You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/277911492

Modeling, Simulation, and Optimization of a High-Pressure Cross-Country


Natural Gas Pipeline: Application of an Ant Colony Optimization Technique

Article  in  Journal of Pipeline Systems Engineering and Practice · May 2015


DOI: 10.1061/(ASCE)PS.1949-1204.0000206

CITATIONS READS

22 344

2 authors, including:

Adarsh Ary
University of Petroleum & Energy Studies
16 PUBLICATIONS   79 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

optimizing city gas distribution of egypt pipeline network View project

All content following this page was uploaded by Adarsh Ary on 11 February 2020.

The user has requested enhancement of the downloaded file.


Modeling, Simulation, and Optimization of a High-Pressure
Cross-Country Natural Gas Pipeline: Application of an Ant
Colony Optimization Technique
Adarsh Kumar Arya 1 and Shrihari Honwad 2
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

Abstract: The process of natural gas transportation through cross-country pipelines, with intermittent repressurization with the help of
compressors that use part of the same gas for energy source, is a very interesting optimization problem that has attracted researchers.
In the present work, an 18-node network connecting a single source to a single delivery point has been selected for analysis. A steady-
state model, incorporating gas flow dynamics, compressor characteristics, and mass balance equations, has been developed. Ant colony
optimization, a comparatively new evolutionary technique in pipeline optimization, has been used for minimizing fuel consumption for
a fixed throughput. A comparison with a similar optimization tool, a solver of a general algebraic modeling system that extracts the principle
of generalized reduced gradient algorithm, indicates an improved solution in terms of fuel consumption minimization. DOI: 10.1061/(ASCE)
PS.1949-1204.0000206. © 2015 American Society of Civil Engineers.
Author keywords: Ant colony optimization; Cross-country gas pipeline; Evolutionary optimization methods; Gas pipeline hydraulics.

Introduction compressors is therefore a very important problem. This paper dis-


cusses and evaluates the application of a relatively new evolution-
Natural gas is a sweeter source of energy present in abundant ary technique—namely, ant colony optimization—for minimizing
amounts (187.3 trillion cubic meters) (Dudley 2013). It is used fuel consumption for a fixed throughput.
not only as a feedstock for the petrochemical industry, but also
as a fuel for domestic, transportation, and thermal power plants.
The efficient use of natural gas will help in reducing the release Existing Pipeline Optimization Methods
of greenhouse gases in addition to yielding economic benefits.
Transportation of natural gas has been an area of major concern. In gas pipeline optimization, techniques such as dynamic program-
For ages, roads, rails, and sea have been important methods for ming (Jamshidifar et al. 2009; Grelli 1985; Osiadacz 1994; Ríos
gas transportation. However, for very long distances, pipelines have Mercado et al. 2002), gradient search techniques (Rozer 2003;
been considered as the most economic, effective, and safe mode of Adeyanju and Oyekunle 2004; Bakhouya and Wolf 2008; Tabkhi
transporting natural gas as compared to other transportation meth- 2007), and heuristic methods (Ferber et al. 1999; Conrado and
ods. The pipeline network system usually includes one or more Rozer 2005; Uraikul and Chan 2004; Diana and Rozer 2002) have
compressor stations. These stations may have one or several com- been quite popular. However, the algorithms used in these tech-
pressors installed in series or parallel. When the gas moves in the niques have a major drawback of getting easily trapped in local
pipeline, pressure and energy are lost due to friction, elevation, etc., optima. Also, the solution obtained from these techniques depends
necessitating the need for recompression through the aid of com- on the initial chosen solution, and these methods are not efficient in
pressors. The energy requirement for running compressors is ob- handling discrete variables. To overcome the aforementioned draw-
tained by the combustion of part of the natural gas that is being back of classical methods, evolutionary methods are now becoming
transported. An estimated 3–5% of the total natural gas transported popular. Evolutionary algorithms are stochastic search methods that
through the pipeline is consumed in turbine-run compressors. The mimic either the natural biological evolution or the social behavior
literature reveals that even a 1% decrease in fuel consumption will of biological species. Genetic algorithm, differential evolution, and
result in a savings of $5 million per year in the United States particle swarm optimization are some of the evolutionary methods
alone (Carter 1996; Summing et al. 2000). Optimizing gas pipe- that have been used in recent years for optimizing pipeline oper-
line operations by minimizing the consumption of natural gas in ations (Elbeltagi et al. 2005). These methods can be applied effi-
ciently at the places where heuristic solutions are not available or
1
Assistant Professor, Dept. of Chemical Engineering, College of Engi- have resulted in unsatisfactory results. The evolutionary methods
neering Studies, Univ. of Petroleum and Energy Studies, Energy Acres, have the potential to hybridize with other methods such as the
Bidholi, Dehradun, Uttarakhand 248007, India (corresponding author). weighted sum method, and have the robustness to handle dynamic
E-mail: adarsharya79@gmail.com changes. These methods have the capability to solve problems
2
Senior Professor, Dept. of Chemical Engineering, College of Engineer- without an initial feasible solution. Ant colony is a comparatively
ing Studies, Univ. of Petroleum and Energy Studies, Energy Acres, Bidholi, newer evolutionary method that is slowly gaining importance in the
Dehradun, Uttarakhand 248007, India. E-mail: shrihari@ddn.upes.ac.in
Note. This manuscript was submitted on October 27, 2014; approved on
field of optimization. The technique mimics the social behavior
March 11, 2015; published online on May 18, 2015. Discussion period open of ants while searching for the shortest route between the nest
until October 18, 2015; separate discussions must be submitted for individual and the food source. The application of ant colony has been used
papers. This paper is part of the Journal of Pipeline Systems Engineering for optimizing pipe diameter (Mohajeri and Taffazzoli 2012) and
and Practice, © ASCE, ISSN 1949-1190/04015008(8)/$25.00. minimizing fuel cost in compressors (Chebouba et al. 2006).

© ASCE 04015008-1 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


The present paper discusses the application of ant colony for min- there are six short pipe segments of very small length (as compared
imizing fuel consumption in compressors in a pipeline network for to the three longest pipelines described earlier). These pipelines are
gas transportation. linked to the inlets and outlets of the compressors. Although the
length of these pipes is very small as compared to the long pipes,
their contribution to pressure drop cannot be neglected. A sche-
Formulation of the Problem matic diagram of a turbo compressor used in the pipeline network
is shown in Fig. 2. The network system comprises one supply node
The optimization problem of minimizing fuel consumption, while and one delivery node. N 0 is designated as the supply node and N 17
obeying all of the transport obligations which would be the con- as the delivery node. The pressure at the supply node and delivery
straints, needs to be considered. The following section looks at node should be within 2% of the specified pressure (in this case,
the basic assumptions under which these are derived. 60 Pa). The pipeline network consists of 18 pressure variables, one
at each node; 15 mass flow rate variables, one in each pipe arc; and
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

Assumptions 6 compressor speed and 6 fuel consumption variables, a pair of


variables for every compressor. A total of 45 decision variables
Following are the basic assumptions in deriving all of the
have been chosen. The nodewise list of pressure and mass flow rate
constraints:
variables along with diameter and length of line segments is listed
1. Steady-state operation.
in Table 1. The variables at the compressor node are listed in
2. Balanced network (nodal material balances are satisfied).
Table 2.
3. Compressor stations including only centrifugal compressors.
The basic model is similar to the one used by Tabkhi (2007),
with modifications that are described in the following section.
The compressor network used in earlier work (Tabkhi 2007) has Objective Function
been selected for comparison of results.
Following is the general definition of minimizing an objective
function:
Network Description Minimize fðxÞ: where x ¼ ½x1; x2; x3; : : : ; xn  is the vector
Fig. 1 shows the gas pipeline network. The network consists of containing optimization variables
three very long pipelines: the first one from the gas supply node subject to
(N 0 ) to the compressor station 1 inlet node (N 1 ), the second from
gi ðxÞ ¼ 0 i ¼ 1; : : : ; m ðequality constraintsÞ
the compressor station 1 outlet node (N 8 ) to the compressor station
2 inlet node (N 9 ), and the third from the compressor station 2 outlet hj ðxÞ ≥ 0 j ¼ 1; : : : ; n ðinequality constraintsÞ
node (N 16 ) to the delivery node (N 17 ).
The two intermediate compressor stations operate to compen- The objective function, as previously described, is subject to
sate for pressure drop in pipelines. Each compressor station in- equality and inequality constraints that will be discussed in the
cludes three parallel centrifugal compressors. At each station, proceeding sections.

N2 C1 N5

N0 N1 N3 C2 N6 N8

N4 N7
C3

N13 C4 N10

N11 N9
N16 N14 C5
N17

N15 C6 N12

Fig. 1. Gas pipeline network

© ASCE 04015008-2 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


Derivation of Objective Function dEcv X  V2

¼ Qcv − W cv þ mi hi þ i þ g × zi
If dEcv =dt is the change in internal energy of the gas per unit time, dt 2
Qcv is the heat transfer to the control volume per unit time, W cv is X  V 2j

the work done on the system per unit time, V i and V j are the veloc- − m j hj þ þ g × zj ð1aÞ
2
ity of gas, hi and hj are the head, zi and zj are the elevations, and mi
and mj are the mass rate per second of gas, each being at the inlet
and outlet of the compressors, then the unsteady-state energy bal- Assumptions
ance equation on the control volume is 1. The process is at steady state, so dEcv =dt ¼ 0, and the mass
flow rate of gas at the inlet of the compressor and at the outlet
of the compressor remains same.
N.G. in Pipeline 2. The process is adiabatic, so Qcv ¼ 0.
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

3. Change in potential energy and kinetic energy is negligible.


Under the aforementioned assumptions, the unsteady-state
Natural Gas. Natural Gas ( energy balance Eq. (1a) reduces to

0 ¼ −W cv þ mj × ðhi − hj Þ ð1bÞ
Centrifugal
Air Combustion Turbine
Compressor Also, the compression process work is done on the constant
Compressor Expander
Chamber mass of gas. Therefore

W comp-ideal ¼ −W cv ¼ mj × ðhj − hi Þ ¼ mj × hij ð1cÞ

In the previous equation, hij , the isentropic head, is a function


Ambient Air Natural Gas (
of the ratio of discharge pressure and suction pressure. The equa-
Fig. 2. Schematic diagram of turbo-compressor used in tion to obtain isentropic head will be explained in the subsequent
pipeline section. The total actual work required in compressing the gas is
obtained by dividing the ideal work by the isentropic efficiency

Table 1. Notation Used in Pipeline Network


Compressibility
Serial Arc joining Diameter of pipe Length of pipe factor of gas in Mass rate of gas in Velocity of gas in
number Node Pressure node N i − N j arc N i − N j arc N i − N j pipe arc N i − N j pipe arc N i − N j pipe arc N i − N j
1 N0 P0 N0 − N1 D1 ¼ 0.787 L1 ¼ 105 z1 m1 v1
2 N1 P1 N1 − N2 D2 ¼ 0.330 L2 ¼ 200 z2 m2 v2
3 N2 P2 N1 − N3 D3 ¼ 0.381 L3 ¼ 300 z3 m3 v3
4 N3 P3 N1 − N4 D4 ¼ 0.330 L4 ¼ 100 z4 m4 v4
5 N4 P4 N5 − N8 D5 ¼ 0.330 L5 ¼ 200 z5 m5 v5
6 N5 P5 N6 − N8 D6 ¼ 0.330 L6 ¼ 100 z6 m6 v6
7 N6 P6 N7 − N8 D7 ¼ 0.330 L7 ¼ 200 z7 m7 v7
8 N7 P7 N8 − N9 D8 ¼ 0.838 L8 ¼ 105 z8 m8 v8
9 N8 P8 N 9 − N 10 D9 ¼ 0.381 L9 ¼ 100 z9 m9 v9
10 N9 P9 N 9 − N 11 D10 ¼ 0.330 L10 ¼ 100 z10 m10 v10
11 N 10 P10 N 9 − N 12 D11 ¼ 0.432 L11 ¼ 100 z11 m11 v11
12 N 11 P11 N 13 − N 16 D12 ¼ 0.330 L12 ¼ 100 z12 m12 v12
13 N 12 P12 N 14 − N 16 D13 ¼ 0.330 L13 ¼ 400 z13 m13 v13
14 N 13 P13 N 15 − N 16 D14 ¼ 0.330 L14 ¼ 100 z14 m14 v14
15 N 14 P14 N 16 − N 17 D15 ¼ 0.889 L15 ¼ 105 z15 m15 v15
16 N 15 P15 — — — — — —
17 N 16 P16 — — — — — —
18 N 17 P17 — — — — — —

Table 2. Notation for Compressor Station


Serial Suction and Isentropic Rotational Fuel
number Compressor discahrge pressure head speed Efficiency consumption
1 C1 P2 − P5 h1 ω1 n1 mf1
2 C2 P3 − P6 h2 ω2 n2 mf2
3 C3 P4 − P7 h3 ω3 n3 mf3
4 C4 P10 − P13 h4 ω4 n4 mf4
5 C5 P11 − P14 h5 ω5 n5 mf5
6 C6 P12 − P15 h6 ω6 n6 mf6

© ASCE 04015008-3 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


   
(nis ), driver efficiency (nd ), and mechanical efficiency (nm ). Hence, Qb × 24 × 3600 Pb
vi ¼ 14.7359 × 3 3 2
×
the total actual work required for compression of gas is ðDo × 10 − 2 × ti × 10 Þ Tb
 2   
10 102 102 zi × T
ðmj × hij Þ × × × ð1dÞ × ð11Þ
nis nd nm Pij × 102

Now, if H m is the lower heating value of gas and mf is the fuel


 
consumed in the compressor, then the total energy released from T Pij
combustion per unit time is zi ¼ 1 þ 0.257 − 0.533 × C × ð12Þ
T PC
H m × mf ð1eÞ
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

Equating Eqs. (1d) and (1e) yields the following Eq. (1f):
Equality Constraints
   2 
mj × hij 10 102 102 Eqs. (13)–(22) are the general mass balance equations that have
mf ¼ × × × ð1fÞ
Hm nis nd nm been obtained by applying the mass balance at each junction of
the pipeline network. Eq. (23) is the equation of motion applied
Minimizing fuel consumption in compressors for fixed through- to each pipeline segment (Tabkhi 2007). Eq. (24) is an isentropic
put is the objective of this paper head equation, and Eq. (25) is an isentropic efficiency equation.
X X mj × hij  Both Eqs. (23) and (24) are applicable to each of the six compres-
minfðmi ; Pi ; Pj Þ ¼ min mf ¼ sors used in gas pipeline network (Tabkhi 2007; Smith and Van
i;j∈Ac
Hm Ness 1998)
 2 
10 102 102
× × × ð1gÞ m1 ¼ m2 þ m3 þ m4 ð13Þ
nis nd nm

m2 ¼ mf1 þ m5 ð14Þ
Equations for Natural Gas Property Calculation
The following Eqs. (2)–(6) are the general equations used for cal- m3 ¼ mf2 þ m6 ð15Þ
culating the properties of natural gas—namely, average molecular
weight [Eq. (2)], critical temperature of natural gas [Eq. (3)], criti-
cal pressure of natural gas [Eq. (4)], heat content of gas mixture m4 ¼ mf3 þ m7 ð16Þ
[Eq. (5)], and isentropic exponent [Eq. (6)].
Similarly, density of natural gas [Eq. (7)], mass flow rate of gas
[Eq. (8)], average pressure of natural gas [Eq. (9)], friction factor m8 ¼ m5 þ m6 þ m7 ð17Þ
[Eq. (10)], velocity of gas in the pipeline [Eq. (11)], and compress-
ibility factor [Eq. (12)] are applicable to each pipe arc (Menon
m8 ¼ m9 þ m10 þ m11 ð18Þ
2005; Mohring et al. 2004):
M ¼ M 1 × y1 þ M 2 × y2 þ M 3 × y 3 ð2Þ
m9 ¼ mf4 þ m12 ð19Þ
T C ¼ T C1 × y1 þ T C2 × y2 þ T C3 × y3 ð3Þ
m10 ¼ mf5 þ m13 ð20Þ
PC ¼ PC1 × y1 þ PC2 × y2 þ PC3 × y3 ð4Þ
m11 ¼ mf6 þ m14 ð21Þ
Hm ¼ ðH1 × y1 × M 1 Þ þ ðH2 × y2 × M 2 Þ þ ðH3 × y3 × M 3 Þ
ð5Þ
m15 ¼ m12 þ m13 þ m14 ð22Þ
ðCp1 × y1 þ Cp2 × y2 þ Cp3 × y3 Þ
k¼ ð6Þ  
ðCp1 × y1 þ Cp2 × y2 þ Cp3 × y3 Þ − R
32 × m2i × zi × R × T × log10 Pi =Pj
P2i − P2j ¼
Pij × M π2 × D4i × M
ρi ¼ ð7Þ 16 × fi × zi × R × T × Li
Zi × R × T − ð23Þ
π2 × D5i × M
mi ¼ ρi × qi ð8Þ
Tabkhi (2007) calculated isentropic head (hij ) and compressor
    efficiency (nij ) using rotational speed. However, the authors feel
2 Pi × Pj
Pij ¼ × Pi þ P j − ð9Þ this model needs to be modified as Eqs. (24) and (25), using com-
3 Pi þ Pj pression ratio ðPj =Pi Þ
 −2      ðk−1Þ=k 
e zi × R × T k Pj
fi ¼ −2log10 ð10Þ hij ¼ × × −1 ð24Þ
3.71 × Di M k−1 Pi

© ASCE 04015008-4 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


 ðk−1Þ=k
Pj Table 3. Prediction of Fuel Consumption for Comparison of GRG Model
−1
Pi with ACO Model (Using np ¼ 1.313)
nij ¼  ðn ð25Þ
Pj p −1Þ=np
Pi −1 Serial number Compressor number Tabkhi model Present model
1 Compressor 1 0.18 0.1887
2 Compressor 2 0.19 0.1917
3 Compressor 3 0.19 0.1923
Inequality Constraints 4 Compressor 4 0.06 0.0587
Eq. (26) represents the pressure limits of the gas in pipeline seg- 5 Compressor 5 0.07 0.0612
ments. Eq. (27) gives the upper limit of velocity of gas in pipelines. 6 Compressor 6 0.06 0.0586
Eq. (28) gives the limit of flow rate at the suction side of the Total fuel consumed 0.75 0.7512
compressor to avoid choking in compressors. Eq. (29) gives the
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

lower and upper limit of the rotational velocity of the compres-


sors (Tabkhi 2007). Eqs. (26) and (27) are applicable to each pipe (2) the mean value of fuel consumption, and (3) the standard
arc, and Eqs. (28) and (29) are applicable to each of the six deviation between the best ant and worst ant has been used for
compressors: guiding the next generation of ants to find the optima (Schlueter
2 × ti × S × E × F × T 0 2012). The algorithm of ant colony optimization is illustrated in
1 < Pi < ð26Þ the appendix.
Di
The computational tests were performed on a personal computer
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi with Intel Core (Intel Corporation, California) 2.4-GHz CPU/2-GB
z ×R×T RAM. Coding was done in MATLAB R 2010 package. The feasible
vi ≤ vei where vei ¼ 122 i ð27Þ
Pij × M solution was obtained in a run of 65 s.

   ðkþ1Þ=2×ðk−1Þ
π 2 2
qimax< × Di × ci × Results and Discussion
4 kþ1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The evolutionary method generates a number of solutions from
k × zi × R × T
where ci ¼ ð28Þ which the best solutions are saved in the solution matrix. The au-
M thors have chosen to save the three best ants out of five and improve
solutions with subsequent iterations. The solutions are compared
166.7 < wi < 450 ð29Þ with the solutions obtained from GRG (Tabkhi 2007), and are
shown in Figs. 3–8.
The objective is to minimize the sum of mf0 s [Eq. (1g)], subject The pressure values at each node are the most important vari-
to equality constraints [Eqs. (13)–(25)] and inequality constraints
ables, as they determine compressor functioning and hence fuel
[Eqs. (26)–(29)]. Various properties of the gas are evaluated from
consumption. The optimal pressures obtained using ant colony op-
Eqs. (2)–(6).
timization at various nodes are compared with those obtained by
the GRG method (Tabkhi 2007) in Fig. 3. It is evident from the
Model Validations figure that, despite the pressures being within the bounds of
The model differs from the earlier model (Tabkhi 2007), as it con- 60  2%, the suction pressures in the authors’ solution in both com-
siders isentropic head and efficiency as a function of discharge pressors are lower than that obtained using the generalized gradient
and suction pressure at the compressors instead of rotational speed. technique. However, the delivery pressure at the outlet of the sec-
Further, in the present work, rotational speed has been considered ond compressor is higher as obtained using ACO. As expected, this
as the action variable. The model is validated using pressure and gives a higher delivery pressure at the delivery station. This is
mass flow rate values of the generalized reduced gradient method clearly evident in Fig. 3. As both the suction and delivery pressures
(GRG) (Tabkhi 2007) to calculate the fuel consumed in each com- obtained by ACO are different from those obtained by GRG, the
pressor. These values are compared with fuel consumed in the ear- compression ratios would be different, as would be the fuel con-
lier case (Tabkhi 2007) in Table 3. The closeness of both numbers, sumptions in the compressors. However, before comparing the fuel
as seen in Table 3, establishes the utility of the current model. consumptions, it is necessary to compare other aspects of the
solution—namely, the amount of gas throughput of the pipe network,
the rotational speeds of compressors, the isentropic heads across the
Ant Colony Optimization compressors, and the isentropic efficiencies of the compressors.
Ant colony optimization (ACO) is a nature-inspired optimization Comparison of the gas throughput of ACO with GRG (Tabkhi
algorithm where populations of ants share some information to 2007), as shown in Fig. 4, clearly indicates that ACO results not
achieve the shortest path. While searching for food, biological ants only comply with the throughput requirement, but are marginally
first start to explore the area around their nest. If an ant succeeds in higher.
finding food, it returns back to the nest, laying down a chemical The rotational speeds of various compressors are compared in
pheromone trail to mark the path. This pheromone trail attracts Fig. 5. ACO rotational speeds are higher than those of GRG
other ants to follow the same path, thus enabling them in finding (Tabkhi 2007) in the second compressor station, whereas at the first
the same food source. The basic idea of ant colony algorithms is to station, they are closer. The ACO rotational speeds are not only
mimic this biological behavior with artificial ants, which randomly higher than GRG rotational speeds, but the rotational speeds for
search at first and then use some pheromone-like parameter to ex- each of the compressors are different. This must be because the
plore the search domain defined by an optimization problem mass flow rate through each of the compressors is different. This
(Dorigo and Stützle 2004). In the present work, a pheromone triplet result gives a natural heuristic angle indicating a need for a different
consisting of (1) the probability of a particular ant to be chosen, operating condition for each compressor at a station.

© ASCE 04015008-5 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Pressure at nodes as obtained using GRG and ACO


Fig. 6. Isentropic head of compressors as obtained using ACO
and GRG

Fig. 4. Mass flow rate in pipe arc as obtained using GRG and ACO

Fig. 7. Isentropic efficiency of compressors using ACO and GRG

The implications of different speeds and different mass flow


rates on compressor heads would be interesting to compare. This
has been done in Fig. 6. As anticipated, ACO heads are the median
of the heads for the GRG solution. Heads of station 1 and those of
station 2 are closer to each other in value, unlike the GRG solution,
where station 1 has much higher heads as compared to station 2.
Apparently, this puts a question mark on the efficiency of the com-
pressors, and these have been compared in Fig. 7, where it is clearly
seen that the ACO efficiencies are consistently higher than GRG
efficiencies and are almost the same for each of the compressors.
Having seen so much operational improvement, it should come as
no surprise that fuel consumption in the ACO solution is lower than
that in GRG (Tabkhi 2007), as seen from Fig. 8.
In economic terms, this reduction would save US$352,076.58
Fig. 5. Comparison of rotational speed as obtained using ACO and GRG
per year, assuming the cost of natural gas per kilogram is US$0.74.

© ASCE 04015008-6 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


Step 5: Use evolutionary operator to generate next generation of
ants. The ant having a higher weight factor is the ant to be com-
pared with the new generation of ants. The ants giving better results
are compared with the previous ant with the highest weight factor.
The three best ants are to be kept in the solution matrix, whereas
others have to be discarded.
Step 6: Evaluate the new generation for fitness. Keep the better
ants (nbest ) in the solution matrix and discard the worse ants.
Step 7: Repeat steps 3–6 until stopping criteria have been
satisfied.
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

Notation

The following symbols are used in this paper:


AC = compressor arc;
D = diameter of pipeline (m);
E = efficiency of pipeline;
F = factor of safety;
Fig. 8. Fuel consumption in compressors as obtained using ACO f = friction factor (dimensionless);
and GRG H m = heat content of gas mixture (J=kg);
h = isentropic head across compressors (kJ=kg);
k = isentropic exponent;
L = length of pipe segment (m);
Conclusion M = molecular weight (kg=mol);
m = mass flow rate in pipe arc (kg=s);
In the present work, a pipeline network simulation model has been
mf = mass of fuel consumed in compressors (kg=s);
developed, which incorporates gas flow dynamics, compressor
nd = driver efficiency;
characteristics, and mass balance equations. For a fixed throughput
ni = adiabatic or isentropic efficiency;
and same gas pipeline network, the ACO technique has been com-
nis = isentropic efficiency;
pared with the GRG technique (Tabkhi 2007) for optimal fuel con-
nm = mechanical efficiency;
sumption. The model used is a variation from the earlier model
np = polytropic exponent;
(Tabkhi 2007), but has been validated to give very similar results.
P = pressure (Pa);
This model has been used for the optimization of fuel consumption
Pb = base pressure (Pa);
for a fixed throughput using the ant colony technique. Comparison
Pc = critical pressure (Pa);
reveals that the fuel savings is improved by 0.015 kg=s, resulting in
Pi = suction pressure at compressor (Pa);
a further savings of approximately US$350,000 per year. The op-
Pij = average pressure in pipe segment, ij (Pa);
erational improvement is also evidenced through comparison of ef-
Pj = discharge pressure of compressor (Pa);
ficiency of compressors. An unforced heuristic solution is also seen
Qb = volumetric flow rate of gas measured under standard
to be forming where the mass flow rates and rotational speeds in
conditions (m3 =s);
each of the three parallel compressors are different, while maintain- q = volumetric flow rate (m3 =s);
ing high efficiencies. The improvement of the optimal solution by R = gas constant (m3 · kPa=kmol · K);
the ACO technique demonstrates the capability of the model used S = specified minimum yield stress (pascal);
and the technique. The usefulness of the optimization technique for T = temperature (K);
multiobjective optimization needs further investigation. T 0 = temperature dearation factor;
T b = base temperature (K);
T c = critical temperature (K);
Appendix. Ant Colony Algorithm ti = thickness of pipeline (m);
v = velocity of gas in pipelines (m=s);
(Stopping criteria: maximum time, specified objective function y = mole fraction of individual component in gas mixture;
value, number of evaluations) z = compressibility factor; and
Step 1: Initiate first generation of ants. ω = rotational speed of compressor (rps).
Step 2: Select ant population size containing ðx; yÞ1 ;
ðx; yÞ2 ; ðx; yÞ3 ; : : : ; ðx; yÞn , number of individuals. The population
of ants remains the same throughout in each iteration.
Step 3: Select number of best solution (nbest ) using fitness Supplemental Data
function. These best solutions are to be kept in a solution matrix.
Tables S1–S13 are available online in the ASCE Library (http://
Step 4: Construct pheromone triplet that guides the ant to search
www.ascelibrary.com).
for optima.
Pheromone triplet contains:
1. Weighted factor: gives the probability of a particular ant to be
References
chosen.
2. Mean value: based on the ant that has a higher probability to be Adeyanju, O. A., and Oyekunle, L. O. (2004). “Optimization of natural gas
chosen. transportation in pipelines.” Petroleum and Gas Engineering Program,
3. Deviation: based on the best and worst ants. Univ. of Logos, Nigeria.

© ASCE 04015008-7 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008


Bakhouya, B., and Wolf, D. (2008). “Solving gas transmission problems by MATLAB R 2010 version 7.10.0.499 [Computer software]. Natick, MA,
taking compressors into account.” HECEcole de Gestiondel’Université Mathworks.
de Liege (ULG), Dunkerque, France. Menon, E. S. (2005). Gas pipeline hydraulics, CRC Press, Taylor and
Carter, R. G. (1996). “Compressor station optimization: Computational Francis Group, Boca Raton, FL.
accuracy and speed.” 28th Annual Meeting of Pipeline Simulation Mohajeri, I., and Taffazzoli, R. (2012). “Optimization of tree-structured
Interest Group, Pennwell Publishing, OK. gas distribution network using ant colony optimization: A case study.”
Chebouba, A., Yalaoui, F., Amodeo, L., Smati, A., and Tairi, A. (2006). “A IJE Trans. A: Basics, 25(2), 141–156.
new method to minimize fuel consumption of gas pipeline using ant Mohring, J., Hoffmann, J., Halfmann, T., Zemitis, A., Basso, G., and
colony optimization algorithms.” Proc., 2006 Int. Conf. on Service Lagoni, P. (2004). “Automated model reduction of complex gas pipeline
Systems and Service Management, IEEE, New York. networks.” Proc., 36th Annual Meeting of Pipeline Simulation Interest
Conrado, B. S., and Rozer, M. (2005). A hybrid meta-heuristic approach Group, Palm Springs, CA.
for natural gas pipeline network optimization, Springer, Berlin. Osiadacz, A. J. (1994). “Dynamic optimization of high pressure gas
Diana, C. Z., and Rozer, M. (2002). “A MINLP model for a minimizing networks using hierarchical systems theory.” 26th annual meeting of
Downloaded from ascelibrary.org by University of California, San Diego on 04/18/16. Copyright ASCE. For personal use only; all rights reserved.

fuel consumption on natural gas pipeline networks.” Memoriasdel XI Pipeline Simulation Interest Group, San Diego.
Ríos Mercado, R., Wu, S., Scott, L., and Boyd, E. (2002). “A reduction
Congreso Latino Iberoamericano de Investigación de Operaciones
technique for natural gas transmission network optimization problems.”
(CLAIO), Springer.
Ann. Oper. Res., 117(1), 217–234.
Dorigo, M., and Stützle, T. (2004). Ant colony optimization, MIT Press,
Rozer, M. (2003). “Efficient operation of natural gas pipeline networks.”
Cambridge, MA.
Computational Finding of High Quality Solutions, Int. Applied
Dudley, B. (2013). “BP statistical reviews of world energy.” 〈http://www.bp Business Research Conf., AccessEcon.
.com/statistical review〉. Schlueter, M. (2012). “Nonlinear mixed integer based optimization
Elbeltagi, E., Hegazy, T., and Grierson, D. (2005). “Comparison among technique for space application.” Ph.D. thesis, Univ. of Birmingham,
five evolutionary based optimization algorithm.” Adv. Eng. Inform., England.
19(1), 43–53. Smith, J., and Van Ness, H. (1998). Introduction to chemical engineering
Ferber, E., Philip, P., William, B., and Ujjal, V. (1999). “CNGT installs thermodynamics, 4th Ed., McGraw-Hill Book Company, Singapore.
fuel minimization system to reduce operating cost.” Pipeline and Gas Summing, W., Rios-Mercado, R. Z., Boyd, E. A., and Scott, L. R. (2000).
Industry, 97–102. “Model relaxation for the fuel minimization of steady state gas pipeline
Grelli, G. J. (1985). “Implementing an optimization program for a natural networks.” Math. Comput. Model., 31(2–3), 197–200.
gas transmission pipeline.” Pipeline Simulation Interest Group, Tabkhi, F. (2007). “Optimization of gas transmission networks.” Ph.D.
Albuquerque, New Mexico. thesis, Grenoble Institute of Technology (INP), Grenoble, France.
Jamshidifar, A., Torbati, H. M., and Kazemian, M. (2009). “GTNOpS, Uraikul, V., and Chan, C. W. (2004). “A mixed- integer optimization
an agent-based optimization software for gas transmission network.” model for compressor selection in natural gas pipeline network system
24th World Gas Conf., Argentina. operations.” J. Environ. Inform., 3(1), 33–41.

© ASCE 04015008-8 J. Pipeline Syst. Eng. Pract.

View publication stats J. Pipeline Syst. Eng. Pract., 2016, 7(1): 04015008

You might also like