ARTICLE IN PRESS

Energy 34 (2009) 235– 244

Contents lists available at ScienceDirect

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

Reliability modeling of multi-carrier energy systems

Gaudenz Koeppel , Göran Andersson
Power Systems Laboratory, Swiss Federal Institute of Technology ETH Zurich, ETL G28, 8092 Zurich, Switzerland

a r t i c l e in fo abstract

Article history: Most of today’s consumed energy can be categorized as either electrical, chemical or thermal.
Received 3 July 2007 Traditionally, the infrastructures for the supply with these energy forms have been engineered separately,
Available online 18 June 2008 often resulting in parallel but uncoordinated supply infrastructures. Various new technologies for
Keywords: conversion between these energy forms however start introducing a mutual dependence and an
Reliability increasing substitution potential between these infrastructures, with combined cycle gas turbines playing
Multi-carrier systems a major role. To a certain extent, the mutual dependence also leads to redundancy effects and influences
Markov model the reliability and availability of supply. This paper presents a model for analyzing and calculating
expected reliability of supply and expected energy not supplied in such multi-carrier energy systems. The
models can be used to identify the benefit and sensitivity of certain conversions as well as their limits in
terms of reliability improvements.
& 2008 Elsevier Ltd. All rights reserved.

1. Introduction Altogether, it is likely that load peaks from the electrical network
will be migrated to the gas network, resulting in a more intensively
Traditionally, the infrastructures for the supply of electrical, and differently used gas network. Similar relations and dependen-
chemical and thermal energy have been treated separately. In cies can be found between thermal networks and electrical or
most cases, both the topologies and the operation strategies have chemical networks when considering e.g. hot water boilers or
been developed independently of the others and in some regions combined heat and power (CHP) gas-fired power stations.
or cities, they are also maintained by different companies. Various Consequently, investigations concerning new and future
developments, however, indicate an increasing mutual depen- energy systems should integrate electrical as well as chemical
dence and competition between these three infrastructures. and thermal energy. The usefulness of such a combined or
This trend is particularly driven by the increasing number of integrated analysis has already been recognized in a few recent
gas-fired power stations, converting natural gas into electricity publications [2–7], focusing mostly on energy flow optimization
and thermal energy [1]. Different technologies exist, both for aspects and on modeling similarities. However, besides potential
small- and large-scale applications, indicating possible inter- economic benefits from a combined power flow optimization, a
dependencies of the energy carriers on transmission as well as on major benefit can be identified in the area of reliability of supply
distribution levels. [8]. A gas-fired power station or an electrical boiler not only
This raises the question if and how in coming years the introduces an interdependence between two energy systems; the
customer supply with electrical, chemical (i.e. gas) and thermal coupling established by the converter makes it possible to supply
energy will change. As gas-fired power stations establish a a load from several independent infrastructures, thus influencing
connection between the electrical and chemical network, a certain reliability of supply.
interchangeability as well as a certain redundancy are being This paper presents and discusses a method, which has been
introduced into the system. At certain times it might be financially developed for the analysis of the reliability of supply in multi-
attractive to generate electricity from gas instead of consuming carrier energy systems [8,9]. The method allows to determine the
directly from the electrical network, e.g. during peak hours. Or reliability of supply as a function of the reliability characteristics
there could be situations, where it is favorable to make use of the of the existing conversions. It will be used to investigate potential
short-term storage flexibility, inherent in the gas network. benefits of the mentioned redundancy effects and to what extent
they influence the availability of supply. Furthermore, the method
is applied to analyze whether a certain overall reliability of supply
 Corresponding author. Tel.: +41 44 632 4188; fax: +41 44 632 1252. can be maintained despite less favorable reliability characteristics
E-mail addresses: g.koeppel@alumni.ethz.ch (G. Koeppel), of particular components. Such investigations are important both
andersson@eeh.ee.ethz.ch (G. Andersson). for maintenance considerations—assuming the maintenance

0360-5442/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.energy.2008.04.012
 ARTICLE IN PRESS

236 G. Koeppel, G. Andersson / Energy 34 (2009) 235–244

Nomenclature L load matrix

lab failure rate of the connection from a to b
e; c; t subscript denoting electrical, chemical and thermal K failure rate matrix
energy, respectively mab repair rate of the connection from a to b
a; b variables for the input and output energy carrier, M repair rate matrix
respectively; a; b 2 ½e; c; t ~
Pb state probability vector for output b
Aab transition rate matrix for the connection from a to b Pib probability of state i of state probability vector ~
Pb
Ab transition rate matrix for the supply of b P state probability matrix
cab coupling factor, coupling a with b Q ab probability of a failed connection from a to b
C coupling matrix Rab probability of an operating connection from a to b
EENS expected energy not supplied S state capacity matrix
Zab efficiency of the connection from a to b t time step
F factor matrix, boolean V matrix containing momentary capacity levels
G factor matrix, boolean, F þ G ¼ I W weighting matrix
Gab connection from a to b zab capacity of the connection Gab
~
Lb load vector at output b fb connection capacity matrix for the supply of b
Lb momentary load demand at output b f connection capacity matrix
~
L momentary load vector, representing the momentary
hub loads

frequency to affect the reliability characteristics—and for discus- The energy hub modeling concept has been developed
sions concerning the benefit of distributed generation. primarily for optimization of the involved energy flows. For this
The following section introduces the energy hub concept, a purpose, the so-called coupling matrix C can be defined, relating
general modeling approach suited for multi-carrier energy each input of the energy hub with each output. This matrix is
systems. Based on this concept, a method for reliability analysis shown in Fig. 2, corresponding to Fig. 1. The input energy carriers
in multi-carrier energy systems is then outlined in Section 3, are assigned to the columns and the output carriers to the rows.
constituting the main contribution of this paper. In Section 4 this The values at the intersections are denoted as coupling factors
method is applied to both demonstrate its applicability as well as cab , coupling the energy carrier a with the carrier b. Generally, the
to analyze sensitivities. Possible applications of the method are coupling factors correspond to the respective conversion effi-
discussed in Section 5. ciency Zab. However, depending on the exact hub structure, it
might be necessary to introduce further factors because of energy
conservation laws [10]. Defining both the input and output power
2. Modeling approach flows as vectors allows to calculate the output flows as the
product of the coupling matrix with the input flows [11]. This
From a system point of view, a unit like a gas-fired power relation can then be used for performing power flow analysis with
station or an electric boiler is simply a converter, converting one given networks, optimizing e.g. the consumption depending on
form of energy into one or several other forms of energy. In this the installed converters and price information. It also allows
context it does not matter, with which technology a conversion is optimizing matrix elements for given loads and power flows, i.e.
performed but simply that a conversion can take place. In order to performing a structural or topological optimization [3].
have a simple formulation for these conversions, the so-called
energy hub modeling concept has been developed [3,10]. With
this approach, any network participant can be reduced to and
represented by the energy conversions it is performing.
The general form of an energy hub is a multi-port, with several 3. Reliability modeling
inputs and several outputs, representing the energy carriers.
Between the ports, the different energy carriers are converted into The following paragraphs introduce and discuss a method for
each other, depending on the available converters. Fig. 1 shows the calculating expected reliability of supply and expected energy not
general two-port form of an energy hub with all theoretically supplied (EENS) for a load, whose supply can be modeled as an
possible conversions displayed, whereas e.g. the expression ‘e–c’ energy hub. The method is intended to be used in matrix-based
represents the electrical–chemical conversion. simulation programs and is therefore defined accordingly.
electrical

chemical

thermal

e-c e-t
electrical electrical

c-e c-t cee cce cte

electrical
chemical chemical
chemical cec ccc ctc
t-c t-e
thermal thermal cet cct ctt
thermal

Fig. 1. Standard representation of a two-port energy hub providing conversions Fig. 2. Coupling matrix C for the energy hub from Fig. 1, providing conversions
between electrical, chemical and thermal energy. between electrical, chemical and thermal energy carriers.
 ARTICLE IN PRESS

G. Koeppel, G. Andersson / Energy 34 (2009) 235–244 237

3.1. Definitions 1
Γe
Γc
The general approach of the above introduced energy hub e t
Γt
concept can also be used for analyzing different aspects of
reliability of supply. For this purpose, each connection between
c
two energy carriers a and b ða; b 2 ½e; c; tÞ is represented with its
reliability characteristics instead of with its conversion efficiency. e c t
Each connection or converter is considered as one component.
2 3 4
Assuming that the operational behavior of an energy hub can be Γe × Γe Γe
described as a stationary Markov process [12] allows the Γc Γc × Γc
introduction of failure and repair rates for all components. Thus, Γt Γt Γt ×
analogous to the coupling matrix C, a failure rate matrix K and a
repair rate matrix M can be defined, containing the failure and c e t c e
t
repair rates, respectively: e c
c t e t
2 3 2 3
lee lce lte mee mce mte 5 6 7
6 7 6 7 Γe × Γe Γe ×
K ¼ 4 lec lcc ltc 5; M ¼ 4 mec mcc mtc 5 (1)
Γc × Γc × Γc
let lct ltt met mct mtt
Γt Γt × Γt ×
Sometimes, not all possible couplings will be occupied. These
particular connections are defined to have a failure rate l ¼ 0 and t e c
a repair rate m ¼ 1. These values have no physical meaning, but
e
they are necessary for the algorithm to perform properly, as will
be seen later. The probability of an operating conversion from 8
Γe ×
energy carrier a to b is denoted as Rab and the probability of a t c
Γc ×
failed conversion as Q ab . They can be defined as follows [12]:
Γt ×
mab lab
Rab ¼ ; Q ab ¼ (2)
mab þ lab mab þ lab Fig. 3. State space diagram for the supply of output Lb from inputs Ie , Ic and It
through the connections Geb , Gcb and Gtb .
It is important to note here that Eq. (2) only considers components
covered by the energy hub in proper. This allows to focus on the
Whether a state is a successful state (i.e. the load is supplied) or
energy hub and the respective converters. If necessary, the
not, depends both on the capacity zab of each supply path Gab as
reliability characteristics of the supplying infrastructures can be
well as on the momentary load Lb . The supply capacity of a state
incorporated by assuming them to form a series system with the
corresponds to the sum of the supply capacities of all components
converters they are supplying [9]. Furthermore, single connections
designated operating in that particular state. As long as the load
inside the hub, e.g. between a converter’s output and the receiving
level Lb is below a state’s supply capacity, the state is considered
network, are assumed to be 100% reliable. This is a reasonable
successful and contributes to the probability of supply. For
assumption if the physical dimensions covered by the hub are
example the supply capacity of state 2 in Fig. 3, with Gcb þ Gtb
small; the energy hub is merely a node and the depicted lines in
operating and Geb failed, equals zcb þ ztb and if the load Lb is above
Fig. 1 are only of illustrative help.
this capacity, the state is considered failed. Thus, although two out
of three components are operating, the load cannot be supplied
3.2. Calculation approach and the probability of state 2 would not contribute to the
reliability of supply, defined as the sum of the probabilities of
According to Fig. 1, each of the three outputs can theoretically all successful states [12].
be supplied from all three networks connected at the hub input. Therefore, it is necessary to also keep track of the supply
These connections are either direct or indirect and sum up to a capacities of all connections and conversions. The connection
total of nine possible connections between input and output. Each capacity matrix f is defined analogous to the failure rate and
output (i.e. the electrical, the chemical and the thermal load) is repair rate matrices as
thus supplied through three supply paths, each of which can be 2 3
operating or non-operating. This allows to model the supply of zee zce zte
6 7
every output with a state space diagram with eight states, f ¼ 4 zec zcc ztc 5 (4)
representing all possible combinations. Fig. 3 shows the state zet zct ztt
space diagram for output Lb , representing any of the three possible
p A converter with several outputs, such as a CHP plant, is most
outputs or loads. The symbol ‘ ’ represents an operating
likely modeled with equal failure rates and repair rates, i.e. lce ¼
connection and ‘’ a failed connection. For example the connec-
lct and mce ¼ mct . The connection capacities are, however, not
tion supplying output b from the chemical network is designated
necessarily identical; the thermal capacity of a CHP usually differs
as Gcb .
considerably from the electrical capacity, i.e. zce azct . These
The probability of residing in one of the states corresponds to
assumptions have to be validated for each technology considered,
the product of the probabilities of the individual components’
particularly for technologies where converters exhibit mutual
state, according to Eq. (2). The probability of e.g. residing in state 6
dependencies.
thus corresponds to the probability of connection Geb being
operating and connections Gcb and Gtb being failed:
3.3. Expected reliability of supply
P 6b ¼ Reb  Q cb  Q tb
meb lcb ltb
¼ (3) The expected reliability of supply for one hub output equals
ðmeb þ leb Þðmcb þ lcb Þðmtb þ ltb Þ the probability of residing in any of the successful states, i.e. the
 ARTICLE IN PRESS

238 G. Koeppel, G. Andersson / Energy 34 (2009) 235–244

sum of the probabilities of all successful states. The simultaneous e

e1 Operating e2 Non-operating
calculation for each output is outlined in this section, focusing on
e → e →
one load level for each output, given as the row vector ~ L: e
~
L ¼ ½Le Lc Lt  (5) c
c1 Operating c2 Non-operating
The supply of each of these three loads can be modeled according c → c →
to Fig. 3. The individual state probabilities can be calculated as c
defined in the context of Eq. (3) and combined in the 8  3 state
probability matrix P: t1 t
Operating t2 Non-operating
2 3 t → t →
P1e P 1c P1t t
6 7
6 P2e P2e P2t 7
6 7
6 P3e P 3c P3t 7 Fig. 4. Individual state space diagrams for the three supply paths for one energy
6 7 hub output b.
6 7
6 P4e P 4c P4t 7
~ ~ ~ 6
P ¼ ½P e P c P t  ¼ 6 7 (6)
7
6 P5e P 5c P5t 7
6 7
6 P6e P 6c P6t 7
6 7 connections. According to Fig. 1, each output is supplied with a
6 P7e P 7c P7t 7
4 5
direct connection and two indirect connections, i.e. through
P8e P 8c P8t
converters from the other energy carriers. The demand at the
The supply capacities of the individual states can be summarized hub input, i.e. the demand from the network, thus corresponds to
analogously in the state capacity matrix S: the sum of the direct load and the loads which are supplied
2 3 indirectly; e.g. the demand from the electrical network equals the
zee þ zce þ zte zec þ zcc þ ztc zet þ zct þ ztt electrical load Le plus the load supplied through the electrical–
6 zce þ zte zcc þ ztc zct þ ztt 7
6 7 chemical converter Gec and through the electrical–thermal
6 7
6 zee þ zte zec þ ztc zet þ ztt 7 converter Get . If this cumulated demand exceeds the capacity of
6 7
6 z þz zec þ zcc zet þ zct 7 the supplying network, a dispatch or priority rule has to be
6 ee ce 7
S¼6 7 (7) defined. The most simple rule would be to give priority to the
6 zte ztc ztt 7
6 7 directly connected load and to evenly distribute the remaining
6 zee zec zet 7
6 7
6 7 supply margin to the remaining converters, as discussed in detail
4 zce zcc zct 5
in [9] and applied in one example in Section 4. However, different
0 0 0 dispatch policies are feasible, depending on the purpose of an
As outlined above, a certain state is only considered successful energy hub, and their discussion would exceed the purpose of this
and contributes to reliability of supply if its capacity is equal to or paper.
beyond the load demand Lb . For this purpose, the 8  3 factor In any case, both the supply capacity matrix f and the factor
matrix F can be defined, containing boolean variables that matrix F have to be redefined for every discrete time step.
designate whether a state is supplying or not: Assuming the three supply systems to be independent of each
other allows the application of Kronecker products and sums
Fij ¼ ~
Lj pSij ; i 2 ½1; . . . ; 8; j 2 ½1; 2; 3 (8) [9,13]. This considerably facilitates the calculation of P and
For example the factor for state 3 of the chemical output, particularly of S, which has to be redefined in every time step as
according to Fig. 3, is defined as follows, whereas j represents well. The failure rates and repair rates of the connections and
the electrical, chemical and thermal output, respectively: converters are not time-dependent and therefore the state
( probabilities only have to be calculated once. However, due to
1; ~ L2 pS32 ¼ ðzec þ ztc Þ both the varying loads and the varying converter capacities it is
F32 ¼ necessary to determine the successful states, i.e. matrix F in every
0; ~ L2 4ðzec þ ztc Þ
time step.
The probability of supply for each output is then defined as the Hence, the supply of a hub output b depends on three supply
diagonal entries of the product of the transposed factor matrix F systems whose individual state space diagrams can be obtained as
and the state probability matrix P: illustrated in Fig. 4.
2 3 2 T 3 The states e1, c1 and t1 are the operating states for each
P eout ðF  PÞ11
connection. The transition rate matrix Aeb and the steady-state
~ 6 7 6 ðFT  PÞ 7
P out ¼ 4 Pcout 5 ¼ 6
4
7
22 5 (9) probability vector ~ P eb for the e ! b connection are given as [12]
Ptout ðFT  PÞ33 " #
leb leb
Eqs. (5)–(9) represent the general calculation approach, i.e. to ~
Peb  Aeb ¼ ½P e1 P e2   ¼0 (11)
meb meb
identify the successful and supplying states and to add their
individual probabilities to give the total probability of successful
The transition rate matrix and the associated state proba-
supply. This approach can now be generalized further by allowing
bility vector for the c ! b and t ! b connections are defined
loads that vary with time. Accordingly, each output is represented
analogously.
with a load vector, containing discrete load levels. Hence, the load
Following the Kronecker approach,1 the transition rate matrix
vector ~L is extended to the n  3 load matrix L, column-wise
Ab , representing the combined state space diagram, is given as2
containing the loads for every time step t of the observation
period of length n with t 2 ½1; . . . ; n:
Ab ¼ Aeb  Acb  Atb (12)
L ¼ ½~
Le ~
Lc ~
Lt  (10)
With varying loads it must also be considered that the supply 1
See Appendix A for explanation of the Kronecker formalism.
capacities could vary, depending on the priority of the individual 2
A Kronecker sum is represented with  and a Kronecker product with .
 ARTICLE IN PRESS

G. Koeppel, G. Andersson / Energy 34 (2009) 235–244 239

The corresponding state probability vector ~

Pb is given accordingly as changing e.g. daily. The justification for doing so can be explained
as follows: all existing converters are assumed to operate
~
Pb ¼ ~
P eb  ~
P cb  ~
Ptb (13) constantly and thus to have reached steady-state conditions. Only
Thus, instead of solving the combined system, the state space their operating points are changing, depending on the different
diagram of each component is solved separately, according to load demands. In other words, the reliability performance is
Eq. (11). The resulting state probability vectors ~ P ab are then calculated for different steady-state load points and the resulting
combined as in Eq. (13), resulting in ~
Pb . reliability values are then summarized according to the relative
The state capacity matrix S can be found as well by using occurrence frequency of each load point during a certain time
Kronecker sums. For each state space diagram in Fig. 4, a matrix interval.
fab can be defined, containing the operating and available capacity
zab of each state in the corresponding diagonal element. With e.g. 3.4. Expected energy not supplied
zeb denoting the power rating of the connection from electrical to
energy carrier b, this results in the matrices: Sometimes it is also important to know the amount of energy
" #
zeb 0 that could not be supplied because of an outage, i.e. the EENS.
feb ¼ (14) A converter might not be able to supply the full load but in case of
0 0
" # an outage of the main supplying connection, the converter can
zcb 0
fcb ¼ (15) still supply a certain load. Thus, the here presented algorithm
0 0 calculates EENS, incorporating load shedding.
" #
ztb 0 For the calculation of EENS, it is not only necessary to know
ftb ¼ (16) which states are successful states but also which the actual supply
0 0
level of each state is. For instance, if the load equals 7 kW and the
Hence, fcb11 corresponds to state c1 of the connection c ! b, with maximum supply capacity of a state is 10 kW, EENS for that state
a power of maximum zcb . These three matrices can now be has to be calculated considering the actual load of 7 kW and not the
combined using Kronecker sums to give the 8  8 matrix fb : suppliable load of 10 kW. To consider this, a second factor matrix G,
fb ¼ feb  fcb  ftb (17) adding up to unity with F, has to be defined, with t designating a
time step, i the different states and j the energy carrier:
The entries of the state capacity matrix S are then just defined as
Gijt ¼ 1  Fijt (21)
the diagonal elements of matrices fe , fc and ft :
2 3 The states’ supply capacities relevant for the calculation of EENS
fe11 fc11 ft11
6f 7 can now be identified with matrix V:
6 e22 fc22 ft22 7
6 7
S¼6 . .. .. 7 (18) Vijt ¼ Gijt  Sijt þ Fijt  Ltj (22)
6 .. . . 7
4 5
fe88 fc88 ft88 Thus, if a state is a successful state, i.e. its supply capacity exceeds
the load demand, the corresponding entry in matrix V equals the
The state number definition of both P, S and f is identical, they, load demand. Otherwise, its entry in V corresponds to its installed
however, do not fully correspond with the state number definition capacity zab, as this equals the load level that could be maintained
chosen in Fig. 3. The concepts outlined in Eqs. (11)–(18) can now in case of an outage situation.
be applied with the general modeling approach given in The probabilities of each state in the state space diagram are
Eqs. (5)–(9). However, in Eq. (9) the reliability is calculated for conditional probabilities. For instance the probability P5e corre-
one time step only. In order to determine the average reliability of sponds to an operating thermal–electrical connection given that
supply during the total investigated period of length n, it is thus both the electrical–electrical and the chemical–electrical connec-
necessary to introduce the 8  3 weighting matrix W: tion have failed. Consequently, EENS—incorporating load shed-
2 Pn Pn Pn 3 ding—corresponds to the difference between the load demand L
t¼1 Fð1; 1; tÞ t¼1 Fð1; 2; tÞ t¼1 Fð1; 3; tÞ
P
6 n Pn P n 7 and the product of V and the individual state probabilities P. In
Fð2; 1; tÞ t¼1 Fð2; 2; tÞ t¼1 Fð2; 3; tÞ 7
166 t¼1 7 order to express EENS in (Wh), it is necessary to apply the time
W¼ 6 .. .. .. 7 (19)
n6 . . . 7 interval duration Dt:
4P Pn Pn 5
n
2 3 2 Pn P8 3
t¼1 Fð8; 1; tÞ t¼1 Fð8; 2; tÞ t¼1 Fð8; 3; tÞ EENSe t¼1 ðL t1  i¼1 V i1t  Pi1 Þ
!
6 7 6 P n P 8 7
Matrix W contains the average contribution of each state to the EENS ¼ 4 EENSc 5 ¼ Dt  6 4 t¼1 ðLt2  i¼1 Vi2t  Pi2 Þ 5
7 (23)
Pn P8
overall probability of supply, as stored in the factor matrix F. As EENSt
t¼1 ðL t3  i¼1 V i3t  Pi3 Þ
mentioned, matrix F is now defined for each time step, in order to
represent the successfully supplying states depending on the Both the calculation of EENS and of the expected reliability of
momentary load level in each time step t. supply will be demonstrated in the following section, which
The average reliability of supply at the three hub outputs is contains two illustrative examples.
now found by multiplying the average weight of each state, stored
in matrix W, with its probability, stored in the state probability
4. Application examples
matrix P, and by summing the products. Analogous to Eq. (9), the
resulting diagonal elements correspond to the probabilities of
supply at each hub output: This section contains two application examples. The first
example focuses on the reliability of supply at one hub output
2 3 2 3
P eout ðWT  PÞ11 only and illustrates its sensitivity to the existing conversions’
6 7 6 T 7
~
P out ¼ 4 Pcout 5 ¼ 6 7 characteristics. The second example demonstrates the application
4 ðW  PÞ22 5 (20)
Ptout T of the presented models for a complete energy hub and illustrates
ðW  PÞ33
the mutual influencing of the energy carriers through the
At first sight, it might be a contradiction to apply steady-state converters. Both examples are constructed to mainly show the
failure and repair rates together with a load profile that is application of the methodology and possible analyses.
 ARTICLE IN PRESS

240 G. Koeppel, G. Andersson / Energy 34 (2009) 235–244

4.1. Calculation and sensitivity examples supplying as well (compare with Fig. 3). Then, during time steps
76–78, i.e. from 19:00 to 19:45, additionally state 2 is a supplying
In Fig. 5 the German standard weekday electrical load profile state: zee is down, but zce þ zte can supply the load. After time step
for a small business is shown, scaled to a total annual consump- 78, state 7 also becomes a supplying state; the load has decreased
tion of 20 MWh [14]. The load curve consists of n ¼ 96 intervals, further such that zce can maintain the redundant supply on its
each with Dt ¼ 15 min duration. own, not anymore requiring zte . Summing up matrix Fe gives the
The load can be supplied both through a direct electrical–electrical relative frequency of each state, as defined in Eq. (19):
connection, rated zee ¼ 10 kW, as well as with indirect connections  
through a chemical–electrical converter, rated zce ¼ 2 kW, and a W~ T ¼ 1 52 1 1 0 1 47 0
e
96 96
thermal–electrical converter, rated zte ¼ 0:5 kW. The failure and
repair rate matrices are defined as The availability of supply at the electrical output is finally found
2 3 2 3 according to Eq. (20):
0:5 1:5 1:5 4380 365 365
6 7 6 7 ~T ~
Peout ¼ W
K¼4 0 0 0 5; M ¼ 4 1 1 1 5 e P e ¼ 0:999947
0 0 0 1 1 1 The availability of supply without the energy hub, i.e. considering
Thus, the chemical–electrical and the thermal–electrical converter fail only the electrical supply, equals P ee ¼ 0:999886. This corresponds
three times as much as the electrical–electrical connection. Further- to an annual outage duration of 59.99 min. With the energy hub,
more, their repair needs 12 times as long, i.e. 24 h. Both the i.e. considering the converters from the chemical and the thermal
connection capacities zce and zte as well as the failure and repair networks, the expected annual outage duration is reduced to
rates are educated guesses [15–17] and chosen such that they help to 27.64 min.
clearly identify their influence in the example.
Using Eq. (2) and the matrix entries (e.g. lce corresponds to 4.1.1. Influence of the converter failure rates
K12 ), the state probability vector ~P e can be found, following the This section investigates the relation between the reliability of
state definition from Fig. 3: supply and the failure rates of the converters from the chemical
2 3 and the thermal networks. The question addressed is how
0:99171800206280 unfavorable the reliability characteristics could be, with the
6 0:00011320981759 7 converters still improving the reliability of supply. Fig. 6 shows
6 7
6 7
6 0:00407555343313 7 the expected annual outage duration as a function of the two
6 7
6 0:00407555343313 7 converters’ failure rates.
~ 6 7
Pe ¼ 6 7 The general sensitivity is visible, the larger capacity of the
6 0:00000046524583 7
6 7 chemical–electrical converter being responsible for its larger
6 0:00001674884973 7
6 7 influence. The expected outage duration without energy hub
6 7
4 0:00000046524583 5
was calculated to be roughly 1 h. Hence, even if the chemical–
0:00000000191197 electrical converter would fail once a week, the outage duration
It is obvious from Fig. 5 that the converters (zce and zte ) only could be reduced to roughly half than before. This result is
contribute during periods where Le ðtÞp2:5 kW. Furthermore, the interesting in so far as both converters only contribute to the
thermal–electrical converter is not powerful enough to supply the reliability of supply during lower load hours. This analysis also
load by itself at all. An excerpt of matrix Fe shows the different raises the question whether the failure rate of the direct
state probability weights between 18:45 and 20:00: connection could be allowed to increase because of the additional
2 3 supply paths. The next subsection concerns this aspect.
        
6 1 0 1 1 0 1 0 0 7 t ¼ 75
6 7 4.1.2. Bandwidth for the failure rate of the direct connection
6 7
6 1 1 1 1 0 1 0 0 7 t ¼ 76 Distributed generation is sometimes claimed to allow for
6 7
6 7
FTe ¼ 6 1 1 1 1 0 1 0 0 7 t ¼ 77 postponing network enhancements or to even make them super-
6 7
6 1 1 1 1 0 1 0 0 7 t ¼ 78 fluous [18]. The assumption behind this claim is that potentially
6 7
6 7 existing distribution network congestions can be alleviated by
4 1 1 1 1 0 1 1 0 5 t ¼ 79
locally generating a certain amount of power. This results in a
        
reduction of the demand from the network and hence, the
This excerpt shows that state 1 is always a supplying state. network enhancement is not necessary anymore or at least not for
Furthermore, as long as Le ðtÞ42:5 kW, states 3, 4 and 6 are the moment. In this context it is interesting to analyze to what

4.5
4.0 Le ()
3.5
3.0
Power, kW

ζce + ζte
2.5
ζce
2.0
1.5
t = 30
t = 31

t = 76
t = 78

1.0
ζte
0.5

04:00 08:00 12:00 16:00 20:00 24:00

time of the day
Fig. 5. Load curve of an electrical load Le supplied with an energy hub.
 ARTICLE IN PRESS

G. Koeppel, G. Andersson / Energy 34 (2009) 235–244 241

46

Outage duration, min

42 ce

38

34
te
30

26
50 100 150 200 250 300 350
Failures per year

Fig. 6. Expected outage duration in minutes for different failure rates of the chemical–electrical and the thermal–electrical converter.

20
Failure rate λee, f/yr

16

12

0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Rating ζce of the chemical-electrical converter Γce, [kW]

Fig. 7. Maximum allowable failure rate for the direct electrical connection for an increasing rating of the chemical–electrical converter and a targeted annual outage
duration of 1 h.

extent a higher failure rate of the electrical connection can be The conversion efficiency for all converters is assumed to be Zab ¼
allowed, because of the converters from the other networks. A 1 and the individual rated converter capacities are given in kW
higher failure rate could be due to e.g. reduced maintenance following the definition in Eq. (4):
intervals. Without these converters and the storage device, the 2 3
10 2 1
expected annual outage duration was found to be almost 1 h. The 6 7
question thus is if the additional converters allow to actually have f¼4 2 8 6 5
a higher failure rate of the electrical connection without risking a 5 5 15
annual outage duration higher than 1 h. Fig. 7 shows the According to Fig. 1, matrix f can be illustrated as displayed in
maximum allowable failure rate lee for the direct connection for Fig. 9. Besides the converters’ capacities, the maximum inlet
different capacities of the chemical–electrical converter. capacities of the individual hub inputs are indicated. These values
The figure shows that the failure rate of the electrical can be arbitrary; for the example they have been chosen to be
connection could increase with an increasing chemical–electrical identical with the capacities of the direct connections, i.e. the
converter capacity. Obviously, the more the chemical–electrical diagonal elements of matrix f. The maximum thermal inlet
converter can supply the load by its own, the higher the failure capacity thus equals 15 kW, which also corresponds to the
rate lee can be. As soon as the capacity zce exceeds the average load maximum capacity of the thermal connection inside the hub.
demand, the electrical connection’s allowable failure rate in- In the following reliability of supply and EENS are calculated
creases significantly; the chemical–electrical converter can now for each time step t 2 ½1; . . . ; 6, with Dt ¼ 0:5 h. The results are
be understood to be the main supplying system. given in Table 1, together with the corresponding values for the
complete time series. In contrast to the reliability of supply, EENS
4.2. Complete hub example corresponds to the sum of all values for EENS in each time step
and not to their average value. In addition to Table 1, Fig. 10 shows
In this example, the reliability of supply of a complete energy the available connection capacities for each time step. The chosen
hub is analyzed. At this point, it is irrelevant whether certain dispatch rule allocates all available converter capacities to support
connections are technically realizable or whether the connections a load at risk. If no load is at risk, the available capacities are
are rather theoretical. To illustrate the behavior of the method, six evenly allocated. The reason for this dispatch rule is as follows. If
segments of a load curve are analyzed, as displayed in Fig. 8. In no load is at risk, i.e. all loads can be supplied through the direct
addition, the figure shows the rated capacities of each direct connections, indirect connections are mainly relevant in case of an
connection, pointing out in which time segments supply through outage of a direct connection by still supplying the load or at least
indirect connections is crucial for maintaining the load supply. a part-load. Thus, the backup-like behavior of the indirect
The reliability characteristics of all hub components, i.e. of all connections influences EENS, with EENS being defined to consider
direct and indirect connections, are given according to Eq. (1), load shedding effects. Evenly allocating available converter
expressed in failures and repairs per year, respectively: capacities therefore results in reduced EENS for all loads.
2 3 2 3 However, if a load is at risk, i.e. the load demand exceeds the
1 5 5 730 365 365
6 7 6 7 supply capacity of the direct connection, supply through indirect
K ¼ 4 5 1 5 5; M ¼ 4 365 121 365 5 connections becomes crucial to maintain a full load supply. The
5 5 1 365 365 121 indirect connections are not anymore backup but they become
 ARTICLE IN PRESS

242 G. Koeppel, G. Andersson / Energy 34 (2009) 235–244

16
electrical load chemical load
14
ζtt

Power, kW
12
ζcc
10
8
ζee thermal load
6

1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Load curve segment Load curve segment Load curve segment

Fig. 8. Load profiles at the hub outputs and the capacities of the direct connections.

Table 1
Expected reliability of supply and expected energy not supplied (EENS) for each
2 5 time step and on average
el el
10
10 Time step Reliability of supply EENS (Wh)
2 5 Electrical Chemical Thermal Electrical Chemical Thermal
ch ch
8
8 t¼1 0.998632 0.999780 0.991803 3.438 0.305 26.778
6 t¼2 0.985137 0.991803 0.991803 13.022 16.504 49.180
1 t¼3 0.971824 0.991803 0.991803 20.066 24.590 57.377
th th
15 t¼4 0.971824 0.991803 0.991803 27.110 8.418 40.984
15
t¼5 0.998450 0.999890 0.991803 5.591 0.332 32.787
t¼6 0.998632 0.999780 0.991803 4.122 0.332 28.799
Fig. 9. Installed inlet, converter and connection capacities, in kW.

t ¼ 1...6 0.987417 0.995810 0.991803 73.351 50.482 235.905

necessary for load supply and all available converter capacities are
thus allocated to support the load at risk. The effect of the chosen
dispatch rule is visible in Fig. 10 and in Table 1 when comparing direct connections (Table 2), clearly identifies the benefits from
e.g. time steps 1 and 2. Clearly, a dispatch rule must comply with being interconnected through the converters. This outcome was to
the overall operation strategy [11] and therefore changes be expected as the indirect connections introduce redundancies in
depending on the given converters, their purpose and limits. In the system. Still, the reduction in EENS is remarkable. Both tables
this context, the dispatch rule is a boundary condition for the showing relations that were expected can be interpreted as a
reliability calculations, however, without influence on the above proof of the proper operation of the models.
proposed methodology.
According to Fig. 8 the electrical load exceeds the installed
capacity of the direct connection during certain time steps; i.e. the 5. Outlook
load can only be supplied with support from the chemical and the
thermal network. This is visible in the decreased reliability of In this paper, a methodology has been developed for the
supply for the electrical load during time steps 2–4. During time calculation and determination of reliability characteristics of
steps 3 and 4, the reliability is lower than during time step 2 as in multi-carrier systems. The implementation of the methodology
total three connections must be operating to maintain the supply and its proper operation have been shown with two examples,
of the load (also illustrated in Fig. 10 with the marked converters chosen rather for demonstrating the application than for being
indicating crucial connections). During the same time intervals, completely realistic. This section therefore outlines possible
the chemical and the thermal load can be supplied only by their applications and interactions with other operational strategies in
direct connections. As they both have the same reliability multi-carrier systems.
characteristics, the expected reliability of supply is identical. Limits to reliability of supply are given by structures and
The chemical load experiences a redundant supply during time systems rather than through operation. Therefore, the methodol-
steps 1, 5 and 6, which is reflected in a comparatively high ogy could be applied mostly for the structural optimization of new
reliability, particularly in time step 5, when the thermal-chemical systems. As the hub concept can be applied to different areas from
converter could supply the full chemical load on its own. e.g. representing a device with multiple conversions up to a
Looking at EENS reveals further aspects of the mutual building or a part of a city, its application range is manifold.
dependency between the energy carriers. For example during Distributed generation and network enhancements are one
time step 3, the chemical load is not supported by the thermal or current application where the models could be useful. Instead
the electrical system. In case of an outage of the direct connection, of enforcing a heavily loaded branch line for an industrial
the load would be lost and no partial load could be supplied. This area, a local CHP could be installed, providing power during peak
is reflected in the comparatively high EENS. Thus, although the load periods as well as process heat. The introduced methodology
chemical load is constant, both the expected reliability of supply can be applied to define the reliability characteristics of the CHP
and EENS change depending on the other loads and converters. such that a certain level of reliability of supply for the electrical
EENS for the thermal load is highest both due to the relatively supply can be guaranteed. The resulting characteristics however
high load level and due to the limited support from the electrical might be contradictory to other requirements, concerning e.g.
and the chemical systems. emissions or costs. In order to find an overall optimum solution,
Nevertheless, comparing the values from Table 1 with the it is thus necessary to perform a multi-objective optimiza-
results from an analysis without converters, i.e. only considering tion, where reliability criteria are one of several aspects to be
 ARTICLE IN PRESS

G. Koeppel, G. Andersson / Energy 34 (2009) 235–244 243

=1 =2 =3

1.5 1.5 - - - -
el el el
7 7 10 12 10 12.5
10 10 10
1 1 2 - 2 -
ch ch ch
6 6 6 6 6 6
8 8 8
5 1 2 1 - 1
th th th
9 9 12 12 14 14
15 15 15

 =4 =5  =6

- - - - 1 1
el el el
10 13 10 11 8 8
10 10 10
2 - 2 - 1 1
ch ch ch
6 6 6 6 6 6
8 8 8
4 1 6 1 th 5 1
th th
10 10 8 8 9 9
15 15 15

Fig. 10. Momentarily available converter and connection capacities and the corresponding load levels, in kW. Converters required for the supply are marked.

Table 2 Appendix A
Expected reliability of supply and expected energy not supplied (EENS) without
indirect supply paths
A.1. Kronecker products and sums
Output Reliability of supply EENS (Wh)
With matrix A of dimension m  n and matrix B of dimension
Electrical 0.332877 4287.620
p  q, the Kronecker product, denoted with the symbol , is given
Chemical 0.991803 147.541
Thermal 0.991803 254.098 as [13]
2 3
a11 B a12 B    a1n B
6 a21 B a22 B    a2n B 7
6 7
AB¼6 6 .. .. .. .. 77 (24)
4 . . . . 5
optimized. Examples of multi-objective optimizations are dis-
am1 B am2 B    amn B
cussed in [11].
Hence, the Kronecker product of both matrices results in a matrix
of dimension mp  nq. The Kronecker sum, denoted with the
6. Conclusion
symbol , is defined as follows, for square matrices A and B with
dimensions n  n and q  q, respectively:
This paper presents a model for the analysis of reliability of
supply in multi-carrier energy systems. The model is based on the A  B ¼ A  Iq þ In  B (25)
energy hub modeling concept, which allows the analysis of
interconnections and dependencies between several energy Whereas matrix Ix represents the identity matrix of dimension x.
carriers at the same time. In terms of reliability, the interconnec- For more than two matrices, associativity holds true:
tions are meaningful both for partially redundant supply and to A  ðB  CÞ ¼ ðA  BÞ  C (26)
reduce the amount of EENS.
The developed models are suitable for analyzing the reliability These relations can be used to combine the transition rate
of supply and EENS for multi-carrier energy systems. Disadvan- matrices of independent state space diagrams to give the
tages and advantages of interconnections between energy carriers transition rate matrix of the combined state space diagram.
can be identified with the models. The models can also be applied The corresponding state probabilities are defined as the Kronecker
for sensitivity analyses, giving the dependency of the reliability of product of the individual state probabilities:
supply from certain components. The application of the models is ~
P ¼~
PA  ~
PB  ~
PC
demonstrated with two examples, in addition showing the
positive influence of interconnections between the energy carriers The transition rate matrix D ¼ A  ðB  CÞ and the state prob-
in terms of reliability. ability vector ~
P now define the new system as
~
PD¼0 (27)
Acknowledgments

This paper is a revised and extended version of a contribution

submitted to the World Energy System Conference in 2006 [8]. References
The documented research work has been performed within the
Vision of Future Energy Networks project, supported by ABB, [1] Li Z. Natural gas for generation: a solution or a problem? Power Energy Mag
IEEE 2005;3(4):16–21.
Areva, Siemens, and the Swiss Federal Office of Energy. See also [2] Shahidehpour M, Fu Y, Wiedman T. Impact of natural gas infrastructure on
http://www.future-energy.ethz.ch. electric power systems. Proc IEEE 2005;93(5):1042–56.
 ARTICLE IN PRESS

244 G. Koeppel, G. Andersson / Energy 34 (2009) 235–244

[3] Geidl M, Andersson G. Operational and structural optimization of multi- [11] Geidl M. Integrated modeling and optimization of multi-carrier energy
carrier energy systems. Eur Trans Elect Power 2006;16(5):463–77. systems. PhD thesis, Swiss Federal Institute of Technology ETH Zurich.
[4] An S, Li Q, Gedra TW. Natural gas and electricity optimal power flow. In: IEEE/ Dissertation ETH no. 17141, 2007.
PES transmission and distribution conference, Dallas, TX, 2003. p. 138–43. [12] Billinton R, Allan RN. Reliability evaluation of power systems. 2nd ed.
[5] Munoz J, Jimenez-Redondo N, Perez-Ruiz J, Barquin J. Natural gas network New York: Plenum Press; 1996.
modeling for power systems reliability studies. In: Power tech conference [13] Rausand M, Høyland A. System reliability theory: models, statistical
proceedings, vol. 4, Bologna, 2003, p. 8. methods, and applications, 2nd ed. Wiley series in probability and statis-
[6] Helseth A, Holen A. Reliability modeling of gas and electric power distribution tics. Applied probability and statistics. Hoboken, NJ: Wiley-Interscience;
systems; similarities and differences. In: Ninth international conference on 2004.
probabilistic methods applied to power systems, Stockholm, Sweden, 2006. [14] Fuenfgeld C, Tiedermann R. Anwendung der repraesentativen VDEW-
[7] Bakken BH, Holen A. Energy service systems: integrated planning case Lastprofile—step-by-step; vdew materialen m05/2000, Brandenburgische
studies. In: Power engineering society general meeting, Denver, CO, USA, Technische Universitaet, 2000.
2004. p. 2068. [15] Krewitt W, Nitsch J, Fischedick M, Pehnt M, Temming H. Market perspectives
[8] Koeppel G, Andersson G. The influence of combined power, gas, and thermal of stationary fuel cells in a sustainable energy supply system—long-term
networks on the reliability of supply. In: Sixth world energy system scenarios for Germany. Energy Policy 2006;34(7):793.
conference, Turin, Italy, 2006. [16] Riffat SB, Ma X. Thermoelectrics: a review of present and potential
[9] Koeppel G. Reliability considerations of future energy systems: multi-carrier applications. Appl Thermal Eng 2003;23(8):913.
systems and the effect of energy storage. PhD thesis, Swiss Federal Institute of [17] Tanrioven M, Alam MS. Reliability modeling and analysis of stand-alone PEM
Technology ETH Zurich. Dissertation ETH no. 17058, 2007. fuel cell power plants. Renewable Energy 2006;31(7):915.
[10] Geidl M, Andersson G. Optimal power flow of multiple energy carriers. IEEE [18] Dugan R, McDermott T. Operating conflicts for distributed generation on
Trans Power Systems 2007;22(1):145–55. distribution systems. In: Rural electric power conference, 2001. p. A3/1–6.