Professional Documents
Culture Documents
Energy
journal homepage: www.elsevier.com/locate/energy
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 node pressures and edge flows. Similarly to electric power net-
works [4], the low tolerance on amount and quality, in terms of
Natural Gas (NG) is today considered the cleaner alternative pressure, of gas fed to end-users for maintaining serviceability
among fossil fuels, thanks to the reduced emissions of greenhouse often results in connectivity-based methods to be unsuitable for
gases [1]. For this reason, its global consumption worldwide is the analysis of gas systems. For flow analysis of gas networks and
growing, and is expected to double by 2030 [2]. The transportation other lifelines, most researchers and practitioners employ com-
of large quantities of NG (referred to as “gas” hereafter) across mercial and non-commercial (open-source) software packages.
countries and regions is carried out by pipeline networks. Gas Examples are given by SynerGEE for gas networks [5], ERACS and
networks, together with other Critical Infrastructure (CI) systems, EDSA for power networks [6], EPANET for water supply systems [7],
such as power, water, transportation (road, railway) and commu- and FLO-2D for natural storm water systems [8]. Most of these tools
nication networks, are lifelines essential to the continued well- focus only on one network and rarely allow an analysis of multiple
being of a modern community. These CI systems are charac- interdependent lifelines, despite the strong interactions existing
terised by multiple mutual interactions. Gas networks have strong between systems of modern infrastructures. Further, commercial
links especially with electric power networks and buildings. An tools are closed programs used as black boxes, and cannot be
appropriate analysis of a gas distribution network is crucial when expanded or modified to meet the increasing needs of researchers.
the analysis scope is not just this network, but the entire set of The goal of this paper is to provide a complete formulation of the
interacting CI systems and buildings of an urban area [3]. steady-state flow analysis of a gas distribution network, thus
As for other networks, the behaviour of a gas network can be providing an important contribution to this field. The algorithm
represented in terms of topological, connectivity analysis or of a was implemented within an open source simulation tool for civil
more refined flow analysis; the latter includes connectivity and infrastructures, namely Object-Oriented Framework for Infra-
involves computation of the system’s operational state, in terms of 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
E-mail address: francesco.cavalieri@uniroma1.it.
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. medium (75 mbare7 bar) and low (<75 mbar) pressures [17]. In
Osiadacz and Pienkosz [11] provide the more detailed flow this paper, the focus is only on distribution (both low and medium
computation algorithm for gas systems, which, however, has limi- pressure) networks.
tations that will be highlighted in the following. An interesting A gas distribution network is a graph made up of nodes and
document on the subject is the thesis by Hwee [12], presenting a edges. The graph is typically undirected (an edge between nodes i
number of flow equations in use for gas networks, as well as simple and j implies the existence of the opposite one, between j and i),
examples that were used here to validate the new proposed algo- since the flow direction is not known a priori (i.e., before running a
rithm. Abeysekera et al. [13] investigate the impact of using flow analysis) and flow in pipes is allowed in both directions. Nodes
different gas supply sources on pressure distribution and gas may be stations, demand (or load) nodes (where gas is delivered to
quality within a network, making use of a flow analysis along the end-users, typically at low pressure) and joints.
lines of Hwee [12]. A further paper focussing on the subject is that Three main types of stations may exist [15]:
by Morais and Lima [14]. Other papers found in the literature are
related to specific aspects of the flow formulation, and will be 1 Metering/Reduction (M/R) stations, which contain metering
referred to within the following sections. equipment for monitoring gas flow, as well as pressure reduc-
To the best of the author’s knowledge, all methods available in tion equipment. In M/R stations gas is derived from the high
the current literature on flow computation for gas networks are pressure transmission network and injected into the medium
limited to a single pressure level. On the other hand, multiple levels pressure distribution network, for which such nodes function as
in other lifelines are treated differently from gas networks: for source nodes;
instance, the presence of multiple voltage levels in electric power 2 Reduction Groups (RGs), which are stations operating pressure
networks is handled with specific equations for transformers, and reduction between medium and low pressure, or between two
the methodology cannot be used for pressure reduction in gas medium or low pressure levels. RGs may also serve as load
systems. The formulation presented in this work fills one gap in the nodes and deliver gas at medium pressure to industrial users,
literature, allowing for the computation of gas flows in the presence and in this case are called RGMs (Reduction Groups and
of multiple pressure levels. This feature is of paramount importance Metering);
in the analysis of real gas systems and specifically in the treatment 3 Metering stations, which are only flow measurement points.
of assessment problems, where a network cannot be decomposed
in pressure-homogeneous portions to be solved independently. The edges between nodes are buried pipelines. The most com-
This advancement represents the main novelty of this work. mon employed materials are steel, polyethylene (high-density,
The second important novelty stands in the application of two HDPE, or low-density, LDPE), ductile iron and copper [17].
features of the methodology, namely the correction for elevation
change in pipes and the pressure-driven formulation, which were 3. Steady-state flow formulation
found in the literature but never applied to case studies. Pressure-
driven formulation, analogue to the head-driven formulation of The topology of a graph with n nodes and m edges is described
water supply networks, is essential in a network assessment by the n m node-edge incidence matrix, A, such that:
problem where, contrary to the situation of a design problem, there
8
is no assurance that demands will be satisfied and allowance must < þ1; if the flow in edge j leaves node i
be made for gas not reaching an end-user due to excessive Aij ¼ 1; if the flow in edge j enters node i (1)
depressurisation along the network; this is the case, for example, of :
0; if edge j is not incident to node i
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
3.1. Kirchhoff’s laws
non-trivial realistic gas network composed of 67 nodes and 88
edges, thus resulting to be computationally effective and hence
Kirchhoff’s first law is the principle of conservation of mass. It
usable within simulation schemes for risk assessment (see e.g.
states that the algebraic sum of flows in a network of edges meeting
Ref. [16]).
at a point is zero, or, equivalently, at any node in a graph, the sum of
The remaining sections of this article are organised as follows. In
ingoing flows equals the sum of outgoing flows. By defining n2 as
Section 2, the taxonomy of gas distribution networks is briefly
the number of source nodes in the network, it is possible to write
outlined, limited to the aspects of interest for the paper. Section 3
Kirchhoff’s first law in the following matrix form [11]:
presents the steady-state flow formulation, up to the definition of
the error function to be minimised in order to obtain the solution. A1 $q ¼ Q (2)
In Section 4, the implemented formulation is validated and applied
to four examples of increasing complexity, through which all the where A1 is the (n-n2) m matrix whose rows are related to load
discussed features are highlighted. Conclusions and future work are nodes (and thus obtained by deleting the rows of A related to
reported in Section 5. 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
2. Taxonomy of gas networks demands).
Kirchhoff’s second law is the principle of conservation of energy.
Gas is supplied across a country through several networks It states that the pressure drop around any closed loop is zero. This
operating at different pressures. A national transmission system law can be expressed in the following matrix form [11]:
conveys gas from the sources to the different regions of the country,
where it is passed to regional transmission systems operating at AT $P ¼ DP (3)
lower pressures. For local conveyance of gas to end-users, the dis-
tribution system is used. The pressure level makes the main dif- where P is a n 1 vector whose generic element, Pi (i.e., the one
ference between transmission and distribution systems. The former related to node i), equals the pressure at node i, pi, for low pressure
operate at high pressures (>7 bar), while the latter operate at nodes, and the pressure at node i squared, p2i , for medium pressure
F. Cavalieri / Energy 121 (2017) 781e791 783
nodes. Similarly, the m 1 vector DP contains the pressure drops (bar) are the gas temperature and pressure, respectively, in stan-
between nodes i and j, (pi e pj), for low pressure nodes, and the dard conditions (Standard Temperature and Pressure, STP), while
pressure drops squared, (p2i e p2j ), for medium pressure nodes. The for the friction factor f the following expression is used:
first term of Eq. (3), AT·P, can be partitioned to separate source sffiffiffi
nodes from load nodes, as: 1 S$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 $D4:848
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
KM
3.2. Common flow equations for pipes
where pi and pj are the pressures (bar) at pipe start and end nodes, function of an elevation adjustment parameter, s, defined as [18]:
respectively, D (mm) is the pipe diameter, L (m) is the pipe length, S
(non-dimensional) is the specific gravity of gas and f (non-dimen- Hj Hi
s ¼ 0:0684$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 F. Cavalieri / Energy 121 (2017) 781e791
A1 $q ¼ Q ðP 1 Þ
independently. The levels can be all in one pressure range (low or (20)
AT $P ¼ fðqÞ
medium) or even spread in both low and medium pressure ranges.
For example, an analyst can perform a flow analysis on a network is composed of (n-n2) continuity equations, one for each load node,
that is driven by two medium pressure levels and one low pressure and m resistance equations, one for each edge. The unknowns are
level. This feature represents the main novelty of this work. the (n-n2) pressures at load nodes and the m flows in the edges. The
In the present formulation, pressure can be only reduced. system is nonlinear in the resistance equations, so that its solution
However, inclusion of pressure increase (e.g., with the presence of must be retrieved by the use of iterative algorithms and the se-
compressor stations) would be straightforward. Pressure is reduced lection of a proper initial guess, X0.
through RG and RGM stations, which are characterised by an input The latter can be partitioned in the two sub-vectors of initial
pressure pin (the highest) and an output pressure pout (the lowest), pressures and initial flows, respectively:
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- P 1;0
X0 ¼ (21)
cerning the pressure reduction they are treated in the same way. In q0
particular, each station node, identified by a node number N, is split
in two nodes, sharing the same position. The first and second nodes which are then established in a sequential manner. The vector of
are assigned the highest (pin) and lowest (pout) pressures, respec- initial pressures, P1,0, contains the nominal pressures (squared for
tively. The second node is identified by the next node number, Nþ1, medium pressure nodes). The latter are assigned starting from the
in the node list. For RGM nodes, the load is kept applied to the first nodes connected to the ones where pressure is known (i.e., sources
node. Between the two nodes a fictitious zero-length edge, called and RG/RGM stations) and then reaching, following an iterative
reduction link, is then added, to preserve the graph connectivity; no process, further nodes that are connected at least to one node
frictional loss takes place in such link, so that its length (zero) and where pressure was already assigned. As far as q0 is concerned,
diameter (which can be assigned arbitrarily) do not affect the Osiadacz and Pienkosz [11] propose to retrieve it by partitioning the
results. matrix A1 in two sub-matrices, referred to as the tree and co-tree
The input and output pressures, pin and pout, for the generic parts. The tree part, having dimensions (n-n2) (n-n2), contains
reduction link are arranged into the variable PR (Pressure Ratio), the columns of A1 related to the tree edges, which are those edges
defined as: 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)
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)
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.
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)
(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.
Table 7
Example 3 with RGM node duplicated: edge data.
# Start node End Node Type pin (bar) pout (bar) L (m) Le (m)
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.
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 791