Journal of Natural Gas Science and Engineering 33 (2016) 255e263

Contents lists available at ScienceDirect

Journal of Natural Gas Science and Engineering

journal homepage: www.elsevier.com/locate/jngse

Layout optimization of natural gas network planning: Synchronizing

minimum risk loss with total cost
Jinyu An*, Shini Peng
College of Urban Construction and Environmental Engineering, Chongqing University, Chongqing 40004, China

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 7 February 2016
Received in revised form
3 May 2016
Accepted 4 May 2016
Available online 12 May 2016

The synchronization of the minimum risk loss and total cost of natural gas pipeline networks at the
planning stage is discussed in this article. Herein, new procedures for optimizing layout are proposed to
minimize the investment cost, operation expense, and the risk loss of the pipeline network. The procedures include two crucial steps: the rst step is tting two risk cost functions (i.e., leakage risk cost
function and corrosion risk cost function), and the second one is achieving the optimal layout by using
the risk cost functions as the edge weight of the minimum spanning tree algorithm. The suggested
method is applied in three different real cases, leading to three distinct optimal layouts, which are more
suitable than that calculated using intelligent algorithms for practical engineering. Then, two optimal
strategies for pipeline network layouts are presented. Different applications that respectively focus on
the leakage risk cost for urban areas and the corrosion risk cost or leakage risk cost for suburban areas are
shown in the above two strategies. These strategies realize a 6.9e21% greater economic benet than that
of the shortest layout.
2016 Elsevier B.V. All rights reserved.

Keyword:
Natural gas network planning
Layout optimization
Nonlinear tting
Risk cost function
Minimum spanning tree

1. Introduction
The investment and operation costs of natural gas pipeline
networks are commonly considered important aspects of the total
investment. Moreover, environmental and social economic losses
should not be ignored at the planning stage of the network design
because the losses caused by natural gas leakage are so huge during
construction and operation stages. Hence, guaranteeing the gas
transmission to individuals with minimum risk loss and effective
cost necessitates optimizing the pipeline network. Safety and reliability at the operation stage of pipeline networks have been
extensively treated previously for their important roles in risk loss
of natural gas transportation (Rios-Mercado and Borraz-Sanchez,
2015).
Pipe diameter and layout are calculated based on the ow horizon and node pressure horizon (Hanbing et al., 2015), respectively, which is a common optimization method in natural gas
network design that has led to a great many optimization models
and algorithms in recent years. The optimization method usually
consists of two steps: rstly, the shortest pipeline layout or

* Corresponding author.
E-mail address: 450044789@qq.com (J. An).
http://dx.doi.org/10.1016/j.jngse.2016.05.017
1875-5100/ 2016 Elsevier B.V. All rights reserved.

minimum pipe laying cost must be identied; secondly, the

appropriate pipe diameter with the lowest possible metal cost is
determined based on the identied layout (Jin and Wojtanowicz,
2010). Operation optimization is a hotspot of the common optimization, and it refers to some parameters, including pipe diameter, pressure and quantity of the compressor, whereas these
parameters are often calculated according to the identied layout.
lu (2014) applied a heuristic random search
ster and Dilaverog
optimization method (Valipour, 2016) to model an existing
network over a long-run planning horizon with optimum discounted operation and capital costs. In Berniers work (2010), a
multi-objective model was proposed by combining a process owsheeting model and a separate process-integration model without
considering the risk loss. Although various optimization models
and improved algorithms have been developed (e.g., Khan and Lee,
2013; Mohammad et al., 2015; Pfetsch et al., 2015; El-Sheikh, 2013;
Sanchez-Ubeda and Berzosa, 2007), risk loss was always ignored
(Schmidt et al., 2015; Zhu et al., 2001; Valipour et al., 2013). As
mentioned previously, safety and reliability are widely considered
at the operation stage of pipeline networks (Rios-Mercado and
Borraz-Sanchez, 2015; Schmidt et al., 2015), while relevant
studies performed at the layout or planning stage are rather
limited. According to the data from developed countries surveying
bureaus, the economic loss caused by corrosion is 2e4% of annual

256

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

gross national economies (Gangya et al., 2007). Every year, economic losses worth 2.8
1011 RMB are caused by corrosion, which
has become a critical factor in inuencing the safety and reliability
of pipeline operation (Borraz-Sanchez and Haugland, 2013). Additionally, the corrosion has even brought about a large amount of
social issues (Ossai et al., 2015).
Synchronizing the layout, risk loss and planning stages in one
model is challenging because the mathematical optimization of
natural gas pipelines is difcult to develop. Previous approaches
usually minimized the risk loss by employing network security
assessment technology and integrity management technology,
which are based on the deterministic layout (Stalheim, 2011;
Suresh et al., 2008; Reitenbach et al., 2015; Wu et al., 2007), and
therefore synchronization of the risk loss and the planning stage
could not be achieved.
Overall, three limitations exist in previous studies according to
the above discussion: 1) common total optimization or operation
optimization is created only after the layout is created; 2) Optimal
layout is obtained by using intelligent algorithms or conventional
optimization algorithms, which are not consistent with the actual
pipeline laying path; 3) Common total optimization does not
consider economic loss caused by environmental disruption or
casualties derived from the risk factors.
Synchronizing total cost, risk loss and the layout at the planning
stage could not be found in any of the literature. The objective of the
present work is to propose the problem of new layout optimization,
in which new procedures are presented and the two risk costs are
developed based on the above three limitations. In addition, two
strategies for optimizing the pipeline network layout are obtained
by applying the procedures.

2. Problem statement of the new layout optimization

The problem statement of the new layout optimization is to
synchronize the minimum of risk loss and total cost of natural gas
pipeline networks at the planning stage. The mathematical model
and the calculation procedures, which constitute the crucial
framework of this problem, are described in the following
subsections.

2.1. Mathematical model

The layout optimization presented by Sanaye and
Mahmoudimehr (2013) is committed to minimizing total costs
(containing investment and operating costs). However, without risk
loss, the layout optimization results are not entirely consistent with
practical application, which is mainly due to the crossover and
mutation of the genetic algorithm (GA) program. GA is based on the
random combination of one group of network data, and hence the
as-obtained optimal layout is not suitable for pipe laying. In our
work, a feasible layout decision is obtained by employing graph
theory because the original and input data are derived from the
actual pipeline laying path.
The edge weight in this article is expressed using cost functions
for leakage risk and corrosion risk according to graph theory, in
which the edge weight of the network graph can be expressed by
different types of practical data. The corrosion prevention cost,
corrosion risk loss cost, leakage risk loss cost and total investment
cost are used in the corrosion risk cost.
Thus, the two costs (i.e., leakage risk cost and corrosion risk cost)
can be summarized by the following formulas.
Leakage risk cost leakage risk loss cost investment cost

(1)

Corrosion risk cost leakage risk loss cost investment

cost operation cost

(2)

The investment cost and operation cost per unit length of

pipeline include material, labor, installation, purchase and transportation costs, which depend on the pipe length, pipe diameter,
and electricity used by the compressor stations. The leakage risk
cost, the economic loss caused by social environmental consequences and casualties, has been substantial in recent years
because accidents have happened frequently with the increasing
number of pipeline transmission networks that have been built.
Particularly, these costs are calculated by using the net present
value of real project. Hence, it is not necessary to estimate the input
parameters for tting the two cost functions. Moreover, the more
comprehensive these costs are, the more accurate the risk cost
function will be. Equation (1) only emphasizes the leakage risk loss
cost that is included in the above two functions, as it is the main
difference between urban and suburban areas.

2.1.1. Risk cost function

On the one hand, the leakage risk cost should be the focus of
middle and low pressure pipeline networks in urban areas because
environmental economic losses caused by corrosion and third party
damage are very large. On the other hand, the corrosion risk cost
should be the focus of long-distance high pressure pipeline networks in suburban areas because the leakage risk loss cost is too
small to be concerning (Biscan and Loncar, 2010). The corrosion risk
cost includes two aspects: one is the pipeline corrosion risk cost
during construction and operation periods, and another is the
economic loss due to corrosion leakage. The two aspects are associated with the pipe networks geographical environment, which is
often affected by random disturbances, such as the ow of people,
surrounding pipelines and buildings around the pipeline. Therefore, the risk levels of soil corrosion can be used as independent
variables to t the corrosion risk function (Zhiping et al., 2014). Soil
components are often treated as an index in buried pipeline
corrosion because buried pipelines are exposed to various types of
soil in different regions (Wu et al., 2014) .
Both corrosion risk cost and leakage risk cost vary with pipeline
soil environment changes, and two chemical components of the
soil, i.e., the resistivity x and the moisture content y, are dened as
the independent variables of the cost function (Valipour and
Montazar, 2012). Thus, the form of leakage risk or corrosion risk
functions is given as follows:
z z0 a*cos(x/w1) b*sin(x/w1) c*cos(y/w2) d*sin(y/
w2);
(3)
where z is the risk cost; x is the resistivity; y is the moisture content; and w1, w2, w3, and w4 are calculated based on the data
(Table 1) tted by the origin 9.0 version. Equation (3) expresses the
basic relationship of the total cost and the risk loss by using soil
constituents as parameters (i.e., x and y).
Equation (3) is the reasonable functional form to calculate the
two risk costs because both the convergence and the minimum
tting error of this equations form are better than others. Although
the error is not as ideal as that calculated by Einsteins energy
equation, it hardly affects the new procedures and the two layout
strategies proposed in this work.
Parameters are solved by algorithm tting, and the closer that
the multiple correlation coefcient R is to 1, the better the astted result. R is calculated by using the following formula:

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

257

Table 1
Actual cost of medium pressure pipe network in the city of Chongqing.
Pipe section number

Resistivity (U m)

Moisture content (%)

Leakage risk cost (m/km)

Laying cost with anticorrosive(m/km)

Corrosion risk cost (m/km)

1
2
3
4
5
6
7
8
9
10
11

8.82
10.0
14.110
17.030
21.200
23.630
24.192
29.250
48.680
87.630
200.000

12.19
14.91
15.06
15.73
16.47
18.61
19.03
20.08
23.39
22.00
25.00

0.014421
0.024726
0.033159
0.037172
0.040689
0.041171
0.042542
0.042947
0.046142
0.055419
0.058798

0.033650
0.074178
0.131791
0.099479
0.163596
0.109395
0.144262
0.123515
0.152270
0.142506
0.140196

0.048071
0.098904
0.16495
0.136651
0.204285
0.150566
0.186804
0.166462
0.198412
0.197925
0.198994



P
yy
y  y b
R q 
2
P b
P
y  y2
yy

(4)

where y is the average, and b

y is calculated by,

cb
cX b
cX
b
yb
0
1 1
k k

(5)

Two variables, i.e., resistivity and moisture content of soil

composition, are included in the leakage risk cost function based on
the data in Table 1, and the function is presented as:
z 0.0476  0.02222*cos(x/22.84191)  0.00242*sin(x/
22.84191)  0.01005*cos(y/1.7634)  0.00549*sin(y/22.84191) (6)
The correlation coefcient R is 0.94194, and the error analysis of
the tted leakage risk cost function is shown in Fig. 1. In Fig. 1E is
the parameter containing z0, a, b, c, d, w1 and w2 in equation (3),
and B is y in equation (3), and Z is the same as z in that equation.
Based on the results of equations (4) and (5), it was found that the
tting degree is high and the reliability is strong.
The error range of the tted cost function (the leakage risk cost
function) is 25% based on the two input parameters in Fig. 2.
Similarly, the corrosion risk cost function is calculated by
equation (7):

Fig. 2. Residual vs. Independent Plot.

z 0.15225  0.09454*cos(x/26.47906) 0.07653*sin(x/

26.47906)  0.04594*cos(y/1.74224)  0.00882*sin(y/1.74224); (7)
where, z is the corrosion risk cost that contain total investment and
operating cost.

2.1.2. Modeling verication

The tting effect of the two cost functions is veried by

Fig. 1. Fitted Curves Plot of the leakage risk cost.

258

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

comparing the actual investment cost and the tted investment

cost. The two groups of data are calculated and shown in Table 2.
The maximum error value of the cost function, which is less
than 25%, was calculated by comparing the tted investment cost
and the actual investment cost according to the scatter plot of the
two groups data, as shown in Fig. 3. The two strategies which is
obtained by the new procedures for decision-makers to plan
optimal network layouts with maximum benets and minimum
risks, could not be affected by a 24% error value. The important
theory basis to verify tted risk cost functions is shown in Fig. 3.
2.2. The new procedures
Based on the two tted cost functions, the new procedures are
depicted in Fig. 4.
Step 1 (test point): Set the test point (i.e., the pit number) according to the principle of interval, 500 m or 1000 m, and record
the resistivity and moisture content of soil chemical elements by
applying GIS. Notice that this step is based on the planning and
building of a pipeline network.
Step 2 (actual cost): Collect the local actual leakage risk loss cost
caused by a leakage accident, corrosion loss cost, investment
cost and operation cost. These base costs act as the basic data for
the tting procedure. Notice that this step is based on the
building of the pipeline network.
Step 3 (two cost function): Choose the form of the two functions
based on the experience of this paper.
Step 4 (tting program): Calculate the two cost functions of risk
loss by applying the original tting program.
Step 5 (minimum spanning tree): Calculate the value of the edge
weight based on the previous procedures and then employ the
minimum spanning tree to gain the optimal layout.
Step 6 (optimal layout): Analyze the optimal results and determine the optimal layout strategy according to the positives and
negatives of the benet value.

2.2.1. Edge weight

The corrosion risk cost and the leakage risk cost are, respectively
used as edge weights. The edge weight value of each pipeline in
various soil environments is different due to two main factors. One
is that soil chemistry elements are independent parameters of the
two risk cost functions that vary in diverse environments. The other
is that the cost used to t the two risk function changes due to
altered areas or markets. The leakage risk loss cost in urban areas is
larger than that of suburban areas because urban populations are
greater. Thus, the edge weight varies with differences in pipeline
area or environment.
The two main factors, which include the soil erosion, third

Fig. 3. Error value of the tted cost function.

parties, and the surrounding construction, relate to the soil chemistry, which contains soil resistivity, redox potential, soil pH value,
soil moisture content, soil salt content, and tube potential. Herein,
two elements, resistivity and moisture content, are selected to be
the independent parameters of the risk cost function to calculate
the values of edge weight.

2.2.2. Minimum spanning tree algorithm

The Kruskal algorithm is a greedy algorithm, in which the
smallest edge weight was selected from the rest of the side and
then added to a set of edges in each step. As the smallest weight of
the edge, which is connected to two trees, one tree, or a new tree, it
is added to the forest, and then, when all of the vertices are connected, the minimum spanning tree is nally determined. Thus, the
program of [T, c] krusf (b, 1) is used in matlab: T is an n row by
2-column matrix, in which n is the number of pipe segments, and
its column numbers are starting nodes and terminal nodes a. c
expresses the sum of weights. Finally, b is an n row by 3-column
matrix, and its column elements are starting nodes, terminal nodes
and edge weights.

3. Case study
The pipeline networks of the three cases are medium pressure
of urban areas, medium-high pressure A of suburban areas, and the
medium-high pressure B of suburban areas. The main differences
are as follows.

Table 2
Actual cost of testing pipelines.
Pipe section number

Fitted leakage risk cost (m/km)

Fitted corrosion risk cost(m/km)

Fitted investment Cost(m/km)

Actual investment cost(m/km)

1
2
3
4
5
6
7
8
9
10
11

0.041902889
0.046715006
0.023411245
0.035308873
0.035085928
0.046280185
0.045220648
0.064144646
0.038245711
0.04545533
0.026107685

0.163934132
0.19026419
0.118398515
0.161472747
0.175518197
0.189309901
0.178814871
0.24665886
0.115152762
0.170932966
0.118014426

0.122031244
0.143549184
0.09498727
0.126163874
0.140432269
0.143029716
0.133594224
0.182514215
0.076907051
0.125477636
0.091906741

0.109395
0.15227
0.099479
0.123515
0.140196
0.144262
0.131791
0.163596
0.074178
0.123515
0.074178

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

test point

two cost

fitting

functions

program

minimum
spanning tree

259

Optimal
layout

actual cost
Fig. 4. Procedure for solving optimal layout of pipe network.

Table 3
The two costs of the rst natural gas pipelines case.
Starting node

Termination node

The actual length (km)

Pit no.

Leakage risk cost (m)

Corrosion risk cost (m)

1
1
1
2
2
2
2
3
3
3
4
4
4
5
5
5
6
6
7
7
7
8
8
9
10
10
4

2
13
15
3
5
15
16
6
14
16
5
7
13
6
9
14
11
12
8
9
13
9
10
12
11
12
15

0.825
1.48
0.979
1.00
1.14
0.862
0.503
0.511
0.446
0.837
0.289
1.32
0.549
0.691
0.781
0.702
1.27
1.16
0.657
0.733
0.832
0.843
0.276
0.539
0.810
0.648
0.183

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

0.022461458
0.072329366
0.050821858
0.041902889
0.053255107
0.020180493
0.017760363
0.017928909
0.020640963
0.037849682
0.018537803
0.050484339
0.024954976
0.01804041
0.054264378
0.031825148
0.070539887
0.029105163
0.041615141
0.035483075
0.050647299
0.034487709
0.014111916
0.032425641
0.030979026
0.029455054
0.009131175

0.091683253
0.30290319
0.188138189
0.163934132
0.216901177
0.10205952
0.081220792
0.089689799
0.084432216
0.149668047
0.071284411
0.152001646
0.093842198
0.081547968
0.237328175
0.153583257
0.214610017
0.13648955
0.135332818
0.140560768
0.193279671
0.149856269
0.058295257
0.128157321
0.093273738
0.110764562
0.03745357

Table 4
Physical and chemical index of soil test point.
Pit no.

Soil texture

The value of pH

Resistivity(U m)

Moisture content (%)

Natural potential(-V)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

Loam
Loam
Clay
Clay
Clay
Clay
Sandy
Sand
Clay
Clay
Clay
Loam
Loam
Sandy
Clay
Clay
Sandy
Sandy
Loam
Clay
Clay
Loam
Loam
Clay
Clay
Clay
Loam
Loam
Loam
Sandy
Sandy
Sandy

8.30
6.46
5.52
6.58
6.99
6.99
5.95
5.50
6.27
7.24
7.05
8.26
8.30
8.25
5.35
8.02
4.84
8.38
8.15
5.52
5.70
7.65
6.31
8.33
4.91
5.38
6.30
4.35
6.89
5.84
7.76
6.79

10.000
8.820
200.000
14.110
24.560
24.560
48.680
87.630
21.200
29.250
24.192
17.030
23.630
27.690
38.880
63.370
105.500
27.810
25.110
54.260
241.900
110.770
21.040
88.900
25.900
58.600
40.350
51.360
78.800
105.500
27.810
48.680

16.47
25.00
15.06
20.08
21.84
21.84
22.00
12.19
18.61
19.03
23.39
15.73
14.91
18.79
10.32
9.61
22.94
30.46
12.13
26.76
22.56
18.01
13.12
20.35
18.33
21.13
11.47
33.36
13.23
22.94
30.46
22.00

0.550
0.542
0.550
0.542
0.540
0.540
0.760
0.772
0.748
0.810
0.785
0.787
0.840
0.820
0.880
0.870
0.825
0.861
0.878
0.850
0.829
0.841
0.880
0.815
0.834
0.690
0.818
0.820
0.821
0.825
0.861
0.760

loam soil

loam soil

loam soil
loam soil

loam soil
loam soil
loam soil

260

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

1 The same area is used for the rst case and second case so that
the form and parameter values of the two cost functions are the
same but the independent variables are not;
2 A different area is used for the third case so that the form of the
cost function is the same as that of the rst and second case but
the parameter values and independent variable are not the
same.
3 The rst case is for urban areas while the second and third cases
depict suburban areas.
Soil data are shown in Table 3, and the node number and the
length of the laying path are shown in Table 4. Data was obtained
from the geographic data of each node.
The test points (Pit.) are selected between each section of the
pipeline, and soil resistivity, moisture content, natural potential, pH
value and soil texture are chosen as the basic soil physical and
chemical index. All of the data of Pit No. are shown in Table 4, and
they were chosen at random. The more test points that are determined, the more accurate the tted cost function and the optimization results become.
3.1. The rst case
The two costs of the rst example are calculated by applying
steps 1 to 5 of the new procedures and are shown in the Table 3.
Then, the results are as follows by using the minimum spanning
tree Kruskal algorithm.
(1) Shortest path layout without focused on soil erosion risk cost
T Columns 1 through 11

4
15

8
10

4
5

3
14

2
16

3
6

9
12

4
13

10
12

7
8

5
6

c 0.3396.
Here, the savings cost is calculated
0.390140213e0.3396 0.050540213 m.

as

follows:

(3) Layout optimization results focused on the corrosion risk

cost.
T Columns 1 through 11

10
12

8
9

4
5

2
16

4
15

3
14

3
6

1
2

9
12

4
13

2
15

Columns 12 through 15

10
11

6
12

7
8

3
16

c 1.4211.
Here, the cost savings are calculated
1.306645949e1.4211 0.114454051 m.

as

follows:

3.2. The second case

The two costs of the second example are calculated by the same
step as the rst case and are shown in Table 5. Then, the results are
as follows by using the same minimum spanning tree Kruskal
algorithm.
T Columns 1 through 11

10
12

8
9

4
5

8
10

10
11

3
14

4
13

2
16

3
6

6
12

7
8

Columns 12 through 15
Columns 12 through 15

5
9

10
11

1
2

3
16

The total length of shortest layout is 8.5450 km, so the leakage

risk cost and corrosion risk cost are equal to 0.390140213 m and
1.306645949 m, respectively.

7
13

4
15

3
16

1
2

c 0.3731.
Here, the cost savings are calculated
0.371955139e0.3731 0.001144861 m.

as

follows:

(3) Layout optimization results focused on corrosion risk cost.

(2) Layout optimization results focused on leakage risk cost.

T Columns 1 through 11

T Columns 1 through 11
4
5
10
12

8
9

2
16

3
6

4
15

4
5

2
15

3
14

1
2

4
13

6
11

10
12

8
9

3
14

2
16

3
6

1
2

4
13

2
15

2
3

4
15

Columns 12 through 15

Columns 12 through 15

10
11

9
12

5
9

7
13

8
10

5
9

7
8

10
11

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

261

Table 5
The two costs of the second natural gas pipelines case.
(1) The length of the pipeline is not changed, so the total length of the shortest layout is also 8.5450 km. However, the area of the network changes as the pipelines are
planned further away from the crowded city such that the leakage risk cost is smaller. The leakage risk cost and the corrosion risk cost are 0.371955139 m and
1.621615599 m, respectively.
(2) Layout optimization results focused on leakage risk cost.
Starting node

Termination node

The actual length (km)

Leakage risk cost (m)

Corrosion risk Cost (m)

1
1
1
2
2
2
2
3
3
3
4
4
4
5
5
5
6
6
7
7
7
8
8
9
10
10
4

2
13
15
3
5
15
16
6
14
16
5
7
13
6
9
14
11
12
8
9
13
9
10
12
11
12
15

0.825
1.48
0.979
1.00
1.14
0.862
0.503
0.511
0.446
0.837
0.289
1.32
0.549
0.691
0.781
0.702
1.27
1.16
0.657
0.733
0.832
0.843
0.276
0.539
0.810
0.648
0.183

0.057322
0.094934
0.062011
0.060874
0.068581
0.047878
0.026112
0.026127
0.022254
0.040905
0.01399
0.061664
0.025408
0.03141
0.035501
0.031825
0.05743
0.048607
0.026878
0.028034
0.03182
0.029765
0.009684
0.014675
0.021147
0.016259
0.004284

0.091683
0.170426
0.112735
0.117663
0.134536
0.10206
0.081221
0.08377
0.075367
0.143071
0.0494
0.231684
0.097593
0.123561
0.147851
0.133565
0.243536
0.222922
0.134464
0.150019
0.17138
0.178054
0.060383
0.125214
0.192593
0.159835
0.05561

T Columns 1 through 11

c 1.5069.
Here, the cost savings are calculated
1.621615599e1.5069 0.114715599 m.

as

follows:

8
9

3.3. The third case

The two costs of the third example are calculated by the same
steps as the above two cases and are shown in the Table 6. Then, the
results are as follows by using the same minimum spanning tree
Kruskal algorithm.
T Columns 1 through 11

10
12

10
12

8
10

2
3

4
5

2
16

2
15

1
2

3
14

3
6

4
15

2
16

4
5

2
3

3
14

2
15

1
2

3
6

Columns 12 through 15

7
13
8
9

8
10

10
11

4
13

7
9

2
5

c 1.2728.
Here,
cost
savings
are
calculated
1.367118542e1.2728 0.094318542 m.

Columns 12 through 15

as

follows:

4. Results and discussion

7
13

4
13

10
11

c 0.3087.
Here, the cost savings are calculated
0.390021523e0.3087 0.081321523 m.

7
9

as

follows:

(3) Layout optimization results focus on corrosion risk cost.

Two results are observed according to the values of matrix T and

positive and negative values of benets in Table 7: One is that the
shortest layout of natural gas pipelines is not the optimal layout for
minimizing leakage risk loss based on matrix T, and the other is that
the choice between two costs is an essential determinant for the
optimal layout based on Table 7. Matrix T for the three cases shows
that the pipe segments of the shortest layout vary among the three
different layouts (for example, the ve pipeline segments [8 10; 4
13; 7 8; 5 6; 3 16] in the rst case). In Table 7, the leakage risk cost of
the optimal layout offers 13% higher economic benets than that of
the shortest layout based on 0.050540213, but the corrosion risk

262

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

Table 6
The two costs of the third natural gas pipelines case.
(1) The length of the pipeline is not changed, so the total length of the shortest layout is also 8.5450 km. However, the area of the network changes such that the moisture
content is different. The leakage risk cost and corrosion risk cost is equal to 0.390021523 m and 1.367118542 m.
(2) Layout optimization focus on leakage risk cost.
Starting node

Termination node

The actual length (km)

Leakage risk cost (m)

Corrosion risk cost (m)

1
1
1
2
2
2
2
3
3
3
4
4
4
5
5
5
6
6
7
7
7
8
8
9
10
10
4

2
13
15
3
5
15
16
6
14
16
5
7
13
6
9
14
11
12
8
9
13
9
10
12
11
12
15

0.825
1.48
0.979
1.00
1.14
0.862
0.503
0.511
0.446
0.837
0.289
1.32
0.549
0.691
0.781
0.702
1.27
1.16
0.657
0.733
0.832
0.843
0.276
0.539
0.810
0.648
0.183

0.021457049
0.09309708
0.070573764
0.018404647
0.023840308
0.021251292
0.021136758
0.023538682
0.021490006
0.045411386
0.019833192
0.053766664
0.02590588
0.027725225
0.049054738
0.041064991
0.070925767
0.061850816
0.04640089
0.043801986
0.030892375
0.025834587
0.007131375
0.011428614
0.03293063
0.027983748
0.008593352

0.091683253
0.370539631
0.267117646
0.085804739
0.112260523
0.091470661
0.084116133
0.096737359
0.089511536
0.215683657
0.085721558
0.155142863
0.122649225
0.064960059
0.157190087
0.148558934
0.313940691
0.282300194
0.192369924
0.185521459
0.124300126
0.105431089
0.031162877
0.051051451
0.093273738
0.110764562
0.037433211

Table 7
The two costs and their benets in three cases.
Name of cost

Case 1

Case 2

Case 3

The two cost of shortest layout(m)

0.390140213
1.306645949
0.3396
0.050540213(13%)
1.4211
0.114454051

0.371955139
1.621615599
0.3731
0.001144861
1.5069
0.1147155(7.1%)

0.390021523
1.367118542
0.3087
0.0813215(21%)
1.2728
0.094318(6.9%)

The leakage risk cost and its benet(m)

The corrosion risk cost(m) and its benet(m)

cost of the former is higher than that of the latter based on

0.114454051 from the rst cases results. The reality, that the
economic loss of environmental pollution and casualties caused by
leakage is very large in urban areas, is entirely consistent with this
conclusion. This conclusion, for which the leakage risk cost is
suitable for the pipe network layout design of urban areas, is
summarized as the rst strategy. In the second case, the corrosion
risk cost of the optimal layout offers a 7.1% higher economic benet
than that of the shortest layout based on 0.1147155. However, the
leakage risk cost of the former is higher than that of the latter based
on 0.001144861 because the economic losses from environmental
pollution and casualties caused by leakage are not large in suburban areas. This conclusion, that corrosion risk loss is suitable for
suburban areas, is summarized as the second strategy. In the third
case, the two risk costs of the optimal layout give 21% and 6.9%
higher economic benets than that of the shortest layout, based on
0.0813215 and 0.094318, respectively, so this conclusion is the
same as the second case. This is the most perfect result because the
two risk costs are suitable for the suburbs. In summary, this optimization layout, obtained by applying the above two strategies,
achieved 6.9e21% higher economic benets over that of the
shortest layout.

5. Conclusions
In the present paper, new efcient optimization procedures are
built to enhance safety and reliability of natural gas pipeline
transportation. The main thrust of this study is the two cost function that was dened based on unique independent parameters to
be used as the edge weight. The new optimization procedures are
applied successfully to three different cases, and then the result
shows that the regular layout cost with the shortest total length is
not less than the optimal layout cost of minimizing pipeline
corrosion risk. As expected, choosing between corrosion risk cost
and leakage risk cost for the deciding factor is crucial to ensure the
best economic and social environmental benets. Finally, leakage
risk cost could be suitable for low pressure pipelines of urban areas,
as shown in case 1; corrosion risk cost could be suitable for higher
pressure pipelines of suburban areas, as shown in case 2; and both
could be suitable for suburban areas, as in case 3. In summary,
leakage risk cost is used as the deciding factor for urban areas, and
corrosion risk cost is suitable for suburban areas, especially because
both corrosion risk cost and leakage risk cost can be used as the
deciding factor for the latter when the two benets values are
positive. This study is distinguished by the problem of the new
layout optimization, in which new efcient optimization

J. An, S. Peng / Journal of Natural Gas Science and Engineering 33 (2016) 255e263

procedures are proposed to obtain a more benecial layout

compared with the shortest layout. In particular, the synchronization of the minimum risk loss and total cost of natural gas pipeline
networks at the planning stage is achieved.
6. Outlook
Notably, although the two cost functions are tted accurately
based on the resistivity and moisture content of the test point, the
inuence of the pipeline environment includes not only the two
factors but also the third party destruction and climate conditions.
Therefore, the selection principle of the independent variables
could be more standardized to make the value of these optimization results more widespread.
The two risk cost function presented in the rst paper can be
tted more successfully by applying an intelligence algorithm, such
as a neural network, to obtain better economic benets of optimal
layout, which is the next step of our research. Thus, leakage risk
cost functions containing more parameters (i.e., diameter and
pressure) will be provided. A new model (Seyyed et al., 2012) and
improved algorithm program, which could synthesize the optimization pipe diameter, layout and city gate station at the beginning
of the planning stage, will be developed to achieve overall optimization of pipe networks. These strategies will be important and
effective support tools for aiding decision makers in optimizing NG
transmission networks.
References
Biscan, D., Loncar, D., 2010. Optimization of waste heat utilization in gas turbine
pipeline compressor station. Strojarstvo 52 (4), 475e487.
Borraz-Sanchez, C., Haugland, D., 2013. Optimization methods for pipeline transportation of natural gas with variable specic gravity and compressibility. Top
21 (3), 524e541.
El-Shiekh, T.M., 2013. The optimal design of natural gas transmission pipelines.
Energy Sources Part B Econ. Plan. Policy 8 (1), 7e13.
Gangya, Z., Shunmin, L., Chunying, C., Huizheng, Sun, 2007. Soil corrosive factors
research for grounding network of Hebei cangzhou. J. soil (6), 1036e1042.
Hanbing, Q., Meiwei, W., Lili, W., Wenhao, Z., 2015. Tree water injection pipe
network optimization design based on EM algorithm. J. Water Resour. Water
Eng. (03)
Jin, L., Wojtanowicz, A.K., 2010. Optimization of large gas pipeline network-a case
study in China. J. Can. Pet. Technol. 49 (4), 36e43.
Khan, M.S., Lee, M., 2013. Design optimization of single mixed refrigerant natural

263

gas liquefaction process using the particle swarm paradigm with nonlinear
constraints. Energy 49 (1), 146e155.
Mohammad, V., Mohammad, A., Saeid, E., 2015. Surface irrigation simulation
models: a review. Int. J. Hydrol. Sci. Technol. 5 (1), 51e70.
Ossai, C.I., Boswell, B., Davies, I.J., 2015. Pipeline failures in corrosive environments
e a conceptual analysis of trends and effects. Eng. Fail. Anal. 53, 36e58.
Pfetsch, M.E., Fuegenschuh, A., Geissler, B., Geissler, N., Gollmer, R., Hiller, B.,
Humpola, J., Koch, T., Lehmann, T., Martin, A., Morsi, A., Roevekamp, J.,
Schewe, L., Schmidt, M., Schultz, R., Schwarz, R., Schweiger, J., Stangl, C.,
Steinbach, M.C., Vigerske, S., Willert, B.M., 2015. Validation of nominations in
gas network optimization: models, methods, and solutions. Optim. Methods &
Softw. 30 (1), 15e53.
Reitenbach, V., Ganzer, L., Albrecht, D., Hagemann, B., 2015. Inuence of added
hydrogen on underground gas storage: a review of key issues. Environ. Earth
Sci. 73 (11), 6927e6937.
Rios-Mercado, R.Z., Borraz-Sanchez, C., 2015. Optimization problems in natural gas
transportation systems: a state-of-the-art review. Appl. Energy 147, 536e555.
Sanaye, S., Mahmoudimehr, J., 2013. Optimal design of a natural gas transmission
network layout. Chem. Eng. Res. Des. 91 (12), 2465e2476.
Sanchez-Ubeda, E.F., Berzosa, A., 2007. Modeling and forecasting industrial end-use
natural gas consumption. Energy Econ. 29 (4), 710e742.
Schmidt, M., Steinbach, M.C., Willert, B.M., 2015. High detail stationary optimization
models for gas networks. Optim. Eng. 16 (1), 131e164.
Seyyed, M., Mohammad, E., Mohammad, V., 2012. Parameters estimate of autoregressive moving average and autoregressive integrated moving average models
and compare their ability for inow forecasting. J. Math. Stat. 8 (3), 330e338, 01
January.
Stalheim, D., 2011. Metallurgical optimization of microalloyed steels for oil and gas
transmission pipelines. J. Iron Steel Res. Int. 18, 669e674.
Suresh, K., Ghosh, S., Kundu, D., Sen, A., Das, S., Abraham, A., 2008. Inertia-adaptive
particle swarm optimizer for improved global search. In: Eighth International
Conference on Intelligent Systems Design and Applications, vol. 2, pp. 253e258.
lu, S., 2014. Optimization for design and operation of natural gas
Ster, H., Dilaverog
transmission networks. Appl. Energy 133, 56e69.
Valipour, M., 2016. Optimization of neural networks for precipitation analysis in a
humid region to detect drought and wet year alarms. Meteorol. Appl. 23,
91e100.
Valipour, M., Banihabib, M., Behbahani, S., 2013. Comparison of the ARMA, ARIMA,
and the autoregressive articial neural network models in forecasting the
monthly inow of Dez dam reservoir. J. Hydrol. 476, 433e441.
Valipour, M., Montazar, A., 2012. Sensitive analysis of optimized inltration parameters in SWDC model. Adv. Environ. Biol. August, p. 2574(8).
Wu, X., Li, C., Jia, W., He, Y., 2014. Optimal operation of trunk natural gas pipelines
via an inertia-adaptive particle swarm optimization algorithm. J. Nat. Gas Sci.
Eng. 21, 10e18.
Wu, Y., Lai, K.K., Liu, Y., 2007. Deterministic global optimization approach to steadystate distribution gas pipeline networks. Optim. Eng. 8 (3), 259e275.
Zhiping, Z., Leijing, W., Chaohui, Y., Fayuan, W., Xu, T., 2014. The application of factor
analysis for substation soil corrosivity appraisal. Chin. J. Corros. Prot. 147e152
(02).
Zhu, G.Y., Henson, M.A., Megan, L., 2001. Dynamic modeling and linear model
predictive control of gas pipeline networks. J. Process Control 11 (2), 129e148.