Energy 121 (2017) 781e791

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Steady-state flow computation in gas distribution networks with

multiple pressure levels
Francesco Cavalieri
Department of Structural and Geotechnical Engineering, University of Rome, La Sapienza, Via Antonio Gramsci 53, 00197, Rome, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Natural gas networks are critical lifelines essential to the continued well-being of a community. What-
Received 2 May 2016 ever its scope, the analysis of a gas network cannot rely on simple connectivity methods: limited
Received in revised form tolerance on quantity and quality (pressure) to maintain serviceability to end-users generates the need
12 October 2016
for a flow analysis. Scarceness of the literature on flow analysis for gas networks and limitations of the
Accepted 10 January 2017
available methods prompted this work. A novel complete steady-state flow formulation is reported, up to
Available online 11 January 2017
the governing nonlinear system of equations and the expression of the error function to be minimised to
find the solution. Important features, such as the correction for elevation change in pipes and the
Keywords:
Flow analysis
pressure-driven mode, are included. The possibility to treat multiple pressure levels, as is the case of real
Flow equations networks, represents the main novelty of this work. The presented procedure was coded into a pro-
Pressure Ratio gramming language and applied to several test cases, one of which being a non-trivial realistic gas
Elevation change network with 67 nodes and 88 edges. Such examples served the purpose to validate the formulation and
Pressure-driven to show its computational performance in the presence of multiple pressure levels.
Numerical solution © 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Natural Gas (NG) is today considered the cleaner alternative
among fossil fuels, thanks to the reduced emissions of greenhouse
gases [1]. For this reason, its global consumption worldwide is
growing, and is expected to double by 2030 [2]. The transportation
of large quantities of NG (referred to as “gas” hereafter) across
countries and regions is carried out by pipeline networks. Gas
networks, together with other Critical Infrastructure (CI) systems,
such as power, water, transportation (road, railway) and commu-
nication networks, are lifelines essential to the continued well-
being of a modern community. These CI systems are charac-
terised by multiple mutual interactions. Gas networks have strong
links especially with electric power networks and buildings. An
appropriate analysis of a gas distribution network is crucial when
the analysis scope is not just this network, but the entire set of
interacting CI systems and buildings of an urban area [3].
As for other networks, the behaviour of a gas network can be
represented in terms of topological, connectivity analysis or of a
more refined flow analysis; the latter includes connectivity and
involves computation of the system’s operational state, in terms of
E-mail address: francesco.cavalieri@uniroma1.it.
node pressures and edge flows. Similarly to electric power net-
works [4], the low tolerance on amount and quality, in terms of
pressure, of gas fed to end-users for maintaining serviceability
often results in connectivity-based methods to be unsuitable for
the analysis of gas systems. For flow analysis of gas networks and
other lifelines, most researchers and practitioners employ com-
mercial and non-commercial (open-source) software packages.
Examples are given by SynerGEE for gas networks [5], ERACS and
EDSA for power networks [6], EPANET for water supply systems [7],
and FLO-2D for natural storm water systems [8]. Most of these tools
focus only on one network and rarely allow an analysis of multiple
interdependent lifelines, despite the strong interactions existing
between systems of modern infrastructures. Further, commercial
tools are closed programs used as black boxes, and cannot be
expanded or modified to meet the increasing needs of researchers.
The goal of this paper is to provide a complete formulation of the
steady-state flow analysis of a gas distribution network, thus
providing an important contribution to this field. The algorithm
was implemented within an open source simulation tool for civil
infrastructures, namely Object-Oriented Framework for Infra-
structure Modelling and Simulation (OOFIMS) [3,9], recently
developed within the European project SYNER-G (2012) [10] and
able to handle system interdependencies.
Despite the importance of flow analysis, current literature

http://dx.doi.org/10.1016/j.energy.2017.01.062
0360-5442/© 2017 Elsevier Ltd. All rights reserved.
782 F. Cavalieri / Energy 121 (2017) 781e791
features only a few works on flow computation in gas systems.
Osiadacz and Pienkosz [11] provide the more detailed flow
computation algorithm for gas systems, which, however, has limi-
tations that will be highlighted in the following. An interesting
document on the subject is the thesis by Hwee [12], presenting a
number of flow equations in use for gas networks, as well as simple
examples that were used here to validate the new proposed algo-
rithm. Abeysekera et al. [13] investigate the impact of using
different gas supply sources on pressure distribution and gas
quality within a network, making use of a flow analysis along the
lines of Hwee [12]. A further paper focussing on the subject is that
by Morais and Lima [14]. Other papers found in the literature are
related to specific aspects of the flow formulation, and will be
referred to within the following sections.
To the best of the author’s knowledge, all methods available in
the current literature on flow computation for gas networks are
limited to a single pressure level. On the other hand, multiple levels
in other lifelines are treated differently from gas networks: for
instance, the presence of multiple voltage levels in electric power
networks is handled with specific equations for transformers, and
the methodology cannot be used for pressure reduction in gas
systems. The formulation presented in this work fills one gap in the
literature, allowing for the computation of gas flows in the presence
of multiple pressure levels. This feature is of paramount importance
in the analysis of real gas systems and specifically in the treatment
of assessment problems, where a network cannot be decomposed
in pressure-homogeneous portions to be solved independently.
This advancement represents the main novelty of this work.
The second important novelty stands in the application of two
features of the methodology, namely the correction for elevation
change in pipes and the pressure-driven formulation, which were
found in the literature but never applied to case studies. Pressure-
driven formulation, analogue to the head-driven formulation of
water supply networks, is essential in a network assessment
problem where, contrary to the situation of a design problem, there
is no assurance that demands will be satisfied and allowance must
be made for gas not reaching an end-user due to excessive
depressurisation along the network; this is the case, for example, of
gas networks affected by earthquake-induced damage [15].
Finally, all referenced papers present small “academic” exam-
ples, whereas the algorithm implemented here was applied to a
non-trivial realistic gas network composed of 67 nodes and 88
edges, thus resulting to be computationally effective and hence
usable within simulation schemes for risk assessment (see e.g.
Ref. [16]).
The remaining sections of this article are organised as follows. In
Section 2, the taxonomy of gas distribution networks is briefly
outlined, limited to the aspects of interest for the paper. Section 3
presents the steady-state flow formulation, up to the definition of
the error function to be minimised in order to obtain the solution.
In Section 4, the implemented formulation is validated and applied
to four examples of increasing complexity, through which all the
discussed features are highlighted. Conclusions and future work are
reported in Section 5.
2. Taxonomy of gas networks
Gas is supplied across a country through several networks
operating at different pressures. A national transmission system
conveys gas from the sources to the different regions of the country,
where it is passed to regional transmission systems operating at
lower pressures. For local conveyance of gas to end-users, the dis-
tribution system is used. The pressure level makes the main dif-
ference between transmission and distribution systems. The former
operate at high pressures (>7 bar), while the latter operate at
medium (75 mbare7 bar) and low (<75 mbar) pressures [17]. In
this paper, the focus is only on distribution (both low and medium
pressure) networks.
A gas distribution network is a graph made up of nodes and
edges. The graph is typically undirected (an edge between nodes i
and j implies the existence of the opposite one, between j and i),
since the flow direction is not known a priori (i.e., before running a
flow analysis) and flow in pipes is allowed in both directions. Nodes
may be stations, demand (or load) nodes (where gas is delivered to
end-users, typically at low pressure) and joints.
Three main types of stations may exist [15]:
1 Metering/Reduction (M/R) stations, which contain metering
equipment for monitoring gas flow, as well as pressure reduc-
tion equipment. In M/R stations gas is derived from the high
pressure transmission network and injected into the medium
pressure distribution network, for which such nodes function as
source nodes;
2 Reduction Groups (RGs), which are stations operating pressure
reduction between medium and low pressure, or between two
medium or low pressure levels. RGs may also serve as load
nodes and deliver gas at medium pressure to industrial users,
and in this case are called RGMs (Reduction Groups and
Metering);
3 Metering stations, which are only flow measurement points.
The edges between nodes are buried pipelines. The most com-
mon employed materials are steel, polyethylene (high-density,
HDPE, or low-density, LDPE), ductile iron and copper [17].
3. Steady-state flow formulation
The topology of a graph with n nodes and m edges is described
by the n
m node-edge incidence matrix, A, such that:
8
< þ1; if the flow in edge j leaves node i
Aij ¼ 1; if the flow in edge j enters node i (1)
:
0; if edge j is not incident to node i
3.1. Kirchhoff’s laws
Kirchhoff’s first law is the principle of conservation of mass. It
states that the algebraic sum of flows in a network of edges meeting
at a point is zero, or, equivalently, at any node in a graph, the sum of
ingoing flows equals the sum of outgoing flows. By defining n2 as
the number of source nodes in the network, it is possible to write
Kirchhoff’s first law in the following matrix form [11]:
A1 $q ¼ Q (2)
where A1 is the (n-n2)
m matrix whose rows are related to load
nodes (and thus obtained by deleting the rows of A related to
source nodes), q is the m
1 vector collecting the edge (or pipe)
flows and Q is the (n-n2)
1 vector of node loads (i.e., the
demands).
Kirchhoff’s second law is the principle of conservation of energy.
It states that the pressure drop around any closed loop is zero. This
law can be expressed in the following matrix form [11]:
AT $P ¼ DP (3)
where P is a n
1 vector whose generic element, Pi (i.e., the one
related to node i), equals the pressure at node i, pi, for low pressure
nodes, and the pressure at node i squared, p2i , for medium pressure
 F. Cavalieri / Energy 121 (2017) 781e791 783

nodes. Similarly, the m
1 vector DP contains the pressure drops (bar) are the gas temperature and pressure, respectively, in stan-
between nodes i and j, (pi e pj), for low pressure nodes, and the dard conditions (Standard Temperature and Pressure, STP), while
pressure drops squared, (p2i e p2j ), for medium pressure nodes. The for the friction factor f the following expression is used:
first term of Eq. (3), AT·P, can be partitioned to separate source sffiffiffi  
nodes from load nodes, as: 1 S$q 0:076
¼ 11:98$E$ (11)
h i   f D
T P
A $P ¼ AT1 AT2 $ 1 ¼ AT1 $P 1 þ AT2 $P 2 (4)
P2
where E is the efficiency factor. Again, in terms of Dp2 one can write:
where A2 is the n2
m matrix complementary of A1, P1 contains the
Dp2 ¼ p2i  p2j ¼ KM $q1:848
(n-n2) unknown pressures (squared for medium pressure nodes) at
load nodes, while P2 contains the n2 known pressures (squared for p2n $S0:848 $T L
¼ $ $q1:848 (12)
medium pressure nodes) at sources. 57:3$108 $Tn2 $143:52 E2 $D4:848
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
KM
3.2. Common flow equations for pipes

Starting from the general steady-state flow equations (see

Ref. [13]), a number of assumptions and simplifications were 3.3. Correction for elevation change in pipes
adopted to derive simplified flow equations, to be used for gas
network analysis. As already said, in this paper the interest is only All the flow equations presented in the previous section were
in low and medium pressure networks. obtained for a horizontal pipeline. In cases of significant elevation
For low pressure networks (p < 75 mbar), Lacey’s equation is difference between the ends of a pipe segment, its effect should be
used. The pipe flow q (m3/h) is expressed as [13]: taken into account, and the flow equations are modified by
v
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi replacing the pipe length with an equivalent pipe length defined as:
u
u p  p $D5
t i j
L$ðes  1Þ
q ¼ 5:72$104 , (5) Le ¼ (13)
f $S$L s

where pi and pj are the pressures (bar) at pipe start and end nodes, function of an elevation adjustment parameter, s, defined as [18]:
respectively, D (mm) is the pipe diameter, L (m) is the pipe length, S  
(non-dimensional) is the specific gravity of gas and f (non-dimen- Hj  Hi
s ¼ 0:0684$S$ (14)
sional) is the friction factor: T$Z
 
12 where Hi and Hj are the upstream and downstream node elevations
f ¼ 0:0044$ 1 þ (6)
0:276$D (m), respectively, and Z is the compressibility factor. Further, a term
es multiplies the downstream node pressure (squared for medium
As will be shown later, it is useful to write flow equations in
pressure equations). For instance, Cox’s equation including the
terms of Dp or Dp2, depending on the pressure level (low or me-
correction for elevation difference is written as:
dium). In the case of Lacey’s equation, one can write:
v
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u
u p2  es $p2 $D5
Dp ¼ pi  pj ¼ KL $q 2
t i j
  q ¼ 1:69$103 $ (15)
S 0:1913 L S$Le
¼ 8
$ 0:0044 þ $ 5 $q2 (7)
32:7184$10 D D
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} Such correction may be not necessary in high and medium
KL pressure systems, since the effect of elevation change on pressure
drop is usually less than 5% of the frictional loss, except for cases of
For medium pressure networks (pressure in the 0.075e7.0 bar
change exceeding several tens metres per kilometre. In low pres-
range), Cox’s and Polyflo equations are usually employed. Cox’s
sure systems, on the contrary, gas flow is more affected by changes
equation does not include a friction factor and is written as [12]:
in atmospheric pressure and gas pressure with elevation [19].
v
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
u
u p2  p2 $D5
t i j 3.4. Pressure-driven formulation
q ¼ 1:69$103 $ (8)
S$L
Generally, in a network design problem, flow analysis is carried
Again, in terms of Dp2 one can write: out in demand-driven mode, meaning that loads are assumed to be
fixed and equal to node demands. However, the satisfaction of
S L
Dp2 ¼ p2i  p2j ¼ KM $q2 ¼ 6 D5
$ $q2 (9) prescribed demands may not be guaranteed in several conditions,
2:86$10
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} such as seismic conditions (when pipes may be damaged due to
KM
ground shaking and permanent deformation) [15] or maintenance
Polyflo equation accounts also for temperature effects and is of a part of the network. For this reason, in a network assessment
given by Ref. [13]: problem the pressure-driven mode should always be adopted,
meaning that loads can be reduced according to node pressure. The
v
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
u latter is the approach followed in the proposed flow formulation.
u p2  p2 $D5
Tn t i j Current literature lacks papers on this subject. To the author’s
q ¼ 7:57$104 $ $ (10)
pn f $S$L$T knowledge, the only attempt to include a pressure-driven formu-
lation in flow analysis of gas networks is the paper by Wang et al.
where T (K) is the average gas flowing temperature, Tn (K) and pn [20], in which the model by Wagner et al. [21] for water supply
784
systems was adapted to gas networks, considering the similarity of
governing equations between the two network types. In particular,
the gas load at node i, Qi(pi), is reduced as follows:
8 p2in
> 0 ; if pi < pi;min
>
> >
>
> sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< pi  pi;min
Qi ðpi Þ ¼ Qi , ; if pi;min  pi < pi;serv
>
> p i;serv  pi;min
>
> pout
>
:
Qi ; if pi  pi;serv
where pi,min is the pressure below which gas delivery is stopped due
to safety reasons, while pi,serv is the pressure above which the de-
mand at node i can be fully satisfied. When pi is between these two
pressure thresholds, it is not sufficient for the delivery of the full
demand Qi; as a consequence, load is reduced. Based on this
formulation, Eq. (2) is modified as:
A1 $q ¼ Q ðP 1 Þ
By allowing the reduction of node loads when pressures are too
low, i.e. by replacing fixed demands with satisfied demands, the
convergence of gradient-based algorithms for the solution of the
nonlinear equations (see Section 3.6) results to be more straight-
forward in the cases of badly designed or damaged networks. As
will be clear from the applications (see Section 4), in the proposed
formulation distinct values are considered for both pressure
thresholds, pmin and pserv, for low pressure and medium pressure
networks.
3.5. Pressure reduction: the Pressure Ratio variable
As already said, the scarce literature on flow computation for gas
networks features only formulations for a single pressure level. The
proposed formulation, on the contrary, allows computing flows in
gas systems in the presence of multiple pressure levels; this makes
it suitable for the analysis of real gas systems and specifically for the
treatment of assessment problems, where a network cannot be
decomposed in pressure-homogeneous portions to be solved
independently. The levels can be all in one pressure range (low or
medium) or even spread in both low and medium pressure ranges.
For example, an analyst can perform a flow analysis on a network
that is driven by two medium pressure levels and one low pressure
level. This feature represents the main novelty of this work.
In the present formulation, pressure can be only reduced.
However, inclusion of pressure increase (e.g., with the presence of
compressor stations) would be straightforward. Pressure is reduced
through RG and RGM stations, which are characterised by an input
pressure pin (the highest) and an output pressure pout (the lowest),
as shown in the examples (see Section 4). RG and RGM stations
differ just for the load, present only in the latter, and hence con-
cerning the pressure reduction they are treated in the same way. In
particular, each station node, identified by a node number N, is split
in two nodes, sharing the same position. The first and second nodes
are assigned the highest (pin) and lowest (pout) pressures, respec-
tively. The second node is identified by the next node number, Nþ1,
in the node list. For RGM nodes, the load is kept applied to the first
node. Between the two nodes a fictitious zero-length edge, called
reduction link, is then added, to preserve the graph connectivity; no
frictional loss takes place in such link, so that its length (zero) and
diameter (which can be assigned arbitrarily) do not affect the
results.
The input and output pressures, pin and pout, for the generic
reduction link are arranged into the variable PR (Pressure Ratio),
defined as:
F. Cavalieri / Energy 121 (2017) 781e791
8
>
> p2in
>
> ; if pin and pout are both MP
>
> p2out
>
>
>
<
PR ¼ ; (18)
> if pin is MP and pout is LP
>
> p out
>
>
>
> pin
(16) >
: ; if pin and pout are both LP
The PR values for all reduction links are then arranged in matrix
form. The m
(n-n2) matrix PR is composed of all ones, except for
the entries related to the end nodes of the reduction links, where
the proper PR values are entered. In this way, PR always multiplies
the downstream node pressure (squared for medium pressure
nodes).
3.6. Error function and numerical solution
(17)
The flow equations presented in Section 3.2, in terms of Dp, can
be arranged in vector form as:
DP ¼ fðqÞ (19)
The m equations in Eq. (19) contain one or two flow equation
types among the available ones (i.e., Lacey’s and Cox’s or Polyflo),
depending on the number and value of pressure levels in the
network. It is important to note that for the generic reduction link k,
the function fk ðqk Þ is set to 0, since pressure is reduced not by
frictional loss, but through PR. Eqs. (17) and (3), together with Eq.
(19), express the flow analysis formulation according to the loop
methods, which are the loop method and loop-node method [11].
The difference between the two methods is in the arrangement of
matrices within the iterative procedure to find the solution; the
current formulation employs a different procedure, and hence it is
valid for the loop methods in general. The alternative formulation,
following the node methods (i.e., node method and node-loop
method), is not of interest in this paper.
The considered system:
A1 $q ¼ Q ðP 1 Þ
(20)
AT $P ¼ fðqÞ
is composed of (n-n2) continuity equations, one for each load node,
and m resistance equations, one for each edge. The unknowns are
the (n-n2) pressures at load nodes and the m flows in the edges. The
system is nonlinear in the resistance equations, so that its solution
must be retrieved by the use of iterative algorithms and the se-
lection of a proper initial guess, X0.
The latter can be partitioned in the two sub-vectors of initial
pressures and initial flows, respectively:
 
P 1;0
X0 ¼ (21)
q0
which are then established in a sequential manner. The vector of
initial pressures, P1,0, contains the nominal pressures (squared for
medium pressure nodes). The latter are assigned starting from the
nodes connected to the ones where pressure is known (i.e., sources
and RG/RGM stations) and then reaching, following an iterative
process, further nodes that are connected at least to one node
where pressure was already assigned. As far as q0 is concerned,
Osiadacz and Pienkosz [11] propose to retrieve it by partitioning the
matrix A1 in two sub-matrices, referred to as the tree and co-tree
parts. The tree part, having dimensions (n-n2)
(n-n2), contains
the columns of A1 related to the tree edges, which are those edges
forming a tree within the network graph, i.e. a subset of the edges
 F. Cavalieri / Energy 121 (2017) 781e791 785

connecting all graph nodes without forming any cycle. The co-tree
part contains the remaining edges, called links. However, it is not
possible to retrieve the tree sub-matrix in the cases of multiple
sources (i.e., n2 > 1), since for a graph of n nodes the number of tree
edges is always (n-1), not (n-n2). For this reason, in the proposed
formulation the initial flows are obtained simply by inverting Eq.
(2), as modified in Eq. (17):
  
q0 ¼ ðA1 Þ1 $  Q P 1;0 (22)

In this way, the initial flows always respect the continuity

equations and result to be balanced.
As already said, the solution of the nonlinear system of equa-
tions must be found iteratively. The problem of solving a nonlinear
system of algebraic equations is equivalent to solving an optimi-
sation problem, with the definition of an error function F(X), taking
vectorial values, of a vectorial variable X. With this position, finding
the solution of the system means finding the root of F(X), that is
finding a point X such that F(X) is zeroed or at least minimised
(finding a global minimum). The error vector expression can be
derived from Eq. (20), by simply moving the right-hand terms to
the left, thus obtaining:

  
P1 A1 $q þ Q ðP 1 Þ
F ¼ ¼0 (23)
q AT1 $PR$P 1  fðqÞ þ AT2 $P 2

where the term Q ðP 1 Þ takes into account the possible reduction of

loads (pressure-driven mode), while the matrix PR, multiplying
element-wise AT1, serves the purpose to reduce pressure throughout
the network. The error vector expression found in Osiadacz and Fig. 1. Sketch of the simple low pressure network (Example 1).
Pienkosz [11] has a similar form, but is a function of the tree and co-
tree parts of A1; also, it misses the matrix PR and the continuity
equations. Table 1
The presented formulation was implemented and coded in Example 1: edge data.

MATLAB® [22] language. For all of the examples in Section 4, the Edge# Start node End Node D (mm) L (mm)
error vector F(X) was minimised by employing the fsolve command, 1 1 2 150 680
with the Trust-Region Dogleg derivative algorithm (the default 2 1 3 100 500
one) as solver: this algorithm, compared to more classical ones such 3 1 4 150 420
as Newton-Raphson, improves robustness when starting “far” from 4 1 5 100 600
5 3 2 100 600
the solution and handles the case when the Jacobian is singular. The
6 3 4 100 340
termination tolerance was set to 109 on both X and the squared 7 5 4 100 420
norm of F(X) (i.e., the scalar function to be minimised). It is
important to note that, since this is a methodological paper, the
focus of Section 3.6 is on the error function expression, Eq. (23), not Table 2
on the employed solver. In other words, given the expression for Example 1 results: node pressures.
F(X), the analyst may use any solver to minimise it.
Node# p (mbar)

2 25.3
4. Applications 3 25.7
4 26.3
The formulation discussed above was applied to four examples 5 25.5
of increasing complexity, presented in the next subsections. The
first two are taken from Ref. [12] and used to validate the algorithm.
The third one, introduced in this work, is again a simple academic
Table 3
example, but is characterised by the presence of two pressure levels Example 1 results: edge flows.
(medium and low). Finally, the fourth example is a more complex
Edge# q (m3/h) D%
gas network, having a realistic topology generated with the
network model proposed by Duen ~ as-Osorio [23]. This paper Hwee (2007)

1 225.825 223.218 1.17

2 86.657 89.794 3.49
4.1. Example 1: simple low pressure network
3 256.686 253.928 1.09
4 80.833 83.060 2.68
Fig. 1 shows a sketch of a simple network, taken from Ref. [12]. 5 24.176 26.782 9.73
Only one pressure level, 30 mbar (low pressure), is present. Of the 6 37.519 36.988 1.44
total n ¼ 5 nodes, one is the source and the remaining ones are load 7 39.167 36.940 6.03

nodes. Since pressure reduction is not operated, the network has no

reduction links, and the m ¼ 7 edges are all pipes.
786 F. Cavalieri / Energy 121 (2017) 781e791

Table 1 reports the edge data for example 1. Elevation is 0 m for Table 4
all nodes. The pressure thresholds, pmin,LP and pserv,LP, were set to Example 2: edge data.

values such that loads were not reduced. Edge# Start node End Node D (mm) L (mm)
Since the network is low pressure, Lacey’s equations was used, 1 1 2 150 680
with S ¼ 0.589, as done by Ref. [12]. Table 2 and Table 3 report the 2 1 3 100 500
results provided by the implemented algorithm, in terms of pres- 3 1 4 150 420
sures at load nodes and flows in all edges, respectively. Negative 4 1 5 100 600
5 1 6 150 600
flows indicate that the flow travels in the direction opposite (from
6 3 2 100 600
node j to node i) to the one initially assumed (from node i to node j). 7 3 4 100 340
The latter table provides also the results found in the reference 8 5 4 100 420
document, which reported only edge flows. 9 6 5 100 600
The comparison shows that flows do not perfectly match, apart
from the two negative signs missing in the [12] results. The reason
of this mismatch stands in the different expression for the KL factor Table 5
in Lacey’s equation, Eq. (7). The KL factor adopted by Ref. [12] has Example 2 results: node pressures.

the form: Node# p (bar)

2 0.954
L
KL ¼ 11:7$103 $ (24) 3 0.960
D5 4 0.967
5 0.966
which can be obtained by introducing a D ¼ 90 mm uniform 6 0.973
diameter in the friction factor expression, Eq. (6). In fact, it was seen
that if Eq. (24) is used in the implemented algorithm for the KL
factor, the results perfectly match. The flows provided by the cur- Table 6
rent formulation, Table 3, are considered the correct ones, since the Example 2 results: edge flows.
variability of pipe diameter in the friction factor is taken into Edge# q (m3/h) D%
account.
This paper Hwee (2007)

1 219.746 219.747 0.000

4.2. Example 2: simple medium pressure network 2 86.890 86.891 0.001
3 238.101 238.107 0.003
4 73.335 73.331 0.005
A sketch of another simple network taken from Ref. [12] is 5 181.928 181.923 0.003
shown in Fig. 2. Also in this case, only one pressure level, 1 bar 6 30.254 30.253 0.003
7 43.364 43.362 0.005
8 14.737 14.695 0.286
9 31.928 31.922 0.019

(medium pressure), is present. There are one source and five load
nodes, for a total of n ¼ 6 nodes. No reduction links are present, so
that the m ¼ 9 edges are all pipes.
Table 4 reports the edge data for example 2. Elevation is 0 m for
all nodes. The pressure thresholds, pmin,MP and pserv,MP, were set to
values such that loads were not reduced.
The entire network is operated at medium pressure: among the
two available equations for medium pressure, the Cox’s one was
used, with S ¼ 0.589, as done in the reference document [12].
Table 5 and Table 6 report the results provided by the implemented
algorithm, in terms of pressures at load nodes and flows in all
edges. The latter table as in the previous case provides also the
reference results from Ref. [12], which once again reported only
edge flows.
The comparison shows that, apart from the two negative signs
missing in the [12] results, the match is very good, up to the second
decimal digit except for one edge. Examples 1 and 2 were used to
validate the formulation implemented in this work.

4.3. Example 3: simple network with two pressure levels

Fig. 3 shows a sketch of a more complex network, with two

Fig. 2. Sketch of the simple medium pressure network (Example 2).
sources, one RGM station, three load nodes and three joints, for a
total of n ¼ 9 nodes. The OOFIMS simulation tool allows analysts to
input manually the reduction links, to the ends of which pin and pout
are assigned to impose the pressure reduction. In this example, two
of the m ¼ 10 edges are reduction links. Thanks to the presence of
 F. Cavalieri / Energy 121 (2017) 781e791 787

Table 7
Example 3 with RGM node duplicated: edge data.

# Start node End Node Type pin (bar) pout (bar) L (m) Le (m)

1 1 2 pipe e e 800 800.35

2 2 3 pipe e e 700 699.51
3 3 4 pipe e e 620 620.27
4 2 5 pipe e e 800 800.35
5 6 10 pipe e e 800 800.35
6 6 1 reduction 3 0.05 0 0
7 9 4 reduction 3 0.05 0 0
8 7 8 reduction 3 0.05 0 0
9 7 9 pipe e e 700 699.88
10 7 10 pipe e e 800 800.07
11 8 5 pipe e e 540 540.33

Table 8
Example 3 results: node pressures.

Node# p (bar)

Cox’s Polyflo

1 0.040 0.050
2 0.030 0.044
3 0.029 0.042
4 0.050 0.050
5 0.037 0.047
6 2.698 3.001
7 2.749 3.000
8 0.042 0.050

Fig. 3. Sketch of the simple network with two pressure levels (Example 3). Table 9
Example 3 results: edge flows.

Edge# q (m3/h)

Cox’s Polyflo

1 57.440 62.985
2 25.117 32.523
3 94.883 87.477
4 47.678 49.538
5 157.440 162.985
6 57.440 62.985
7 94.883 87.477
8 47.678 49.538
9 153.693 217.415
10 143.985 516.953
11 47.678 49.538

pressure).
Following the procedure explained above, the RGM node is
automatically split in two nodes, with a reduction link between
them: the resulting topology is shown in Fig. 4, in which hatched
and grey nodes are at medium and low pressures, respectively.
Table 7 reports the edge data for the expanded example 3 (i.e.,
with the RGM node duplicated). For this network, the matrix PR has
the form:

2 3
1 1 1 1 1 1 1 1
6 1 1 1 1 1 1 1 1 7
6 7
6 1 1 1 1 1 1 1 1 7
6 7
6 1 1 1 1 1 1 1 1 7
6 7
6 1 1 1 1 1 1 1 1 7
6 7
Fig. 4. Sketch of the simple network with two pressure levels (Example 3). The RGM PR ¼ 6
6 180 1 1 1 1 1 1 1 77 (25)
6 1 1 1 180 1 1 1 1 7
node, formerly node 7, has been now duplicated into nodes 7 and 8, at medium and 6 7
low pressures, respectively. A reduction link between these nodes has been added. 6 1 1 1 1 1 1 1 180 7
6 7
6 1 1 1 1 1 1 1 1 7
6 7
4 1 1 1 1 1 1 1 1 5
pressure reduction devices, the network operates at two different 1 1 1 1 1 1 1 1
pressure levels, 3 bar (medium pressure) and 0.05 bar (low
788 F. Cavalieri / Energy 121 (2017) 781e791

The effect of changes in elevation between the nodes was little in the last step, meaning that the reported point may not be
computed assuming T ¼ 293 K and Z ¼ 0.8 (see e.g. Ref. [24]). The near the solution. The same message was output for other solvers.
difference between pipe length L and the equivalent length Le, and Accepting the provided point as the solution, it can be seen that
consequently the impact of elevation change on results, are negli- pressure losses in pipes are underestimated, compared to Cox’s
gible; however, the aim here is just to show and exemplify the results. Finally, the author noted that for demands increased by
features of the algorithm. The pressure thresholds, pmin,LP and 50%, Polyflo equation did not even provide a solution. For this
pserv,LP for the low pressure part, and pmin,MP and pserv,MP for the reason, Polyflo proved to be less robust in finding a solution.
medium pressure part of the network, were set to values such that This deviation between the two flow equations is somehow
loads were not reduced. All edges were assigned a D ¼ 300 mm expected, since the development of most of equations is based on
diameter. the results of gas flow experiments; hence, they are applicable only
Lacey’s and Cox’s equations, with S of natural gas set to 0.6048 in a limited range of pipe diameters, as well as inner surface and
[13], were used for the low and medium pressure parts of the flow conditions. Lacey’s equation has a good behaviour for smooth
network, respectively. Tables 8 and 9 report the solution provided pipe flow law and diameters of 100 mm or smaller. Cox’s equation is
by the implemented algorithm, in terms of pressures at all nodes valid for pressure higher than 350 mbar and velocity lower than
(except for sources, where pressure is a fixed boundary condition) 20 m/s in all pipes [12]. For Polyflo equation, an applicability range
and flows in all edges, respectively. The solution is balanced and was not found in the literature. Further research is needed. It is
energetically congruent. In Table 8 are evident the different pres- important to note again that this is a methodological paper, pro-
sure levels between nodes 6 and 7 (at medium pressure), and the posing a general methodology that is built upon physics-based
remaining ones (at low pressure). governing equations (continuity and resistance) and as such
The network of example 3 was also solved using the second flow
equation available for medium pressure, Polyflo, to gain insight into
the applicability ranges of these two equations. The values chosen Table 10
Example 4: node data.
for the parameters affecting the KM factor are the following: E ¼ 0.9
(it normally ranges between 0.8 and 1, see Ref. [12]), Tn ¼ 273 K, Node# X (km) Y (km) Type pin (bar) pout (bar) Q (m3/h)
pn ¼ 1 bar, T ¼ 293 K. Tables 8 and 9 also report the solution ob- 1 1.071 1.071 joint e e e
tained by using Polyflo equation. 2 1.071 3.214 RG 3 2 e
As can be noted, for this example the two equations lead to quite 3 1.071 5.357 load e e 30
different solutions. Cox’s equation, being function only of S (apart 4 1.071 7.500 RGM 2 0.05 50
5 1.071 9.643 joint e e e
from the parameters common to all equations, i.e. q, L end D), is 6 1.071 11.786 RGM 3 2 60
easier to use and less affected by the choice of parameter values; 7 1.071 13.929 RGM 2 0.05 50
moreover, it resulted to be more robust in providing an accurate 8 3.214 1.071 RGM 3 2 70
solution, against changes in demands, diameters, lengths etc. Pol- 9 3.214 5.357 RG 2 0.05 e
10 3.214 7.500 joint
yflo equation, on the other hand, is function of more parameters e e e
11 3.214 9.643 RG 3 2 e
and much affected by the selected value of efficiency. Further, the 12 3.214 13.929 RG 3 2 e
solution provided for this example is inaccurate: in fact, MATLAB® 13 5.357 1.071 joint e e e
outputs a warning message, saying that the solver found a point 14 5.357 3.214 load e e 20
where the sum of squares of function values is less than the square 15 5.357 5.357 RG 2 0.05 e
16 5.357 7.500 RG 3 2 e
root of the tolerance; however, the sum of squares changed very 17 5.357 9.643 load e e 20
18 5.357 11.786 joint e e e
19 5.357 13.929 load e e 10
20 7.500 1.071 RGM 2 0.05 50
sources RG stations RGM stations loads nodes joints 21 7.500 3.214 RG 3 2 e
22 7.500 5.357 joint e e e
7 12 19 25 32 37 44 23 7.500 9.643 joint e e e
24 7.500 11.786 RG 3 2 e
25 7.500 13.929 RG 2 0.05 e
6 46 18 24 31 48 43 26 9.643 1.071 load 30
12 e e
27 9.643 3.214 RG 2 0.05 e
28 9.643 5.357 RG 2 0.05 e
29 9.643 7.500 RG 3 2 e
10 5 11 17 23 30 36 42 30 9.643 9.643 load e e 20
31 9.643 11.786 load e e 10
32 9.643 13.929 load e e 10
8 4 10 16 49 29 35 41 33 11.786 1.071 RGM 3 2 60
Y (km)

34 11.786 5.357 RG 3 2 e
35 11.786 7.500 joint e e e
6 36 11.786 9.643 RGM 3 2 70
3 9 15 22 28 34 40 37 11.786 13.929 RGM 2 0.05 50
38 13.929 1.071 joint e e e
39 13.929 3.214 RGM 3 2 60
4
2 45 14 21 27 47 39 40 13.929 5.357 joint e e e
41 13.929 7.500 RGM 3 2 50
42 13.929 9.643 load e e 10
2 43 13.929 11.786 joint e e e
1 8 13 20 26 33 38 44 13.929 13.929 load e e 10
45 3.214 3.214 source 8 3 e
0 46 3.214 11.786 source 8 3 e
0 2 4 6 8 10 12 14 47 11.786 3.214 source 8 3 e
X (km)
48 11.786 11.786 source 8 3 e
49 7.500 7.500 source 8 3 e
Fig. 5. Sketch of the realistic network with multiple pressure levels (Example 4).
 F. Cavalieri / Energy 121 (2017) 781e791 789

Table 11 together with the indication of node types. As can be noted, the
Example 4: edge data. graph is characterised by a grid or mesh-like topological structure,
Edge# Start node End Node L (m) typical in urban areas where the main arcs connect suburbs or
1 1 2 2112.7
districts, and can be considered as a transmission/distribution (TD)
2 3 4 2114.3 system. The topology was generated using the network model
3 4 5 2223.9 developed by Duen ~ as-Osorio [23]. Such model, already used by
4 8 45 2112.7 Franchin and Cavalieri [25], aims to represent real TD systems,
5 10 11 2112.7
based on the ideal class of the d-lattice-graph, an unweighted, un-
6 11 46 2225.4
7 46 12 2114.3 directed, regular graph of dimension d with vertices joined to their
8 14 15 2114.3 lattice neighbours according to specified rules. First, the number of
9 15 16 2114.3 vertices n is fixed in order to obtain a square grid, since TD models
10 21 22 2225.4
exist on adjacency matrices of square topologies. Then, m edges of
11 24 25 2114.3
12 27 28 2114.3
the complete graph (an aperiodic TD substrate) are retained with a
13 31 32 2114.3 probability of existence equal to pm. This probability can be esti-
14 33 47 2112.7 mated empirically for each network typology, and the expression
15 47 34 2225.4 provided by Duen ~ as-Osorio [23] for gas systems is:
16 35 36 2112.7
17 36 48 2225.4
18 39 40 2114.3
19 43 44 2114.3
20 1 8 2137.9
21 2 45 2137.3
22 3 9 2139.5 pm ¼ 0:50$n0:05 (26)
23 5 11 2135.3
24 6 46 2137.5 The total number of nodes is 49, five of which are sources and
25 7 12 2136.9 the remaining ones are RG/RGM stations, load nodes and joints. The
26 46 18 2134.6 algorithm can handle a user-defined number of pressure levels: in
27 13 20 2137.9
this case, three levels are present, in particular two medium pres-
28 18 24 2137.5
29 19 25 2136.8 sure (3 bar and 2 bar) and one low pressure (0.05 bar). A total of 65
30 22 28 2136.6 HDPE pipes connect the nodes. No reduction links are initially
31 49 29 2135.9 present: pressure reduction is operated only by reduction groups,
32 28 34 2139.4 which are automatically split in two nodes as explained above.
33 29 35 2138.8
34 33 38 2140.7
Tables 10 and 11 report the node and edge data, respectively, for
35 47 39 2140.1 example 4 (not yet expanded with duplicated RG/RGM nodes).
36 37 44 2133.9 Elevation is 0 m for all nodes. The input pressure pin for source
37 2 4 4228.6 nodes is fictitious and only indicates that such nodes derive gas
38 8 9 4337.4
from the high pressure (>7 bar) network. The minimum pressures,
39 10 46 4337.4
40 13 15 4226.2 pmin,LP and pmin,MP, were set to 20 mbar and 0.2 bar, respectively,
41 20 22 4337.4 based on standard requirements in Italy (see e.g. Ref. [26]). For the
42 21 49 4339.7 remaining pressure thresholds, pserv,LP and pserv,MP, no references
43 22 23 4226.2 were found: the chosen values, pserv,LP ¼ 35 mbar and
44 27 29 4228.6
45 28 30 4337.4
pserv,MP ¼ 1 bar, were assigned arbitrarily, given the illustrative
46 36 37 4339.7 character of the application, and with the aim to show the reduc-
47 38 40 4226.2 tion of loads in pressure-driven mode. The coordinates of nodes,
48 42 44 4228.6 expressed in km from a fictitious origin in Table 10, were converted
49 4 16 4273.4
into geographical coordinates, since in the OOFIMS tool all points
50 45 21 4276.0
51 10 49 4273.3 are georeferenced: pipe lengths in Table 11 were computed
52 46 24 4270.7 accordingly. All pipes were assigned a D ¼ 300 mm diameter.
53 12 25 4269.4 Lacey’s and Cox’s equations, with S set to 0.6048, were used for
54 13 26 4277.2 the low pressure and medium pressure parts of the network,
55 14 27 4275.9
56 16 29 4273.3
respectively. As done for example 3, all RG and RGM nodes were
57 17 30 4271.9 duplicated, and reduction links were added between them,
58 19 32 4269.4 resulting in a graph composed of n ¼ 67 nodes and m ¼ 88 edges.
59 20 33 4277.2 The computational time required for obtaining the solution was
60 21 47 4275.9
0.7 s, using a laptop with 8 GB RAM and a 2 GHz quad core pro-
61 23 36 4271.9
62 25 37 4269.3 cessor. Table 12 reports the pressures at non-source nodes, the
63 27 39 4275.8 demands for the expanded network and the possibly reduced loads.
64 29 41 4273.2 Seven loads, delivered at nodes operating at low pressure, result to
65 31 43 4270.6 be reduced, compared to the respective demands: indeed, it can be
noted that their pressures are lower than pserv,LP ¼ 35 mbar. The
three different pressure levels operated in the network are evident.
always valid. Given the methodology, the analyst will select flow Table 13 reports the edge flows. The three edges with zero flow are
equations to be used on a case-by-case basis. reduction links derived from the split of RGM stations: flow is zero
because the end nodes of such links are dead ends, and the station
4.4. Example 4: realistic network with multiple pressure levels loads were automatically applied to their start nodes, following the
procedure explained above. Also in this example, the solution re-
The topology of the network for example 4 is sketched in Fig. 5, sults to be balanced and energetically congruent.
790 F. Cavalieri / Energy 121 (2017) 781e791

Table 12 Table 13
Example 4 results: node pressures, demands and final loads. Example 4 results: edge flows.

Node# p (bar) Q (m3/h) Q(p) (m3/h) Edge# q (m3/h)

1 1.928 0 0 1 6.676
2 2.921 0 0 2 2.844
3 1.947 0 0 3 23.084
4 0.042 30 30 4 8.246
5 1.861 50 50 5 11.895
6 0.043 0 0 6 8.906
7 1.929 0 0 7 9.110
8 2.886 60 60 8 8.307
9 1.924 0 0 9 23.964
10 1.779 50 50 10 13.510
11 0.040 0 0 11 16.322
12 2.861 70 70 12 2.844
13 1.908 0 0 13 50.150
14 1.880 0 0 14 20.311
15 0.044 0 0 15 50.150
16 2.996 0 0 16 13.269
17 2.987 0 0 17 13.269
18 1.992 0 0 18 29.840
19 2.858 0 0 19 36.369
20 1.905 0 0 20 0.000
21 0.039 0 0 21 60.000
22 0.042 20 20 22 0.000
23 1.837 0 0 23 3.579
24 0.042 0 0 24 20.311
25 2.790 0 0 25 66.421
26 1.860 0 0 26 16.731
27 0.027 20 13.510 27 16.731
28 2.983 0 0 28 16.731
29 0.038 10 10 29 36.369
30 1.784 50 50 30 16.553
31 0.040 0 0 31 36.369
32 2.986 0 0 32 19.815
33 1.990 0 0 33 67.247
34 1.887 0 0 34 67.247
35 1.873 0 0 35 50.000
36 2.967 0 0 36 21.766
37 1.978 0 0 37 7.750
38 1.876 0 0 38 14.017
39 0.044 0 0 39 18.827
40 0.030 30 23.964 40 2.939
41 1.879 0 0 41 21.766
42 0.044 0 0 42 9.947
43 1.887 0 0 43 9.947
44 0.045 0 0 44 31.582
45 2.814 0 0 45 45.486
46 1.876 0 0 46 15.199
47 0.030 20 16.193 47 45.486
48 0.034 10 9.520 48 15.087
49 0.034 10 9.646 49 15.200
50 2.800 60 60 50 46.173
51 1.866 0 0 51 23.084
52 2.947 0 0 52 23.088
53 1.965 0 0 53 26.322
54 1.867 0 0 54 26.322
55 2.788 70 70 55 17.247
56 1.859 0 0 56 46.173
57 1.734 50 50 57 12.251
58 0.038 0 0 58 12.251
59 1.870 0 0 59 29.703
60 2.823 60 60 60 2.009
61 1.882 0 0 61 12.491
62 1.878 0 0 62 29.703
63 2.643 50 50 63 7.414
64 1.762 0 0 64 76.241
65 0.032 10 9.110 65 18.827
66 0.034 0 0 66 5.671
67 0.034 10 9.589 67 13.085
68 19.459
69 28.365
70 79.459
5. Conclusions and future work 71 8.906
72 40.185
In the analysis of a distribution network of natural gas, whatever 73 40.185
74 40.185
its scope, the computation of the system’s operational state, in
 F. Cavalieri / Energy 121 (2017) 781e791
Table 13 (continued )
Edge# q (m3/h)
75 9.466
76 13.085
77 11.895
78 79.466
79 21.543
80 37.098
81 34.445
82 21.543
83 14.336
84 74.336
85 5.430
86 8.906
87 0.000
88 50.000
terms of node pressures and edge flows, is crucial, and represents
the basis for most scopes involving gas networks. Despite this
importance, the currently available papers providing a detailed
description of the whole formulation and of its practical application
are very few. For flow analysis of gas networks and other lifelines,
most researchers and practitioners employ commercial and non-
commercial (open-source) software packages. Most of them focus
only on one network, and rarely allow an analysis of multiple
interdependent lifelines, despite the strong interactions existing
between systems of modern infrastructures. Further, commercial
tools are used as black boxes, and cannot be expanded or modified
to meet the increasing needs of researchers.
This article presented a formulation for the steady-state flow
analysis of gas networks. The presence of multiple pressure levels
represents the main novelty, which allows analysts to compute
edge flows and node pressures for realistic networks with pressure
reduction at some stations. The contribution provided by the paper
is important also for presenting together, in one organic document,
several relevant features, spread over different papers in the cur-
rent literature. Such features include the correction for elevation
change in pipes and the load reduction according to pressure
(pressure-driven formulation). While other papers present small
academic examples, the algorithm implemented here was applied
to a non-trivial realistic gas network (with 67 nodes and 88 edges).
Further three example applications were presented, two of which
were used to validate the software.
The algorithm was implemented within an open source simu-
lation tool for civil infrastructures, (OOFIMS) [3,9], recently devel-
oped within the European project SYNER-G [10] and able to handle
system interdependencies. Future work aims to extend the current
formulation to simulate earthquake-induced physical damage to
gas pipes, so to allow the analysis of a pipeline gas network sub-
jected to earthquakes. In seismic conditions, with pipes possibly
damaged due to ground shaking and permanent deformation, the
satisfaction of prescribed demands may not be guaranteed: the
implemented pressure-driven formulation, imposing loads reduc-
tion according to node pressure, will provide realistic results.
Further, OOFIMS will be used for risk assessment of a real network,
possibly considering the interaction with other infrastructural
systems; for instance, the computation of gas demands will be
based on served population, and the malfunctioning of natural gas
power plants due to damage to gas delivery stations will be taken
into account.
Funding
This work was supported by the European Commission, for the
791
SYNER-G collaborative research project [grant number 244061].
Acknowledgments
The author gratefully acknowledges: Dr. Simona Esposito for the
support in defining the taxonomy of gas networks; and Prof. Paolo
Franchin for the discussion on the flow formulation.
References
[1] Tabkhi F. Optimisation de re seaux de transport de gaz. Toulouse: INPT; 2007.
Ph.D. thesis.
[2] Riva A, D’Angelosante S, Trebeschi C. Natural gas and the environmental re-
sults of life cycle assessment. Energy 2006;31(1):138e48.
[3] Franchin P. A computational framework for systemic seismic risk analysis of
civil infrastructural systems. In: Pitilakis K, Franchin P, Khazai B, Wenzel H,
editors. SYNER-G: systemic seismic vulnerability and risk assessment of
complex urban, utility, lifeline systems and critical facilities. Geotechnical,
geological and earthquake engineering, vol. 31. Netherlands: Springer
Netherlands; 2014. p. 23e56. http://dx.doi.org/10.1007/978-94-017-8835-9_
2.
 Corte
[4] Cavalieri F, Franchin P, Buritica s JAM, Tesfamariam S. Models for seismic
vulnerability analysis of power networks: comparative assessment. Com-
puter-Aided Civ Infrastructure Eng 2014a;29(8):590e607.
[5] Cimellaro GP, De Stefano A, Villa O. Serviceability of natural gas distribution
networks after earthquakes. J Earthq Tsunami 2013;7(2):1350005.
[6] Hasan KN, Rao KSR, Mokhtar Z. Analysis of load flow and short circuit studies
of an offshore platform using ERACS software. In: Proceedings of the power
and energy conference, 2008. PECon 2008. IEEE 2nd International. IEEE; 2008.
p. 543e8.
[7] Tabesh M, Yekta AA, Burrows R. An integrated model to evaluate losses in
water distribution systems. Water Resour Manag 2009;23(3):477e92.
[8] De Risi R, Jalayer F, De Paola F, Iervolino I, Giugni M, Topa ME, et al. Flood risk
assessment for informal settlements. Nat Hazards 2013;69:1003e32.
[9] OOFIMS (Object-Oriented Framework for infrastructure modelling and
simulation). 2010-2017. Accessed 12 October 2016, https://sites.google.com/
a/uniroma1.it/oofims/home/.
[10] SYNER-G. Collaborative research project, funded by the European union
within Framework programme 7 (2007e2013), under grant agreement no.
244061. 2012. Accessed 12 October 2016, http://www.syner-g.eu/.
[11] Osiadacz AJ, Pienkosz K. Methods of steady-state simulation for gas networks.
Int J Syst Sci 1988;19(7):1311e21.
[12] Hwee PH. Application of computer based simulation in gas network system
using graph theory algorithms and numerical methods. 2007 [Master thesis].
[13] Abeysekera M, Wu J, Jenkins N, Rees M. Steady state analysis of gas networks
with distributed injection of alternative gas. Appl Energy 2016;164:
991e1002.
[14] Morais MS, Lima JM. Natural gas network pricing and its influence on elec-
tricity and gas markets. In: Proceedings of the power tech conference, Bologna
(vol. 3). IEEE; 2003.
[15] Esposito S, Iervolino I, d’Onofrio A, Santo A, Cavalieri F, Franchin P. Simulation-
based seismic risk assessment of gas distribution networks. Computer-Aided
Civ Infrastructure Eng 2015;30(7):508e23.
[16] Cavalieri F, Franchin P, Pinto PE. Application to selected transportation and
electric networks in Italy. In: Pitilakis K, Franchin P, Khazai B, Wenzel H,
editors. SYNER-G: systemic seismic vulnerability and risk assessment of
complex urban, utility, lifeline systems and critical facilities. Netherlands:
Springer; 2014b. p. 301e30. http://dx.doi.org/10.1007/978-94-017-8835-9_
10. Online ISBN: 978-94-017-8835-9.
[17] Snow DA. Plant engineer’s reference book e second Edition. Butterworth-
Heinemann; 2002.
[18] Menon ES. Gas pipeline hydraulics. CRC Press; 2005.
[19] Wilson GG. Flow calculations and sizing of mains, chapter 5 in IGT home study
course d gas distribution, 62. Chicago: Institute of Gas Technology; 1963.
[20] Wang Y, Huang H, Su BN. Vulnerability analysis of urban natural gas network
under cascading failures. In: Proceedings of the 2014 International workshop
on urban planning and management, Kanazawa, Japan; 2014.
[21] Wagner JM, Shamir U, Marks DH. Water distribution reliability: simulation
methods. J Water Resour Plan Manag 1988;114(3):276e94.
[22] The MathWorks Inc. MATLAB® R2011b, Natick, MA. 2011.
~ as-Osorio L. Interdependent response of networked systems to natural
[23] Duen
hazards and intentional disruptions. Georgia Institute of Technology; 2005.
Ph.D. thesis.
[24] Coker AK. Ludwig’s applied process design for chemical and petrochemical
plantsvol. 1. Elsevier Gulf Professional Publishing; 2007.
[25] Franchin P, Cavalieri F. Probabilistic assessment of civil infrastructure resil-
ience to earthquakes. Computer-Aided Civ Infrastructure Eng 2015;30(7):
583e600.
[26] JRC. Fornitura di gas naturale per il Centro comune di ricerca di Ispra (VA) (in
Italian). Available at:. 2015. Accessed 12 October 2016 https://web.jrc.ec.
europa.eu/callsfortender/index.cfm?action¼app.tender&id¼2890.