method to
analyze transient and non-isothermal flow within a pipe and a
quasi-steady flow assumed at each time step of the numerical
9 GMC 2004
solution for centrifugal compressor equations to simulate
compressor stations under non-isothermal conditions. The
results show that:
The simulation approach that is developed here is
adequate for supporting numerical optimization.
Numerical optimization is an effective tool for
optimizing compressor speeds, and can yield
significant reductions in fuel consumption. This, in
turn, will increase throughout.
Determination of the optimal number of
compressors to shutdown in a compressor station
and selection of optimal speeds for the remaining
compressors can be done simultaneously using the
methods developed herein.
The result is a compressor station map that can be
used to optimize the fuel consumption as a function
of mass flow rate through a compressor station.
ACKNOWLEDGMENTS
The authors graciously acknowledge the funding of this
project, Virtual Pipeline System Testbed to Optimize the U.S.
Natural Gas Transmission Pipeline System from the U.S.
Department of Energy, National Energy Technology
Laboratory, Grant No.DE-FC26-01NT41322.
REFERENCES
1. Botros, K. K.; Campbell, P., J.; Mah, D. B., 1989,
Dynamic Simulation of Compressor Station
Installations Including Control Systems, 21st Annual
Meeting Pipeline Simulation Interest Group (PSIG),
19-20 Oct., El Paso, Texas.
2. Botros, K. K.; Campbell, P., J.; Mah, D. B., 1991,
Dynamic Simulation of Compressor Station Operation
Including Centrifugal Compressor and Gas Turbine,
Transactions of the ASME, Vol. 113, April, PP. 300-
311.
3. Botros, K. K., 1994, Transient Phenomena in
Compressor Stations During Surge, Journal of
Engineering for Gas Turbines and Power, Vol. 116,
Jan., PP.133-142.
4. Bryant, M., 1997, Complex Compressor Station
Modeling, 29th Annual Meeting Pipeline Simulation
Interest Group (PSIG), 15-17 Oct., Tucson, Arizona.
5. Stanley, R. A.; Bohannan, W. R., 1977, Dynamic
Simulation of Centrifugal Compressor Systems,
Proceedings of the Sixth Turbomachinery Symposium,
PP. 123-131.
6. Turner, W.J.; Simonson, M.J., 1984, A compressor
Station Model for Transient Gas Pipeline Simulation,
16th Annual Meeting Pipeline Simulation Interest
Group (PSIG), 18-19 Oct., Chattanooga, Tennessee.
7. Turner, W.J.; Simonson, M.J., 1985, Compressor
Station Transient Flow Modeled, Oil & Gas Journal,
Technology, May 20, PP.79-83.
8. Schultz, J. M., 1962, The Polytropic Analysis Of
Centrifugal Compressors, Journal of Engineering for
Power, Jan., PP. 69-82.
9. Odom, F.M., 1990,Turorial on Modeling of Gas
Turbine Driven Centrifugal Compressors, 22nd
Annual Meeting Pipeline Simulation Interest Group
(PSIG), 18-19 Oct, Baltimore, Maryland.
10. Carter, R. G., 1996, Compressor Station
Optimization: Computational Accuracy and Speed,
28th Annual Meeting Pipeline Simulation Interest
Group (PSIG), 23-25 Oct., San Francisco, California.
11. Letnowski, F. W., 1993, Compressor Station
Modeling in Networks, 25th Annual Meeting Pipeline
Simulation Interest Group (PSIG), Oct. 14-15,
Pittsburgh, Pennsylvania.
12. Jenicek, T., Kralik, J., 1995, Optimized Control of
Generalized Compressor Station, 27th Annual
Meeting Pipeline Simulation Interest Group (PSIG),
October 18-20, Albuquerque, New Mexico
13. Botros, K.K., 1990, Thermodynamic Aspects of Gas
Recycling During Compressor Surge Control, ASME
Proceeding Pipeline Engineering Symposium, New
Orleans, LA. PP. 57-65.
14. Boyd, E.A., Scott, L.R., Wu, S., 1997,Evaluating the
Quality of pipeline Optimization Algorithms, 29th
Annual Meeting Pipeline Simulation Interest Group
(PSIG), 15-17 Oct., Tucson, Arizona.
15. Carter, R., 1998, Pipeline Optimization: Dynamic
Programming after 30 Years, 30th Annual Meeting
Pipeline Simulation Interest Group (PSIG), 28-30
Oct., Denver, Colorado.
16. Wu, S., Rios-Mercado, R. Z., Boyd, E. A., Scott, L.
R., 2000, model Relaxation for the Fuel Cost
Minimization of Steady-State Gas Pipeline Networks,
Mathematical and Computer Modeling, Vol. 31, No.
2-3, PP. 197-220.
17. Cobos-Zaleta, D., Rios-Mercado, R. Z., A MINLP
Model for a Minimizing Fuel Consumption on Natural
Gas Pipeline Networks, XI Latin-Ibero-American
Conference on Operations Research, 27-31 Oct.
Concepcin, Chile.
18. Siregar, S., Nababan, S. M., Saragih, R., Nuraini, N.,
Boestami, A., 2000, The Importance of Gas Pipeline
Network Optimization, Proceedings of the Sixth
AEESEAP Triennial Conference, Aug. 23-25, Kuta,
Bali, Indonesia.
19. Edgar, T. F., Himmelblau, D., Bickel, T. C., 1978,
Optimal Design of Gas Transmission Networks,
Society of Petroleum Engineers of ASME, April, PP.
96-104.
20. Osiadacz, A. J., 1994, Dynamic Optimization of
High Pressure Gas Networks Using Hierarchical
Systems Theory, 26th Annual Meeting Pipeline
Simulation Interest Group (PSIG), 13-14 Oct., San
Diego, California.
21. Chapman, K. S. and Abbaspour, M., 2003,
Development of a Virtual Pipeline System Testbed
Using Non-isothermal Transient Simulation, GMRC
2003 Conference, Oct. 6-8, Salt Lake City, UT.
22. Chapman, K. S..; Abbaspour, M., and Keshavarz, A.,
Non-isothermal Transient Gas Pipeline Simulation
Using Fully Implicit Method, Numerical Heat
Transfer Journal, NHT04/3737, submitted.
23. Abbaspour, M.; Chapman, K.S. and Keshavarz, A.,
2004, Dynamic modeling of non-isothermal gas
pipeline system, International Pipeline Conference,
ASME, IPC- 0081, Oct. 4-8, Calgary, Canada.
10 GMC 2004
24. Dranchuck, P.M., Purvis, R. A. and Robinson, D.B.,
1974, Computer Calculations of Natural Gas
Compressibility Factors Using the Standing and Katz
Correlation, Institute of Petroleum Technical series,
No. IP 74-008, PP.1-13.

Biographies of Authors

Dr. Kirby S. Chapman, a professor and director of the
NGML, researches methods and technologies to monitor and
reduce pollutant emissions in turbochargers and internal
combustion engines. He serves as advisor to the GMC, teaches
turbomachinery courses to industry professionals and K-State
students, and mentors graduate students. He earned his Ph.D.
in mechanical engineering at Purdue University under GRI
guidance.
Dr. Prakash Krishnaswami is a professor of
Mechanical and Nuclear Engineering at Kansas State
University, his areas of expertise include computer-aided
mechanical design and optimization of systems. He received his
undergraduate degree from the Indian Institute of Technology,
Madras, his Master's degree from the State University of New
York at Stony Brook. and his doctorate from theUniversity of
Iowa
Mohammad Abbaspour, a Ph.D. candidate and
associate researcher at the NGML, his areas of expertise
include single and two phase gas pipeline simulation,
compressor station analysis, thermal system simulation and
optimization. He received his B.S. and M.S. in mechanical
engineering from the Isfahan University of Technology (IUT) in
IRAN and now he is Ph.D. student in mechanical engineering
at Kansas State University.