Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

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

net/publication/316501631

Characterization of Cutsets in Networks With Application to Transient Stability


Analysis of Power Systems

Article  in  IEEE Transactions on Control of Network Systems · September 2018


DOI: 10.1109/TCNS.2017.2698266

CITATIONS READS

6 316

3 authors:

Yue Song David J. Hill


The University of Hong Kong The University of Hong Kong
50 PUBLICATIONS   237 CITATIONS    515 PUBLICATIONS   22,913 CITATIONS   

SEE PROFILE SEE PROFILE

Tao Liu
The University of Hong Kong
76 PUBLICATIONS   1,222 CITATIONS   

SEE PROFILE

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

National Basic Research Program of China View project

National Basic Research Program (973) View project

All content following this page was uploaded by Yue Song on 12 December 2019.

The user has requested enhancement of the downloaded file.


IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS 1

Characterization of Cutsets in Networks with


Application to Transient Stability
Analysis of Power Systems
Yue Song, Student Member, IEEE, David J. Hill, Life Fellow, IEEE, and Tao Liu, Member, IEEE

Abstract—A cutset is a concept of importance in both graph In the context of power systems, transient stability problems
theory and many engineering problems. In this paper, cutset are closely linked to cutsets. Transient stability refers to the
properties are studied and applied to transient stability problems ability of a power system to maintain angle synchronism when
in power systems. First, we reveal that cutsets are intrinsically
linked to cycles, another important concept in graph theory. The subjected to large disturbances [3]. A typical cutset-related
main theoretical results include that any pair of edges in a cutset phenomenon in transient stability is that the loss of synchro-
are concyclic (i.e., contained in the same cycle), and any edge nism of a power system initiates from the angle separation
together with some of its concyclic edges can form a cutset. Then, across the transmission lines in some cutsets in the underlying
with the cycle constraint for the bus angles in power networks, power network, called the critical cutsets. This critical cutset
we give a theoretical explanation of a common phenomenon in
transient stability that the loss of synchronism initiates from phenomenon is an indicator of transient instability that is
the angle separation across the critical cutset in the underlying frequently observed in simulations and real cases. There have
power network. This phenomenon is frequently observed but no been some studies that attempt to explore the instability mecha-
proof has been presented yet. Moreover, an improved cutset index nism from the perspective of cutsets. For the small-disturbance
(ICI) is proposed based on these cutset properties. This index stability problem, some equilibrium points have a feature that a
can better identify the vulnerable cutset and help to estimate the
cutset-relevant unstable equilibrium points and critical energy for cutset is formed by transmission lines with angle differences
stability region determination. Numerical studies on two IEEE beyond 90◦ . It is shown in [4–6] that the equilibrium point
test systems show that the formation of critical cutset coincides with the presence of such a cutset is an unstable equilibrium
with the explanation, and the ICI has better performance than point (UEP). Also, the normal form method has been adopted
the conventional cutset index especially in heavy load cases. in [7] to investigate the angle separation in the neighbourhood
Index Terms—power network, transient stability, cutset, graph of a UEP. However, the linearization approach used in small-
theory. disturbance stability analysis is not applicable to transient
stability problems where the disturbances are large and the
I. I NTRODUCTION nonlinear dynamics become significant. A recent progress
HE concept of cutsets is important in graph theory and has been made in the cluster synchronization of dynamical
T network science. A cutset is a group of edges whose networks, which is similar to the critical cutset phenomenon.
deletion will split a graph into two disjoint subgraphs (the It shows that the cluster synchronization can be interpreted as
rigorous definition will be presented later). Many concepts the solution to a saddle-point problem [8]. This result provides
and properties in graphs and networks are closely related new insights into the system response in the large-disturbance
to cutsets. For instance, the edge-connectivity of a graph is scenario, but the proposed model in [8] is not applicable to
defined as the cardinality of its minimum cutset. In a flow power systems. The problem of how the critical cutset is
network, the total capacity of the minimum cutset is equal to formed in case of transient instability remains open.
the maximum amount of flow passing from the source to the The close link between cutsets and transient stability has
sink, which is known as the max-flow min-cut theorem [1]. also inspired useful stability evaluation and control methods.
In addition, to find the maximum cutset is a combinatorial The cutset index (CI), which is a sum of the potential energies
optimization problem with both theoretical value and wide of transmission lines in a cutset, is proposed in [9] to measure
range of applications that has been extensively studied in cutset vulnerability and estimate the critical energy. After-
discrete mathematics [2]. wards, the CI has been incorporated into active power dispatch
problems as a security objective [10]. The potential energy of
Manuscript received May 12, 2016; revised November 7, 2016; accepted cutsets along post-fault trajectories has been investigated in
April 10, 2017. This work was supported by the Hong Kong PhD Fellowship
Scheme (HKPFS) and the Research Grants Council of the Hong Kong Special [11]. Moreover, an online critical cutset detection approach
Administrative Region under the General Research Fund (GRF) through by using voltage and current measurements is proposed in
Project No. 17202414. [12]. In [13], the researchers apply energy function analysis by
Y. Song and T. Liu are with the Department of Electrical and Elec-
tronic Engineering, The University of Hong Kong, Hong Kong (e-mail: using data from phasor measurement units (PMUs) to monitor
yuesong@eee.hku.hk; taoliu@eee.hku.hk). the angle dynamics across key cutsets. The identification of
D. J. Hill is with the Department of Electrical and Electronic Engineering, the cutset of lines whose contingency causes severe power
The University of Hong Kong, Hong Kong, and also with the School of
Electrical and Information Engineering, The University of Sydney, Sydney, outage and the corresponding optimal control strategy are
NSW 2006, Australia (e-mail: dhill@eee.hku.hk; david.hill@sydney.edu.au). proposed in [14, 15]. In [16], graphical observers are designed
2 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

to estimate the state of key cutsets in power networks. The and E are |V| = n and |E| = l, respectively. The notation
controlled islanding techniques are established in [17–19] to ek = (i, j) ∈ E, k = 1, 2, .., l denotes the edge k connecting
find out the optimal cutset decision that can stabilize the node i and node j, where (i, j) is an unordered pair of nodes.
system and minimize the power imbalance within each island. In order to define the incidence matrix, each edge of G(V, E)
Nevertheless, these contributions are based on approximations is fictitiously assigned an arbitrary but fixed orientation, and
and/or heuristic reasoning. For further understanding transient we say the edge ek = (i, j) starts from node i and ends at node
stability, more theoretical and practical properties of cutsets j. Then, the incidence matrix E ∈ Rn×l is defined such that
need to be explored. ∀ek = (i, j) ∈ E, Eik = 1, Ejk = −1 and Emk = 0, m 6= i, j.
In this paper, we study cutsets by using algebraic graph
theory, and apply the obtained results to address the afore- II. P ROBLEM FORMULATION
mentioned open problems in power system transient stability. A. Structure preserving model & topological Lyapunov func-
The main contributions of this paper are threefold. Firstly, new tion of power systems
theorems on cutsets are derived, which are summarized as
Consider a power network with synchronous generators
follows:
and loads. The synchronous generators are modeled as con-
1) If a cutset consists of a single edge, then this edge is
stant voltage sources (i.e., generator internal buses) behind
non-cyclic, i.e., it is not contained in any cycles.
transient reactances. In the structure preserving model, the
2) If a cutset consists of multiple edges, then for any
power network is augmented with fictitious buses and lines
arbitrary two edges in the cutset, there exists a cycle that
to represent the corresponding generator internal buses and
contains these two edges but does not contain the other
transient reactances, respectively. The loads are assumed to
edges in the cutset.
be frequency dependent. The power network is connected.
3) Any edge together with some of its concyclic edges can
The transmission lines in the network are modeled as series
form a cutset.
reactances, and transformers and transient reactances are also
These new properties reveal a close relationship between
regarded as lines. Denote VG as the set of generator internal
cutsets and cycles. Secondly, with the cycle constraint in power
buses, and VL as the set of the remaining buses that are without
networks that the sum of bus angle differences across any
generators and may connect loads. Then, the system dynamics
cycle is zero, we present an explanation of how the critical
can be described by [9]
cutset is formed in the transient response of a power system. X
The critical cutset is usually considered to be closely related Mi θ̈i + Di θ̇i = Pi − Vi Vj |Yij | sin(θi − θj ), i ∈ VG
to node dynamics, while the role of network topology has not j∈Ni
(1)
been paid as much attention. Our result implies that the power
X
Di θ̇i = Pi − Vi Vj |Yij | sin(θi − θj ), i ∈ VL
network structure is a key factor in causing this phenomenon. j∈Ni
Thirdly, motivated by the new properties of cutsets, we propose
an improved cutset index (ICI) to better identify the vulnerable where θi , Vi denote the phase angle and voltage magnitude of
cutset and estimate the critical energy for stability region bus i, respectively; θ̇i denotes the angular frequency of bus i
determination. and henceforth we will use ωi = θ̇i to represent it; Ni denotes
The remainder of the paper is organized as follows. Power the set of adjacent buses of bus i, the notation j ∈ Ni means
system modeling from a graph-theoretic perspective and back- bus i and bus j are connected by a line; −Yij denotes the
ground knowledge of transient stability and critical cutsets are susceptance of line (i, j). For bus i ∈ VG , Pi > 0 represents
given in Section II. Cutset properties are studied in Section III, the active power generation, and Mi > 0 and Di > 0
which builds up a bridge between cutsets and cycles in graphs. denote the inertia and damping coefficient of the synchronous
Then, the theoretical results are applied to transient stability generator, respectively. For bus i ∈ VL , the term −Pi + Di θ˙i
issues in Section IV and Section V, where an explanation represents the active power load with −Pi > 0 being the
of the critical cutset phenomenon and the ICI are proposed, rated load and Di being the load frequency coefficient. Unlike
respectively. Section VI verifies the results on two IEEE test the network reduced model where the set of buses VL are
systems, Section VII concludes the paper. eliminated via Kron reduction [20], the structure preserving
Notations: The following notations will be used throughout model contains the information of the original physical power
this paper. For simplicity, a vector x = [x1 , x2 , ..., xp ]T ∈ Rp network structure, which is preferable for the cuset-based
is denoted as x = [xi ] ∈ Rp , and a diagonal matrix A = analysis in this paper.
diag{a1 , a2 , ..., ap } ∈ Rp×p is denoted as A = diag{ai } ∈ We list the assumptions for system (1) below and make
Rp×p . The notation sin x ∈ Rp represents the function vector some remarks before proceeding further.
sin x = [sin x1 , ..., sin xp ]T , 1p ∈ Rp represents the vector (A1) The voltage magnitude of each bus is constant;
whose entries are all equal to one, and Ip ∈ Rp×p represents (A2) The loads have positive frequency coefficients.
the identity matrix. We slightly abuse the notation | · |. It Remark 1: Assumption (A1) coincides with the high-voltage
represents the absolute value when applied to a number and transmission system characteristics that angle dynamics and
it represents the set cardinality when applied to a set. We voltage dynamics are highly decoupled. Also the increasing
consider undirected graphs in this paper. Let G(V, E) denote use of power electronic voltage controllers can add further
an undirected graph where V denotes the set of nodes and justification. Moreover, the simplification made by (A1) will
E ⊆ V × V denotes the set of edges. The cardinality of V not cause any limitations of the results in the following
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 3

sections as they reveal some universal laws for power networks. existence of such a point is a basis for transient stability
The obtained results also apply to other structure preserving analysis. In addition, a post-fault power system described by
models that include voltage variations, excitor control loops, (2) is said to be transiently stable if the system trajectory
reactive compensation devices, etc [21–24]. converges to the operating point.
Remark 2: The positive load frequency coefficient in As- Next we briefly review the topological Lyapunov function
sumption (A2) reflects the real situations, which has been of system (2) and its link to transient stability (referring to [9]
commonly adopted in the literature [25–29]. The term Di θ˙i for the details). Define a Lyapunov function candidate
in (1) describes the typical load response during a transient
process where the frequency usually has small fluctuation. The V (σ, ω) = VK (ω) + VP (σ) (3)
load model can be extended to voltage-dependent loads when where VK (ω) = 21 ω T M ω is the kinetic energy of generator
voltage variations are considered, e.g., see [22, 23]. It is also rotors and
worth pointing out that power system dynamics is generally
different from dynamical networks such as consensus proto- VP (σ) = 1Tl B(cos σ s − cos σ) + (σ s − σ)T B sin σ s (4)
cols [30] and Kuramoto oscillators [31]. The latter one is a
is the total potential energy of transmission lines so that
differential equation (DE) system, while the former one could
V (σ, ω) is the total system energy. We also define
be a differential-algebraic equation (DAE) system in some
cases. For example, an algebraic equation occurs if there exists VP,k (σk ) = Bk (cos σks − cos σk ) + Bk (σks − σk ) sin σks (5)
a bus i ∈ VL such that Di = 0, i.e., bus i has no load or the
load is frequency-independent. However, the zero frequency as the potential energy of line ek . It follows that
coefficients will not impact the obtained results in this paper. X
VP (σ) = VP,k (σk ). (6)
This is because the Lyapunov function (detailed expression
ek ∈E
will be given later), which is a basis for the ICI, is commonly
independent of frequency coefficients and applies to the cases Note that the potential energy can be re-expressed as
where both positive and zero frequency coefficients exist. We VP = 1Tl B[cos(E T θ s ) − cos(E T θ)] + (θ s − θ)T P (7)
refer to [24, 32] for a comprehensive review of Lyapunov
functions with frequency-independent loads. which has also been used in the literature [21, 22, 25]. We
We now reformulate (1) with the graph explicit. Denote adopt form (4) as it is more suitable for cutset-based analysis.
V = VG ∪ VL as the set of buses and E as the set of lines1 It proves that V̇ (σ, ω) = −ω T Dω ≤ 0 and V (σ, ω) > 0,
with |V| = n and |E| = l. Thus, the power network can be ∀σ ∈ Rσ and ∀ω, where
regarded as a graph G(V, E). We denote E as the incidence Rσ = {σ| σkmin < σk < σkmax }
matrix of the graph G(V, E), θ = [θi ] ∈ Rn as the vector of
bus angles, and σ = E T θ = [σk ] ∈ Rl as the vector of angle σkmin = −π − σks (8)
differences across lines where σk = θi − θj , ∀ek = (i, j) ∈ E. σkmax =π− σks .
In addition, we introduce the vector of angular frequencies Thus V (σ, ω) is a Lyapunov function ∀σ ∈ Rσ and ∀ω.
ω = [ωi ] ∈ Rn and line coupling strength Bk = Vi Vj |Yij |, Further, it is called the topological Lyapunov function as it
∀ek = (i, j) ∈ E. We set Mi = 0, ∀i ∈ VL . Then, by taking consists of the potential energy of each transmission line that
(σ, ω) as the state variables, system (1) can be re-expressed characterizes the power network topology. Denote (σ u , 0) as
as (referring to [33] for the detailed derivation) the lowest UEP, i.e., the UEP with the minimum system
σ̇ = E T ω energy, or equivalently, the minimum potential energy as
(2) ω = 0 at an equilibrium point. The set Ω = {(σ, ω)| σ ∈
M ω̇ = P − Dω − EB sin σ
Rσ , V (σ, ω) < VP (σ u )} gives an estimation of the stability
where M = diag{Mi } ∈ Rn×n , D = diag{Di } ∈ Rn×n , region. Any system trajectory starting from Ω will converge to
P = [Pi ] ∈ Rn and B = diag{Bk } ∈ Rl×l . the operating point (σ s , 0) and thus achieves transient stability.
At an equilibrium point of system P (2), the synchronized We henceforth refer to the potential energy at the lowest UEP
n Pn
angular frequency is given byPω e = i=1 Pi / i=1 Di . It as the critical energy for transient stability.
n
is convenient to assume that i=1 Pi = 0 and ω e = 0 as it
can be achieved by a coordinate translation (refer to [9] for B. Critical cutset: background & motivation
the details). Thus, we denote (σ e , 0) as an equilibrium point
of system (2), which satisfies the power flow equation P = The critical cutset phenomenon is frequently observed in
EB sin σ e . Further, we make one more assumption below case of transient instability, as depicted in Fig. 1. We summa-
rize this phenomenon into the following proposition.
(A3) There exists a unique equilibrium point, say (σ s , 0)
Proposition 1: The loss of synchronism of a power system
such that |σks | < π2 , ∀ek ∈ E.
originates from angle separation across (at least) one critical
The equilibrium point satisfying Assumption (A3) is asymp- cutset in the underlying power network. The angle differences
totically stable [9] and serves as the operating point. The across the lines in the critical cutset go unbounded.
1 As a convention, we will use two groups of terms “bus, line, network” and
We also refer to [12] for some relevant statements on
“node, edge, graph” interchangeably. The former group is for power systems the critical cutset phenomenon. Note that there appears no
and the latter group is for graphs. agreement on the number of critical cutsets from which
4 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

transient instability is initiated. It is rare but observed in some III. C HARACTERIZATION OF CUTSETS
cases that more than one critical cutsets are formed during
A. Preliminaries on graph cutsets and cycles
the post-fault dynamics, i.e., bus angles are separated into
three or more groups [7]. One viewpoint allows the case of We present some background knowledge of cutsets and cy-
multiple critical cutsets, while another viewpoint regards that cles as a basis (referring to [39–41] for the detailed discussions
the loss of synchronism still initiates from the angle separation and proofs). Let G(V, E) be an undirected graph defined before.
across a unique critical cutset and the other critical cutsets are We first introduce the concept of a path in G. A path is a set of
developed from this cutset afterwards [24, 34]. Nevertheless, edges that connect a sequence of nodes which are all distinct
this debate will not influence Proposition 1 where a general from one another, except that the starting node and ending
statement is used. node are possibly the same. A graph is connected if there is
Proposition 1 is commonly adopted in power system com- a path between every pair of nodes. We consider connected
munities to guide and simplify transient stability analysis of graphs in this paper. A tree is an undirected graph where any
multi-machine systems. One application is the extended equal two nodes are connected by exactly one path. For a connected
area criterion (EEAC), see [35] and references therein. We graph G(V, E), we can select a spanning tree T (V, ET ) with
take Fig. 1 to illustrate the EEAC. The critical cutset in Fig. E T ⊆ E and |E T | = n − 1 that contains every node of
1 splits the power network into two subnetworks, and the G. Deleting E T from G(V, E) gives the remaining subgraph,
generators in each of the two subnetworks are regarded as denoted as C(V, E C ), where E C = E\E T and |EC | = l − n + 1.
coherent generators [36, 37]. As shown in Fig. 2, the two A cycle is a path whose starting node is also the ending
cyc
groups of coherent generators are approximately represented node, denoted as E . It is also called a simple cycle or
by two generators, and the system then finally equivalent to a circuit in some literature [42, 43]. The cycles have the
a single-machine infinite-bus system where the classical equal following properties. In algebraic graph theory, a cycle E cyc
cyc l
area criterion applies. We refer to [32] for the details of the can be expressed by an edge selection vector x ∈ R whose
cyc
classical equal area criterion. Another common application entries can be 0, ±1, where xcyc k = 0 means the edge ek is not
links the critical cutsets to UEP profile, which originates from contained in the cycle, and x k = 1 or xcyck = −1 means the
an important observation that each critical cutset corresponds edge e k is contained in the cycle with its assigned orientation
to certain relevant UEPs [9, 22, 38]. For example, the cutset or the reversed orientation. It follows that Excyc = 0 if a
cyc
in Fig. 1 corresponds to the relevant UEP (σ e , 0) that satisfies vector x 6= 0 represents a cycle. Henceforth, the notations
cyc
σke ≈ σkmax or σke ≈ σkmin for line ek between the two E and xcyc will be used interchangeably to represent a
subnetworks and σke ≈ σks for line ek within each subnetwork. cycle. For any ek ∈ EC , the set {ek } ∪ ET contains a unique
As the exact UEP may be hard to find, this property leads to cycle, namely the fundamental cycle. The orientation of the
a fast estimation of UEPs and critical energy. fundamental cycle is taken to be consistent with the assigned
orientation of ek ∈ EC . There are totally l − n + 1 fundamental
cycles for a given T (V, ET ), and the corresponding l − n + 1
edge selection vectors form the fundamental cycle matrix,
denoted as O ∈ Rl×(l−n+1) . Also, we henceforth refer to
C(V, EC ) as the cycle subgraph as it completes cycles in G.
A cutset is defined as the set of all edges with one
endpoint in V1 and the other endpoint in V2 , denoted as
E cut , where V1 , V2 6= φ is an arbitrary partition of nodes
such that V1 ∪ V2 = V, V1 ∩ V2 = φ. The cutsets have the
Figure 1. An example of the critical cutset. following properties. The deletion of E cut splits G(V, E) into
two connected components G1 (V1 , E1 ) and G2 (V2 , E2 ) where
E1 ∪ E2 = E\E cut and E1 ∩ E2 = φ. Furthermore, for any
ek ∈ E cut , the graph G ′ (V, E ′ ) with E ′ = {ek } ∪ E1 ∪ E2
is connected. Similarly, we also use an edge selection vector
xcut ∈ Rl to represent the cutset: xcut k = 0 if edge ek is not in
the cutset, and xcut k = 1 or x cut
k = −1 according to whether
the fictitiously assigned orientation of edge ek is directed from
V1 to V2 or from V2 to V1 . Henceforth, the notations E cut and
Figure 2. Dynamical equivalent of coherent generators. xcut will be used interchangeably to represent a cutset. For any
ek ∈ ET , the set {ek }∪EC contains a unique cutset, namely the
Despite that the critical cutset phenomenon leads to many in- fundamental cutset, where we assume ek is directed from V1
spiring results as above, to the best of the authors’ knowledge, to V2 . There are totally n − 1 fundamental cutsets for a given
its theoretical foundation has not been fully addressed. In the T (V, ET ), and the corresponding n − 1 edge selection vectors
following sections, we will first study cutsets from a graph- form the fundamental cutset matrix, denoted as Φ ∈ Rl×(n−1) .
theoretic perspective. Then we apply the obtained theorems Also note that different selections of T (V, ET ) and C(V, EC )
to prove Proposition 1 and develop an improved way for lead to different combinations of the fundamental cycles and
estimating UEPs and critical energy. fundamental cutsets, and thus different matrices O and Φ.
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 5

B. Main results Theorem 1: The following statements hold for cutsets:


With the concepts and notations introduced above, a graph- 1) Any non-cyclic edge forms a simple cutset.
theoretic analysis of cutsets is conducted in this subsection that 2) For any composite cutset, there exists at least one cycle
leads to insightful observations. We begin with a more detailed that contains at least two edges in the cutset. Further, for
description of the fundamental cycle matrix and fundamental any arbitrary two edges in a composite cutset, there exists
cutset matrix. Let T (V, ET ) and C(V, EC ) be a spanning tree a cycle that contains these two edges but does not contain
and the cycle subgraph of G(V, E), respectively. Without loss the other edges in the cutset.
of generality, the edges in ET are numbered from 1 to n−1 and Proof: 1) It is trivial to show that deleting a non-cyclic
the edges in EC are numbered from n to l. Then, by definition, edge from a connected graph will split it into two connected
the fundamental cycle matrix O can be expressed as subgraphs.

Of
 2) We prove the first part by contradiction. Suppose there
O = [O1 , O2 , ..., O l−n+1 ] = ∈ Rl×(l−n+1) (9) is a composite cutset E cut such that the edges in any sub-
Il−n+1
set of E cut are not concyclic. Without loss of generality,
where O i is the i-th column of O representing the i-th we assume an edge ej ∈ E cut . By Lemma 2, E cut is a
fundamental cycle, O f ∈ R(n−1)×(l−n+1) and Il−n+1 are fundamental cutset with respect to a specific spanning tree
the sub-matrices with respect to ET and EC , respectively. The T (V, ET ) and cycle subgraph C(V, EC ) such that ej ∈ ET and
entries of O f can be ±1, 0. Similarly, the fundamental cutset (E cut \ej ) ∈ EC . Then, the edge selection vector representing
matrix Φ can be expressed as E cut is Φj with Φjj = 1. By (9)-(11), we have O T Φ = 0,
Pl

I
 i.e., k=1 Oki Φkj = 0, i = 1, 2, ..., l − n + 1. From those
Φ = [Φ1 , Φ2 , ..., Φn−1 ] = n−1 ∈ Rl×(n−1) (10) equations we can claim that Oki Φkj = 0, k = 1, 2, ..., l,
Φf
i = 1, 2, ..., l − n + 1. Otherwise, since the value of Oki Φkj
where Φi is the i-th column of Φ representing the i-th can only be 0, ±1, there will exist a fundamental cycle O i and
fundamental cutset, In−1 and Φf ∈ R(l−n+1)×(n−1) are the at least two edges, say ek1 , ek2 , in the cutset E cut such that
sub-matrices with respect to ET and EC , respectively. The Ok1 i Φk1 j 6= 0 and Ok2 i Φk2 j 6= 0. This implies that the edges
entries of Φf can be ±1, 0. ek1 , ek2 ∈ E cut are concyclic in the fundamental cycle O i so
With the above expressions of the fundamental cycle matrix that a contradiction yields. Since Oki Φkj = 0, k = 1, 2, ..., l,
and fundamental cutset matrix, we present the following i = 1, 2, ..., l − n + 1, applying Φjj = 1 to these equations
lemmas and definitions. The main results—Theorem 1 and gives that Oji = 0, i = 1, 2, ..., l − n + 1. And by (11) we
Theorem 2, will then be proved. have Φkj = 0, k = n, ..., l. Further, it follows from (10) that
Lemma 1: [41, Lemma 5.4] The following equality holds only the j-th entry of Φj is nonzero, which implies that the
cutset E cut is a simple cutset and again makes a contradiction.
Φf = −OTf . (11)
Then we prove the second part. For any arbitrary two
Lemma 2: Any cutset in the graph G(V, E) is a fundamen- edges ek1 , ek2 ∈ E cut , by Lemma 2, we can select the
tal cutset with respect to a properly selected spanning tree T (V, ET ) and C(V, EC ) such that ek1 ∈ ET and ek2 ∈ EC .
T (V, ET ). Without loss of generality, assume ek1 is numbered as the
Proof: Consider an arbitrary cutset E cut in the graph first edge in T (V, ET ) and ek2 is numbered as the first edge
G(V, E). Deleting E cut splits the graph into two connected in C(V, EC ). Thus we have k1 = 1, k2 = n and the numbers
components G1 (V1 , E1 ) and G2 (V2 , E2 ) where V1 ∪ V2 = V, of the other edges in E cut range from n + 1 to l. Then, the
V1 ∩ V2 = φ, E1 ∪ E2 = E\E cut and E1 ∩ E2 = φ. edge selection vector representing E cut is (by proper edge
In addition, for any ek ∈ E cut , the graph G ′ (V, E ′ ) with orientation assignment)
E ′ = ({ek } ∪ E1 ∪ E2 ) ⊆ E is connected. Thus, we can select  T
Φ1 = 1 0 ··· 0 −1 ···
a spanning tree T (V, ET ) and the cycle subgraph C(V, EC )
with ET ⊆ E ′ and EC = E\ET such that ek ∈ ET and ↓ ↓ ··· ↓ ↓ ··· (12)
(E cut \{ek }) ⊆ EC . Therefore, the cutset E cut is a fundamental Φ11 Φ21 · · · Φn−1,1 Φn1 · · ·
cutset with respect to the selected spanning tree T (V, ET ).
where we do not care about the values of Φn+1,1 , ..., Φl1 . By
Lemma 2 provides an alternative way to analyze a cutset
(9) and (11), the first column of the corresponding fundamental
by transforming it into a fundamental cutset. Then, the funda-
cycle matrix O has the following form
mental cutset matrix, fundamental cycle matrix and properties
T
shown in Lemma 1 can be used, which will lead to significant

O1 = 1 ··· 1 0 ··· 0
simplification in the analysis. Also, to better describe our (13)
↓ ··· ↓ ↓ ··· ↓
results, the following concepts are introduced.
Definition 1: In a graph, an edge is a cyclic edge if it is O11 · · · On1 On+1,1 · · · Ol1
contained in at least one cycle, otherwise it is a non-cyclic where On+1,1 , ..., Ol1 are all zero and we do not care about
edge. A set of edges are said to be concyclic if they are the values of O21 , ..., On−1,1 . This implies that ek1 , ek2 are
contained in a common cycle. concyclic in the fundamental cycle O 1 . Furthermore, since
Definition 2: A cutset is a composite cutset if it is formed the (n+1)-th to l-th entries of O 1 are all equal to zero, the
by multiple edges, otherwise it is a simple cutset. fundamental cycle O 1 does not contain any other edges in
6 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

the cutset E cut . Thus we find such a cycle that satisfies the If the edge eki connects Gi′ (Vi , Ei′ ) and G1b′ ′
(V1b , E1b ),
required conditions. merge them in a similar way. Update Eek by Eek ←
Next we reveal another relationship between cutsets and Eek \eki .
cycles. For any cyclic edge ek in a graph G(V, E), suppose Step 3: If there exist some sub-graphs Gi′ (Vi , Ei′ ), 2 ≤
there are totally m different cycles, say E1cyc , ..., Em cyc
such i ≤ g that are not merged into G1a ′ ′
(V1a , E1a ) or
cyc cyc ′ ′
that ek ∈ Ei , i = 1, ..., m. For each cycle Ei , select an G1b (V1b , E1b ), go back to Step 1. Otherwise, stop the
edge eki such that eki ∈ Eicyc and eki 6= ek . Define Eek = manipulation.
{ek1 , ek2 , ..., ekm } (If eki = ekj for some 1 ≤ i, j ≤ m, merge Since the graph G(V, E) is connected after deleting ek , the
them into one). Then, we have the following theorem. above manipulations will finally satisfy the stop criterion. Then
Theorem 2: For any cyclic edge ek in the graph G(V, E), all those subgraphs are merged into two disjoint subgraphs
there exists at least one composite cutset in the set {ek } ∪ Eek . G1a ′ ′ ′ ′ ′
(V1a , E1a ), G1b (V1b , E1b ) with V1a ∪ V1b = V and E1a ∪

Further, among all these composite cutsets in the set {ek }∪Eek , E1b = E\({ek } ∪ Eek ) (Hereafter V1a , E1a , V1b , E1b and Eek ′ ′

there exists at least one cutset that contains edge ek . refer to the updated ones after the mergence). Obviously, the
Proof: We prove the first part by contradiction. If there graph G ∗ (V, E ∗ ) is connected, where E ∗ = E1a ′
∪ E1b ′
∪ {ek },
exists no cutset in the set {ek } ∪ Eek , the graph G(V, E) is still and the edge ek forms a simple cutset in G (V, E ∗ ). Thus, ∗

connected after deleting all the edges in the set {ek } ∪ Eek . there exists at least one subset Ee′ k ⊆ Eek such that {ek } ∪ Ee′ k
By Lemma 2, we can select the T (V, ET ) and C(V, EC ) such is a cutset in G(V, E). This completes the proof.
that ET ⊆ E\({ek } ∪ Eek ) and ({ek } ∪ Eek ) ⊆ EC . Then, We take the graph in Fig. 3 to illustrate Theorem 1 and
the property of cycles in Section III-A gives that the set Theorem 2. The fictitious orientations are assigned to the
{ek } ∪ ET contains a fundamental cycle. On the other hand, edges. If we select ET = {e1 , e2 , e3 } (the black edges) and
by the definition of Eek , for any cycle Eicyc (i = 1, 2, ...m) EC = {e4 , e5 , e6 } (the blue edges), then the fundamental cycle
that contains edge ek , we can find at least one edge eki 6= ek , matrix and fundamental cutset matrix are
eki ∈ Eicyc such that eki ∈ Eek and eki ∈ EC . It implies 
−1 1 1
 
1 0 0

that the set {ek } ∪ ET contains no cycle, which makes a −1 0 1 0 1 0
contradiction. So there exists at least one cutset in the set
   
 0 1 1 0 0 1
{ek } ∪ Eek . Moreover, by Theorem 1, all the cutsets in the set O= 
,Φ =  1
  . (15)
 1 0 0 1 0
{ek } ∪ Eek are composite cutsets as they consist of multiple
 
 0 1 0 −1 0 −1
cyclic edges. 0 0 1 −1 −1 −1
Then we prove the second part. Since for any cycle Eicyc ,
i = 1, 2, ...m, there exists an edge eki 6= ek , eki ∈ Eicyc such The set {e1 , e4 , e5 , e6 }, which is marked by the red dash
that eki ∈ Eek , we can conclude that ek is non-cyclic in the line, forms a cutset as well as a cycle coincidentally. This
graph G ′ (V, E ′ ) where E ′ = E\Eek . Consider the following implies that there may exist a cycle that contains more than
two cases about G ′ (V, E ′ ): two edges in a composite cutset, which does not contradict
1) G ′ (V, E ′ ) is connected. By Theorem 1, the non-cyclic to Theorem 1. On the other hand, according to Theorem 1,
edge ek forms a simple cutset in G ′ (V, E ′ ). Thus, there exists we can always find the cycles that contain arbitrary two and
at least one subset Ee′ k ⊆ Eek such that {ek }∪Ee′ k is a cutset in only two edges in this cutset {e1 , e4 , e5 , e6 }, e.g., the cycle
G(V, E). This is the case in the second part of the statements. {e1 , e2 , e4 } contains e1 , e4 , the cycle {e1 , e5 , e3 } contains
2) G ′ (V, E ′ ) is not connected. It implies that the set Eek e1 , e5 , the cycle {e1 , e2 , e6 , e3 } contains e1 , e6 , the cycle
forms at least one cutset in G(V, E) so that G ′ (V, E ′ ) consists {e2 , e4 , e3 , e5 } contains e4 , e5 , the cycle {e3 , e4 , e6 } contains
of several connected components, say Gi′ (Vi , Ei′ ), i = 1, 2, ..., g e4 , e6 , and the cycle {e2 , e6 , e5 } contains e5 , e6 . Next we
such that i=1 Vi = V, i=1 Ei′ = E ′ and Vi ∩ Vj = φ, show how Theorem 2 works. Consider the cyclic edge e1
Sg Sg
Ei′ ∩ Ej′ = φ, ∀i 6= j, 1 ≤ i, j ≤ g. Without loss of generality, that is contained in four different cycles, i.e., {e1 , e5 , e3 },
we assume ek ∈ E1′ , and then applying Theorem 1 gives {e1 , e2 , e4 }, {e1 , e2 , e6 , e3 } and {e1 , e5 , e6 , e4 }. If we select
that ek forms a simple cutset in G1′ (V1 , E1′ ). Let G1a ′
(V1a , E1a′
) one edge other than edge e1 from each of these four cycles to
and G1b (V1b , E1b ) denote the two connected components after form Ee1 , e.g., Ee1 = {e5 , e2 , e4 } or Ee1 = {e3 , e4 }, then it is
′ ′

deleting ek in G1′ (V1 , E1′ ), where V1a ∪V1b = V1 , V1a ∩V1b = φ, easy to find cutsets containing edge e1 from the set {e1 } ∪ Ee1 .
′ ′
E1a ∪ E1b = E1′ \{ek } and E1a ′
∩ E1b′
= φ. Then, we re-
′ ′ ′ ′
cursively merge those subgraphs G1a (V1a , E1a ), G1b (V1b , E1b ),
′ ′ ′ ′
G2 (V2 , E2 ), ..., Gg (Vg , Eg ) by the following steps:
Step 1: Find an edge eki ∈ Eek with one endpoint in
Gi′ (Vi , Ei′ ), 2 ≤ i ≤ g and the other endpoint in
′ ′ ′ ′
G1a (V1a , E1a ) or G1b (V1b , E1b ). The existence of such
an edge is guaranteed by that the graph G(V, E) is
still connected after deleting ek .
Step 2: If the edge eki connects Gi′ (Vi , Ei′ ) and G1a ′ ′
(V1a , E1a ), Figure 3. An example of cutset and cycle.
′ ′
merge them into a new graph G1a (V1a , E1a ) by
Remark 3: Theorem 1 and Theorem 2 reveal that the
′ ′
V1a ← V1a ∪ Vi , E1a ← E1a ∪ Ei′ ∪ {eki }. (14) two important concepts in graphs—cutsets and cycles, which
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 7

seem to have no correlation, are intrinsically linked. The two IV. A PPLICATION I: ANALYSIS OF CRITICAL CUTSET
theorems characterize cutsets in terms of cyclic edges. It can be In this section, we show that the properties of cutsets
inferred from Theorem 1 that a simple cutset only consists of revealed by Theorem 1 and Theorem 2 can shed light on the
a single non-cyclic edge, while a composite cutset consists of critical cutset phenomenon stated in Proposition 1. To see this,
multiple cyclic edges. Further, the edges in a composite cutset we first introduce a physical constraint for power networks. For
are concyclic in pairs by choosing proper cycles. Theorem any cycle xcyc , it follows from the property Excyc = 0 that
2 shows that any cyclic edge ek with some of its concyclic
edges can form a composite cutset. These new properties (xcyc )T σ = (xcyc )T E T θ = (Excyc )T θ = 0. (19)
help to better understand the role that cutsets and cycles
play in the transient process of power systems, which will be This implies the sum of the angle differences across a cycle is
detailed in the following sections. Moreover, the applications zero. We refer to condition (19) as the cycle constraint. With
are even beyond power networks as the theorems are universal the fundamental cycle matrix O, the cycle constraint can be
to generic networks. generalized as
Further, we derive the explicit expression of the fundamental O T σ = 0. (20)
cutset matrix Φ and fundamental cycle matrix O. With the
We recall that lines in a composite cutset are concyclic by
spanning tree T (V, ET ) and cycle subgraph C(V, EC ) selected, Theorem 1 and Theorem 2. Thus, the angle differences across
the incidence matrix E can be expressed as E = ET EC
the lines in a composite cutset in the power network are
where ET ∈ Rn×(n−1) and EC ∈ Rn×(l−n+1) are the
subjected to the cycle constraint rather than evolving arbitrarily.
incidence matrices with respect to T (V, ET ) and C(V, EC ),
This inspires a proof of Proposition 1 below.
respectively. We will show that Φ and O can be expressed
Proof of Proposition 1: Suppose a fault happens and the
in terms of ET and EC . To do so, we first introduce the
post-fault system (2) is unstable. In this case, the angles of a
following lemma.
group of generators will diverge from those of the remaining
Lemma 3: [44] EC can be re-expressed as EC = ET TC ,
group of generators (see [32]). Thus, the angle differences
where TC is the transformation matrix defined below
across some lines will approach to infinity, e.g., some lines
TC = (ETT ET )−1 ETT EC ∈ R(n−1)×(l−n+1) . (16) in the path between the two groups of generators. We select
one line, say ek , from those lines with unbounded angle
Lemma 3 describes the relationship between ET and EC . differences, and consider the following two cases:
By the definition of fundamental cycle, the entries of TC take 1) If line ek is non-cyclic, then it forms a simple cutset
values as follows: (TC )k1 k2 = 1 or (TC )k1 k2 = −1 if the edge by the first statement of Theorem 1. It is a critical cutset
ek1 ∈ ET and edge ek2 ∈ EC are concyclic in a fundamental because of its unbounded angle difference. This is the case
cycle. The sign depends on their assigned fictitious orienta- in Proposition 1.
tions. Otherwise (TC )k1 k2 = 0. By this transformation, we 2) If line ek is cyclic, then we show that a composite critical
can obtain the explicit form of O f in terms of ET and EC , cutset will be formed by line e and some of its concyclic lines.
k
which is addressed in the following theorem. Suppose there are totally m different cycles, say E1cyc , ..., Em
cyc
Theorem 3: O f = −TC = −(ETT ET )−1 ETT EC . cyc
that contain ek , i.e., ek ∈ E , i = 1, ..., m. Due to the cycle
Proof: By the property of cycles given in Section III-A, constraint, for each cycle Ei cyc , there exists at least one line
i
we have EO = ET Of + EC = 0. Substituting EC = ET TC e 6= e , e ∈ E cyc such that the angle difference across eki
ki k ki i
into this equation gives is also unbounded. Then, all such lines form the set Eek as used
ET (O f + TC ) = 0. (17) in Theorem 2. According to Theorem 2, we can conclude that
at least one composite cutset is formed by line ek together
For a connected graph, ET ∈ R(n−1)×l has full column rank with some lines in Eek , and obviously it is a critical cutset. 
as rank(ET ) = n− 1 [41]. Then, (17) implies that O f + TC = Remark 4: The above analysis reveals the formation mech-
0 [45], which leads to O f = −TC . anism of critical cutset. The loss of synchronism over cutsets,
There is another expression of O f in terms of the first which is indeed a collective behaviour of bus angles, is much
n − 1 rows of the incidence matrix E, which is from a node attributed to the new properties of cutsets and cycle constraint.
perspective (see [41, Theorem 5.6] for the details). Theorem 3 Further, this proof makes Proposition 1 a more solid statement
provides an equivalent but new way to construct O f from the for transient stability analysis, so does it for the results induced
perspective of edges. By Theorem 3, we obtain the explicit by Proposition 1 as aforementioned, e.g., the coherent groups
forms of O and Φ as follows of generators and dynamical equivalents [36, 37] that are
−(ETT ET )−1 ETT EC
 
commonly used for the simplification of stability analytics and
O= (18a)
Il−n+1 simulation.
 
In−1
Φ= . (18b)
ECT ET (ETT ET )−1 V. A PPLICATION II: A N IMPROVED CUTSET INDEX
In the next two sections, the above results will be applied Another important question related to critical cutsets is how
to the analysis of power system transient stability that lead to to identify the vulnerable cutset and the relevant UEP that has
a theoretical explanation of critical cutset phenomenon and an low potential energy. Answers to this question are useful for
improved cutset index. critical energy estimation. The identification is usually done
8 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

by using the CI [9, 10]. In this section, we will discuss the Despite the intuitive interpretation of the CI, it has a
deficiencies of the CI by the properties of cutsets and cycle problem that the angle differences across the lines in the cutset
constraint, and then make some improvements. are assigned independently (see (21)). As already shown by the
properties of cutsets and cycle constraint, the angle differences
across the lines in a composite cutset are interrelated. Directly
A. The cutset index and its deficiencies
summing up the area Ak+ or Ak− may not appropriately
We first give a brief review of the CI (referring to [9] for the reflect the real situations, for instance, it may not be physically
details). To find the entire set of UEPs is a hard problem as realizable that the angle differences across two lines ek1 , ek2
the number of UEPs is large even for small-size systems. The reach emax max
k1 , ek2 at the same time. In the zero power flow
lack of efficient and reliable numerical approaches adds further case where P = 0 and σ s = 0, it can be shown that the
difficulties [46]. The CI provides an alternative to approximate points (σ + (xcut ), 0) and (σ − (xcut ), 0) accurately represent
the potential energy at UEPs, which avoids the difficulties in the relevant UEPs [9], and we can also expect them to be
directly calculating UEPs. As mentioned in Section II-B, UEPs good approximations of the relevant UEPs in light load cases.
are closely related to cutsets. For a cutset xcut , the relevant However, in heavy load cases, significant error may occur
UEPs can be approximated by the points (σ + (xcut ), 0) and as σ + (xcut ) and σ − (xcut ) may seriously violate the cycle
(σ − (xcut ), 0), where σ + = [σk+ ] ∈ Rl and σ − = [σk− ] ∈ Rl constraint, which appeals for improvements.
are in terms of σkmax , σkmin and σks as follows
 max
 σk , xcut k =1
+
σk = σkmin , xcut
k = −1 (21a)
 s
σk , xcut
k = 0
 max
 σk , xcut k = −1
σk− = σkmin , xcut
k =1 (21b)
 s
σk , xcut
k = 0.

Accordingly, the CI is defined as the potential energy at these


approximate points
CI(xcut ) = min{VP+ (xcut ), VP− (xcut )} (22a) Figure 4. An illustration of the conventional cutset index.
X X
VP+ = VP,k (σkmax ) + VP,k (σkmin ) (22b)
xcut
k =1 xcut
k =−1

VP− =
X
VP,k (σkmin ) +
X
VP,k (σkmax ) (22c) B. The improved index
xcut
k =1 xcut
k =−1
For the cutset E cut , we define the ICI with the cycle
constraint included as follows
and the potential energy at the lowest UEP is estimated by the
minimum CI among all cutsets ICI(xcut ) = min {ICI+ (xcut ), ICI− (xcut )} (24)
min cut
CI = min CI(x ) (23)
xcut ∈X cut where ICI+ (xcut ) is the optimal value to
where X cut is the set of line selection vectors representing
X
ICI+ (xcut ) = max VP,k (σk ) (25a)
cutsets. Note that the UEPs relevant to multiple cutsets usually σ
xcut
k 6=0
have greater potential energy than those relevant to one cutset,
only the latter ones are considered in the CI. s.t. OT σ = 0 (25b)
Moreover, the CI can be interpreted geometrically. For line σk = σks , xcut
k = 0 (25c)
ek , VP,k (σkmax ) and VP,k (σkmin ) are equal to the area Ak+ and π 3π
≤ σk ≤ , xcut
k =1 (25d)
Ak− in Fig. 4, respectively, which are indeed the maximum 2 2
potential energy increment for the two possible angle deviation 3π π
− ≤ σk ≤ − , xcut k = −1 (25e)
directions. Further, since V̇ (σ, ω) ≤ 0, i.e., the system energy 2 2
is non-increasing with time, the potential energy increase
and ICI− (xcut ) is the optimal value to
indicates the deceleration of generator rotors. Thus, the area
Ak+ and Ak− can be regarded as the “deceleration area”
X
ICI− (xcut ) = max VP,k (σk ) (26a)
provided by line ek . The two candidates VP+ and VP− , which σ
xcut
k 6=0
sum up the potential energy of each line in the cutset, can be
thought of as the “deceleration area” provided by the cutset s.t. O T σ = 0 (26b)
[10]. The CI takes the smaller one between VP+ and VP− to σk = σks , =0xcut
k (26c)
characterize the more vulnerable angle deviation direction. A 3π π
− ≤ σk ≤ − , xcut
k = 1 (26d)
small CI indicates a vulnerable cutset as its deceleration area 2 2
may be too small to decelerate generator rotors so that angle π 3π
≤ σk ≤ , xcut
k = −1. (26e)
separation occurs. 2 2
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 9

The potential energy at the lowest UEP is estimated by the Remark 5: The inclusion of cycle constraint makes that all
minimum ICI the feasible points for problems (25) and (26) are physically
realizable. Thus, the optimization process for (25) and (26) can
ICImin = min ICI(xcut ). (27)
xcut ∈X cut be regarded as the system trajectory reaching a local maximum
on the potential energy surface. The optimal solution to (24),
We explain problem (25) and similar interpretations apply to
(26). Problem (25) aims to find the maximum potential energy say σ ∗ , is the local maximum with the “lowest energy barrier
to escape”. This is similar to the concept of the maximum
of the cutset E cut given a specific angle deviation direction.
Constraint (25b) is the cycle constraint. Constraint (25c) potential energy along the fault-on trajectory in the potential
energy boundary surface (PEBS) method. By the results on
implies that the angle difference across any line ek ∈ / E cut
s the PEBS method [48, 49], the point (σ ∗ , 0) lies close to the
is fixed as σk . This setting is justified by the aforementioned
observation on UEPs [38]. Constraints (25d) and (25e) deter- relevant UEP. Hence the ICI gives a good estimation of the
potential energy at the relevant UEP.
mine that the angle deviation direction of line ek ∈ E cut with
Remark 6: Since the point σ ∗ satisfies the cycle constraint,
xk = 1 is from π2 to 3π 2 , and the angle deviation direction of it can also provide a quality initial value for Newton-Raphson
line ek ∈ E cut with xk = −1 is opposite, i.e., from − 3π 2 to iteration to find the relevant UEP. We show that the initial
− π2 . Here we use the intervals [ π2 , 3π ] and [− 3π
, − π
] instead
2 2 2 value of bus angle vector θ ∗ can be found from σ ∗ by the
of specific values σkmax , σkmin as in the definition of CI. The
relation σ ∗ = E T θ ∗ . Select a spanning tree T (V, ET ) and
adoption of these two intervals originates from two aspects.
take bus 1 as the reference bus, i.e., θ1∗ = 0. It follows that
First, to keep the angle differences across lines lying in the
σT∗ = (ETr )T θr∗ , where ETr ∈ R(n−1)×(n−1) consists of the
interval [− π2 , π2 ] is a basic requirement for system normal
last n − 1 rows of ET and θr∗ ∈ Rn−1 consists of the last
operation, and such an interval is not suitable to approximate
n−1 rows of θ ∗ . Since the matrix ETr is nonsingular [41], we
UEPs. Second, it follows from the literature [28, 47] that the T
obtain θ ∗ by θ ∗ = 0 (σT∗ )T (ETr )−1 . As aforementioned,

intervals [ π2 , 3π 3π π
2 ] and [− 2 , − 2 ] indicate high risk of loss of
synchronism and hence are reasonable to describe the angle (σ ∗ , 0) is close to the relevant UEP and thus θ ∗ is a good
separation developing from the cutset. initial guess.
In addition, problem (25) can be further simplified by using Remark 7: Note that it is not cost effective to traverse over
the properties of cutsets. For a simple cutset, the ICI gives all possible cutsets (approximately 2n ) to rigorously find the
the same answer as the CI since the angle difference across minimum ICI of a practical power system. In practice, the
the non-cyclic line in the cutset is not subjected to the cycle cutsets of interest are those with few lines and heavy power
constraint. For a composite cutset E cut , as Theorem 1 implies flow [50]. The efficient graph-partitioning algorithm proposed
that the lines in it are concyclic in pairs, we can accordingly in [14], which seeks the maximum power flow-minimum
define a matrix O x ∈ Rl×nx to simplify the expression of the cutset in power networks, can be a proper prefilter to select
cycle constraint, where nx is the number of different line pairs candidate cutsets for the ICI analysis.
in the cutset. Suppose the j-th column of O x represents the
constraint for line pair ek1 , ek2 ∈ E cut , then it takes value as


 1, k = k1
1, k = k2 , xcut cut
k1 xk2 = −1

(O x )kj = cut cut (28)

 −1, k = k2 , xk1 xk2 = 1
0, k 6= k1 , k2 .

By using the matrix O x , constraints (25b) and (25c) can be


replaced by
O Tx σ = O Tx σ s . (29) Figure 5. An illustration of the improved cutset index.

As the number of lines in a cutset is usually much smaller


VI. C ASE STUDY
than the number of total fundamental cycles, (29) significantly
reduces the problem dimension especially for highly meshed A. IEEE 6-bus test system
networks. We first take the IEEE 6-bus test system to verify the
To illustrate the ICI, we recall the cutset in Fig. 3 and obtained results. This small-size system provides good visual-
suppose σ1s =30◦ , σ5s =-10◦. When the angle difference across ization for the analysis. The system diagram is shown in Fig.
line e1 moves from 30◦ to 150◦ to obtain the deceleration 6, where the transmission lines are assigned with fictitious
area A1+ , the angle difference across line e5 actually moves orientations. Bus 1, bus 2 and bus 3 are the augmented
from -10◦ to -130◦ due to the cycle constraint, as shown generator internal buses and the lines e12 , e13 , e14 represent
in Fig. 5 (here we fix the angle differences across the other transient reactances. The bus and line parameters are listed in
lines just for illustration). The deceleration area contributed by Table I and Table II, respectively. We select the five generator-
line e5 is A∗5− rather than A5− . Thus, the ICI avoids such a oriented cutsets marked in Fig. 6 to test the performance of
contradiction happens in the CI that the angle difference across ICI in a light load case and a heavy load case given in Table II.
line e1 reaches 150◦ and meanwhile the angle difference across For simplicity, the data related to lines e12 , e13 , e14 are omitted
line e5 reaches -170◦ . in the following analysis as they are not real physical links.
10 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

In the light load case, the CI and ICI data are listed in Table I
the 3rd and 4th columns in Table III. The ICI and CI are IEEE 6- BUS SYSTEM : BUS PARAMETERS
almost the same. Both of the two indices identify Cutset 1
Bus Pi in light case Pi in heavy case Vi Mi Di
as the most vulnerable cutset, the ICI gives 25.45 as the
1 1.05 8.40 1.05 10 2
estimation of the potential energy at the lowest UEP and the CI
2 0.55 4.40 1.05 20 2
gives 25.46. Considering that the actual potential energy at the
3 0.65 5.20 1.07 30 2
lowest UEP is 25.45, both the ICI and CI perform well. The
4 -0.05 -0.40 1.05 – 2
angle differences across lines at the operating point, lowest
5 -0.05 -0.40 1.05 – 2
UEP, minimum CI and minimum ICI are listed in Table IV.
6 -0.05 -0.40 1.07 – 2
As expected, there exist only slight differences among the last
7 -0.70 -5.60 1.00 – 2
three columns. The ICI gets a little closer to the lowest UEP 8 -0.70 -5.60 1.00 – 2
than the CI, however, it is not remarkable. 9 -0.70 -5.60 1.00 – 2
We proceed to the heavy load case, the corresponding CI
and ICI data are listed in the 5th and 6th columns in Table
III. The effect of the cycle constraint on the cutsets leads to Table II
significant differences between the ICI and CI. Both the ICI IEEE 6- BUS SYSTEM : LINE PARAMETERS
and CI identify Cutset 1 as the most vulnerable one, the ICI
Line Terminals |Yij | Line Terminals |Yij |
gives 7.71 as the estimation of the potential energy at the
e1 (4,5) 5.00 e8 (6,8) 3.85
lowest UEP and the CI gives 8.31. Direct calculation shows
e2 (4,7) 5.00 e9 (6,9) 10.00
that the potential energy at the lowest UEP is 7.59, so the ICI
e3 (4,8) 3.33 e10 (7,8) 2.50
provides a better estimation. For more detailed comparison,
e4 (5,6) 4.00 e11 (8,9) 3.33
we list the angle differences across lines at the operating point,
e5 (5,7) 10.00 e12 (1,3) 50.00
lowest UEP, minimum CI and minimum ICI in Table V. It can e6 (5,8) 3.33 e13 (2,4) 50.00
be seen that the angle differences across lines e2 , e3 reach their e7 (5,9) 5.00 e14 (3,6) 50.00
maximum values emax 2 , emax
3 at the minimum CI, while they
max max
exceed e2 , e3 at the minimum ICI and get much closer to
the lowest UEP. Also, we note that the exact UEPs in Table IV Table III
and Table V can be found by using the angle differences across IEEE 6- BUS SYSTEM : CI AND ICI IN LIGHT AND HEAVY CASES
lines at the minimum ICI as the initial value. It validates the
Light load Heavy load
adoption of intervals [ π2 , 3π 3π π
2 ] and [− 2 , − 2 ] in ICI to obtain Cutset No. Set of lines
CI ICI CI ICI
a more precise approximation of UEPs.
1 {e1 , e2 , e3 } 25.46 25.45 8.31 7.71
To further show the higher effectiveness of ICI than CI,
2 {e1 , e4 , e5 , e6 , e7 } 56.98 56.92 48.83 44.14
we set the following system initial state: the angle differences
3 {e4 , e8 , e9 } 36.76 36.74 25.26 24.23
across lines are equal to their values at the minimum ICI (see 4 {e1 , e3 , e5 , e10 } 43.14 43.04 40.03 33.24
the last column in Table V) and the rotor speed deviations of 5 {e4 , e7 , e8 , e11 } 34.09 34.05 33.13 30.66
the generators at bus 1, bus 2 and bus 3 are 0.091, 0.110 and
0.023 rad/s, respectively. Then, the total system energy at such
an initial state is 7.88, which is greater than the ICI but less
than the CI. The system is judged to be stable by the CI, while In addition, we simulate another disturbance in the heavy
the ICI indicates the risk of instability. The ICI turns out to load case to illustrate the critical cutset phenomenon. The
be better as Fig. 7 shows that the system trajectories finally unstable post-fault trajectories after bus 5 is severely disturbed
diverge and Cutset 1 becomes the critical cutset. are depicted in Fig. 8. In the early period of post-fault dynam-
ics (see the zoomed figure in the top left), the disturbance at
bus 5 induces large angle deviations on lines e1 , e4 , e5 , e6 , e7 ,
i.e., Cutset 2. In a larger time scale, the angle differences
across all these lines but line e1 gradually converge. As the
angle difference across line e1 approaches to infinity, the angle
differences across another two lines e2 , e3 , which are concyclic
with line e1 , diverge as well. Consequently, Cutset 1 becomes
critical and instability occurs. The whole process coincides
with our deduction on Proposition 1.
Furthermore, the ICI indicates the critical cutset in this case.
At the post-fault initial point, the kinetic energy is 8.59 and
the potential energies of Cutset 1 and Cutset 2 are 3.12 and
1.52, respectively. Then, the possible potential energy increase
Figure 6. Diagram of the IEEE 6-bus system (including generator internal for Cutset 1 can be evaluated by the difference between its ICI
buses). and initial potential energy 7.71-3.12=4.59, which is less than
the initial kinetic energy. For Cutset 2, the possible potential
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 11

energy increase is 44.14-1.52=42.62, which is greater than the


initial kinetic energy. It indicates that Cutset 2 has the capacity 800
to “absorb” the kinetic energy when its line angles tend to 700
separate, while Cutset 1 has a high risk of becoming a critical

Line angle difference/degree


cutset. To see it in detail, the trajectories of the potential 600
energies of Cutset 1 and Cutset 2 and kinetic energy are 500
depicted in Fig. 9. We observe that the ICI gives a quite precise
upper bound estimation of the potential energy trajectories. As 400

aforementioned, the angle differences across Cutset 2 continue Cutset 1


300
growing and tend to separate in the early post-fault period. line e line e1
3
Meanwhile, the potential energy of Cutset 2 surpasses the 200
line e2
kinetic energy and reaches a maximum around 0.2s (see the 100
zoomed figure in the bottom left in Fig. 9), which consequently
0
prevents angle separation. We also check that the signs of
(σk − σks )σ̇k of the lines e4 , e5 , e6 , e7 in Cutset 2 change from −100
0 1 2 3 4 5 6 7 8
positive to negative around 0.2s, which implies the tendency Time/s
to converge and coincides with the potential energy trajectory.
On the other hand, as the angle differences across Cutset 1 Figure 7. The system trajectories starting from the set initial values.
grow later, Cutset 1 fails to decelerate generator rotors so that
loss of synchronism occurs. This case shows the possibility
of online stability prediction by combining the ICI and PMU 800 200
line e1

Angle difference across line/degree


data, which will be a direction in future works. 700 100
line e
7
Table IV 600 0
IEEE 6- BUS SYSTEM : COMPARISON OF ANGLE DIFFERENCE IN LIGHT
LOAD CASE 500−100 line e4 line e6 line e
5
0 0.1 0.2 Cutset 1
400
σk at the σk at the σk at the
Line σks
lowest UEP minimum CI minimum ICI 300 line e1
e1 2.71◦ 174.08◦ 177.29◦ 174.67◦ line e2
200
e2 4.49◦ 176.55◦ 175.51◦ 176.45◦ line e3
e3 5.38◦ 178.12◦ 174.62◦ 177.34◦ 100
e4 0.29◦ 0.89◦ 0.29◦ 0.29◦
0
e5 1.79◦ 2.47◦ 1.79◦ 1.79◦
e6 2.67◦ 4.04◦ 2.67◦ 2.67◦ −100
0 1 2 3 4 5 6 7 8
e7 2.70◦ 3.27◦ 2.70◦ 2.70◦ Time/s
e8 2.39◦ 3.14◦ 2.39◦ 2.39◦
e9 2.42◦ 2.38◦ 2.42◦ 2.42◦ Figure 8. The system trajectories after severe disturbance at bus 5.
e10 0.89◦ 1.57◦ 0.89◦ 0.89◦
e11 0.03◦ -0.76◦ 0.03◦ 0.03◦
50
ICI of Cutset 2
Table V
IEEE 6- BUS SYSTEM : COMPARISON OF ANGLE DIFFERENCE IN HEAVY
0
LOAD CASE ICI of Cutset 1

σk at the σk at the σk at the


Energy

Line σks
lowest UEP minimum CI minimum ICI −50
e1 23.68◦ 129.84◦ 156.32◦ 134.43◦ 40

e2 38.42◦ 149.13◦ 141.58◦ 149.18◦ 30


e3 46.14◦ 162.14◦ 133.86◦ 156.90◦
−100 20
e4 2.63◦ 6.98◦ 2.63◦ 2.63◦
10 Potential energy of Cutset 1
e5 14.75◦ 19.29◦ 14.75◦ 14.75◦
Potential energy of Cutset 2
0
e6 22.47◦ 32.30◦ 22.47◦ 22.47◦ 0 0.05 0.1 0.15 0.2 Kinetic energy
e7 22.35◦ 26.56◦ 22.35◦ 22.35◦ −150
0 1 2 3 4 5 6 7 8
e8 19.83◦ 25.32◦ 19.83◦ 19.83◦ Time/s
e9 19.72◦ 19.58◦ 19.72◦ 19.72◦
e10 7.72◦ 13.01◦ 7.72◦ 7.72◦ Figure 9. The potential and kinetic energy after severe disturbance at bus 5.
e11 -0.12◦ -5.74◦ -0.12◦ -0.12◦
12 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

B. IEEE 39-bus test system Table VIII


IEEE 39- BUS TEST SYSTEM : COMPARISON OF ANGLE DIFFERENCE
We also test on the IEEE 39-bus system to show the
proposed approach applies to systems with practical size. σk at the σk at CI σk at ICI
Line |Yij | σks
The system diagram is shown in Fig. 10. The active power relevant UEP of Cutset 1 of Cutset 1
generations and loads are listed in Table VI, which is a heavy (26,28) 21.10 -39.99◦ -123.28◦ -140.01◦ -125.22◦
load case. We set the voltage magnitude of all buses to be (26,29) 16.00 -57.16◦ -144.19◦ -122.84◦ -142.39◦
1.0 p.u., and we refer to the MATPOWER package [51] for σk at the σk at CI σk at ICI
line parameters. Two candidate cutsets are marked in Fig. 10, Line |Yij | σks
relevant UEP of Cutset 2 of Cutset 2
both of which have few lines and heavy power flow by the (21,22) 71.43 -14.54◦ -155.44◦ -165.46◦ -161.36◦
selection criterion in [14]. For these two cutsets, the CI, ICI (23,24) 28.57 21.72◦ 182.37◦ 158.28◦ 168.54◦
and potential energy at the relevant UEP are shown in Table
VII, where the ICI performs better than the CI. This is further
validated by the angle difference data listed in Table VIII. It VII. C ONCLUSION
can be seen that the estimation given by the ICI is much closer
to the relevant UEP than that by the CI. We have revealed some new characterization of cutsets in
undirected graphs. A close link between cutsets and cycles
has been established. We prove that a single non-cyclic edge
forms a simple cutset. For any pair of edges in a composite
cutset, there exists a cycle that contains these two edges
but does not contain the other edges in the cutset. We also
show that any cyclic edge can form a composite cutset with
some of its concyclic edges. These properties are universal
to any connected undirected graphs. Then, we apply these
new properties of cutsets to power system transient stability
analysis. We give a theoretical explanation of the widely
adopted proposition that the transient instability is developed
from the angle separation across the critical cutset. We find that
this instability pattern is attributed directly to the topological
properties of the underlying power network. Furthermore, we
propose the ICI that can be used to identify the vulnerable
cutset and estimate the critical energy. The obtained results
Figure 10. Diagram of the IEEE 39-bus system. have been validated by the numerical simulations on two IEEE
test systems, which shows that the ICI performs better than
the conventional cutset index, especially in heavy load cases.
Table VI Also, the examples of unstable post-fault trajectories match up
IEEE 39- BUS SYSTEM : BUS PARAMETERS well with our explanation of the critical cutset phenomenon.

bus Pi bus Pi bus Pi bus Pi


ACKNOWLEDGEMENT
1 -3.00 11 0.00 21 -9.00 31 21.00
2 0.00 12 0.00 22 0.00 32 19.50 The authors wish to thank the reviewers and editors for their
3 -9.00 13 0.00 23 -7.50 33 18.00 constructive comments that improve this manuscript.
4 -18.00 14 0.00 24 -9.00 34 15.00
5 0.00 15 -10.50 25 -13.50 35 16.50 R EFERENCES
6 0.00 16 -15.00 26 -18.00 36 19.50 [1] B. Bollobás, Modern Graph Theory. Springer, 1998.
7 -16.50 17 0.00 27 -9.00 37 28.50 [2] M. X. Goemans and D. P. Williamson, “Improved approximation algo-
rithms for maximum cut and satisfiability problems using semidefinite
8 -18.00 18 -16.50 28 -6.00 38 37.50 programming,” Journal of the ACM, vol. 42, no. 6, pp. 1115–1145, 1995.
9 0.00 19 0.00 29 -4.50 39 9.00 [3] P. Kundur, J. Paserba, V. Ajjarapu, G. Andersson, A. Bose, C. Canizares,
10 0.00 20 -19.50 30 18.00 N. Hatziargyriou, D. Hill, A. Stankovic, C. Taylor, T. Van Cutsem,
and V. Vittal, “Definition and classification of power system stability
IEEE/CIGRE Joint Task Force on stability terms and definitions,” IEEE
Trans. Power Syst., vol. 19, no. 3, pp. 1387–1401, Aug 2004.
[4] D. J. Hill and A. R. Bergen, “Stability analysis of multimachine power
Table VII networks with linear frequency dependent loads,” IEEE Trans. Circuits
IEEE 39- BUS SYSTEM : CI AND ICI Syst., vol. 29, no. 12, pp. 840–848, Dec 1982.
[5] Y. Song, D. J. Hill, and T. Liu, “Small-disturbance angle stability
Potential energy analysis of microgrids: A graph theory viewpoint,” in Proc. IEEE Conf.
Cutset No. Set of lines CI ICI Control Appl., 2015, pp. 201–206.
at relevant UEP
[6] Y. Song, D. J. Hill, and T. Liu, “Network-based analysis of small-
1 {(26,28), (26,29)} 9.40 10.60 9.52 disturbance angle stability of power systems,” IEEE Trans. Control Netw.
2 {(21,22), (23,24)} 117.50 118.95 118.34 Syst., to be published. doi: 10.1109/TCNS.2017.2654162.
[7] Y. Ni, V. Vittal, and W. Kleimann, “System separation mechanism in
the neighbourhood of a relevant type-n UEP using the normal form of
SONG et al.: CHARACTERIZATION OF CUTSETS IN NETWORKS WITH APPLICATION TO TRANSIENT STABILITY ANALYSIS OF POWER SYSTEMS 13

vector fields,” in IEE Proc. Gener. Transm. Distrib., vol. 145, no. 2, Mar to (optimal) frequency regulation in power grids with time-varying
1998, pp. 139–144. voltages,” Automatica, vol. 64, pp. 240–253, 2016.
[8] M. Bürger, D. Zelazo, and F. Allgöwer, “Hierarchical clustering of [34] M. Pavella, D. Ernst, and D. Ruiz-Vega, Transient Stability of Power
dynamical networks using a saddle-point analysis,” IEEE Trans. Autom. Systems: A Unified Approach to Assessment and Control. Springer
Control, vol. 58, no. 1, pp. 113–124, Jan 2013. Science & Business Media, 2000.
[9] A. R. Bergen and D. J. Hill, “A structure preserving model for power [35] Y. Xue, L. Wehenkel, R. Belhomme, P. Rousseaux, M. Pavella, E. Eu-
system stability analysis,” IEEE Trans. Power App. Syst., vol. PAS-100, xibie, B. Heilbronn, and J.-F. Lesigne, “Extended equal area criterion
no. 1, pp. 25–35, Jan 1981. revisited,” IEEE Trans. Power Syst., vol. 7, no. 3, pp. 1012–1022, August
[10] K. S. Chandrashekhar and D. J. Hill, “Dynamic security dispatch: Basic 1992.
formulation,” IEEE Trans. Power App. Syst., no. 7, pp. 2145–2154, July [36] F. F. Wu and Y.-k. Tsai, “Identification of groups of ǫ-coherent gen-
1983. erators,” IEEE Trans. Circuits Syst., vol. 30, no. 4, pp. 234–241, Apr
[11] G. Cai, K. Chan, W. Yuan, and G. Mu, “Identification of the vulnerable 1983.
transmission segment and cluster of critical machines using line transient [37] F. F. Wu and N. Narasimhamurthi, “Coherency identification for power
potential energy,” Int. J. Elect. Power Energy Syst., vol. 29, no. 3, pp. system dynamic equivalents,” IEEE Trans. Circuits Syst., vol. 30, no. 3,
199–207, 2007. pp. 140–147, Mar 1983.
[12] K. Padiyar and S. Krishna, “Online detection of loss of synchronism [38] F. Prabhakara and A. El-Abiad, “A simplified determination of transient
using energy function criterion,” IEEE Trans. Power Del., vol. 21, no. 1, stability regions for Lyapunov methods,” IEEE Trans. Power App. Syst.,
pp. 46–55, Jan 2006. vol. 94, no. 2, pp. 672–689, March 1975.
[13] J. H. Chow, A. Chakrabortty, M. Arcak, B. Bhargava, and A. Salazar, [39] S. Seshu and M. B. Reed, Linear Graphs and Electrical Networks.
“Synchronized phasor data based energy function analysis of dominant Addison-Wesley, 1961.
power transfer paths in large power systems,” IEEE Trans. Power Syst., [40] C. D. Godsil and G. Royle, Algebraic Graph Theory. Springer, 2001.
vol. 22, no. 2, pp. 727–734, May 2007. [41] R. B. Bapat, Graphs and Matrices. Springer, 2010.
[14] B. C. Lesieutre, S. Roy, V. Donde, and A. Pinar, “Power system extreme [42] V. G. Vizing, “Some unsolved problems in graph theory,” Russian
event screening using graph partitioning,” in Proc. North Amer. Power Mathematical Surveys, vol. 23, no. 6, pp. 125–141, 1968.
Symp., Sept 2006, pp. 503–510. [43] J. W. Essam and M. E. Fisher, “Some basic definitions in graph theory,”
[15] A. Pinar, J. Meza, V. Donde, and B. Lesieutre, “Optimization strategies Reviews of Modern Physics, vol. 42, no. 2, p. 271, 1970.
for the vulnerability analysis of the electric power grid,” SIAM Journal [44] J. Sandhu, M. Mesbahi, and T. Tsukamaki, “Cuts and cycles in relative
on Optimization, vol. 20, no. 4, pp. 1786–1810, 2010. sensing and control of spatially distributed systems,” in Proc. Amer.
[16] E. Scholtz, G. Verghese, and B. Lesieutre, “Observer-based monitors Control Conf., June 2005, pp. 73–78.
for electromechanical dynamics in power networks,” in Proc. Power [45] R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge University
Syst. Comput. Conf., Aug 2005, pp. 1–7. Press, 2012.
[17] H. You, V. Vittal, and X. Wang, “Slow coherency-based islanding,” IEEE [46] D. K. Molzahn, B. C. Lesieutre, and H. Chen, “Counterexample to a
Trans. Power Syst., vol. 19, no. 1, pp. 483–491, Feb 2004. continuation-based algorithm for finding all power flow solutions,” IEEE
[18] G. Xu and V. Vittal, “Slow coherency based cutset determination Trans. Power Syst., vol. 28, no. 1, pp. 564–565, February 2013.
algorithm for large power systems,” IEEE Trans. Power Syst., vol. 25, [47] C. J. Tavora and O. J. Smith, “Stability analysis of power systems,”
no. 2, pp. 877–884, May 2010. IEEE Trans. Power App. Syst., vol. PAS-91, no. 3, pp. 1138–1144, May
[19] J. Li, C.-C. Liu, and K. P. Schneider, “Controlled partitioning of a 1972.
power network considering real and reactive power balance,” IEEE Trans. [48] R. A. Schlueter, “A local potential energy boundary surface method for
Smart Grid, vol. 1, no. 3, pp. 261–269, Dec 2010. power system atability assessment,” in Proc. IEEE Conf. Dec. Control,
[20] A. H. El-Abiad and K. Nagappan, “Transient stability regions of Dec 1985, pp. 58–64.
multimachine power systems,” IEEE Trans. Power App. Syst., vol. PAS- [49] H.-D. Chiang, F. F. Wu, and P. Varaiya, “Foundations of the potential
85, no. 2, pp. 169–179, Feb 1966. energy boundary surface method for power system transient stability
[21] N. Tsolas, A. Arapostathis, and P. P. Varaiya, “A structure preserving analysis,” IEEE Trans. Circuits Syst., vol. 35, no. 6, pp. 712–728, Jun
energy function for power system transient stability analysis,” IEEE 1988.
Trans. Circuits Syst., vol. 32, no. 10, pp. 1041–1049, Oct 1985. [50] V. Donde, V. López, B. Lesieutre, A. Pinar, C. Yang, and J. Meza,
[22] N. Narasimhamurthi and M. Musavi, “A generalized energy function “Severe multiple contingency screening in electric power systems,” IEEE
for transient stability analysis of power systems,” IEEE Trans. Circuits Trans. Power Syst., vol. 23, no. 2, pp. 406–417, May 2008.
Syst., vol. 31, no. 7, pp. 637–645, Jul 1984. [51] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas, “Mat-
[23] I. A. Hiskens and D. J. Hill, “Energy functions, transient stability and power: Steady-state operations, planning, and analysis tools for power
voltage behaviour in power systems with nonlinear loads,” IEEE Trans. systems research and education,” IEEE Trans. Power Syst., vol. 26, no. 1,
Power Syst., vol. 4, no. 4, pp. 1525–1533, October 1989. pp. 12–19, February 2011.
[24] K. Padiyar, Structure Preserving Energy Functions in Power Systems:
Theory and Applications. CRC Press, 2013.
[25] R. J. Davy and I. A. Hiskens, “Lyapunov functions for multimachine
power systems with dynamic loads,” IEEE Trans. Circuits Syst. I, vol. 44,
no. 9, pp. 796–812, September 1997.
[26] T. Odun-Ayo and M. L. Crow, “Structure-preserved power system
transient stability using stochastic energy functions,” IEEE Trans. Power
Syst., vol. 27, no. 3, pp. 1450–1458, Aug 2012.
[27] F. Dörfler, M. Chertkov, and F. Bullo, “Synchronization in complex
oscillator networks and smart grids,” Proc. Nat. Acad. Sci., vol. 110,
no. 6, pp. 2005–2010, 2013.
[28] N. Ainsworth and S. Grijalva, “A structure-preserving model and suffi-
cient condition for frequency synchronization of lossless droop inverter-
based AC networks,” IEEE Trans. Power Syst., vol. 28, no. 4, pp. 4310–
4319, Nov 2013. Yue Song (S’14) received B.S. and M.S. degrees in
[29] T. L. Vu and K. Turitsyn, “A framework for robust assessment of power electrical engineering from Shanghai Jiao Tong Uni-
grid stability and resiliency,” IEEE Trans. Autom. Control, vol. 62, no. 3, versity, China, in 2011 and 2014, respectively. He
pp. 1165–1177, March 2017. is currently working toward the Ph.D. degree in the
[30] R. Olfati-Saber, A. Fax, and R. M. Murray, “Consensus and cooperation Department of Electrical and Electronic Engineering,
in networked multi-agent systems,” Proc. IEEE, vol. 95, no. 1, pp. 215– University of Hong Kong, Hong Kong.
233, January 2007. He received the Hong Kong Ph.D. Fellowship
[31] F. Dörfler and F. Bullo, “Synchronization in complex networks of phase from the Research Grants Council of Hong Kong,
oscillators: A survey,” Automatica, vol. 50, no. 6, pp. 1539–1564, 2014. and the CLP Fellowship in electrical engineering
[32] M. A. Pai, Energy Function Analysis for Power System Stability. from the University of Hong Kong during his Ph.D.
Springer Science & Business Media, 2012. studies. His research interests include power systems,
[33] S. Trip, M. Bürger, and C. De Persis, “An internal model approach stability analysis and dynamical networks.
14 IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

David J. Hill (M’76-SM’91-F’93-LF’14) received Tao Liu (M’13) received the B.E. degree from
the B.E. degree in electrical engineering and the Northeastern University, China, in 2003 and the
B.Sc. degree in mathematics from the University of Ph.D. degree from the Australian National Univer-
Queensland, Australia, in 1972 and 1974, respec- sity, Australia, in 2011.
tively, and the Ph.D. degree in electrical engineering From January 2012 to May 2012, he was a Re-
from the University of Newcastle, Australia, in 1976. search Fellow in the Research School of Engineering
He is the Chair of Electrical Engineering with the at the Australian National University. During this
Department of Electrical and Electronic Engineering, period, he also held a visiting scholar position in the
University of Hong Kong, where he directs the Cen- Centre for Future Energy Networks at the University
tre for Electrical Energy Systems and is the Program of Sydney, Australia. From June 2012 to August
Coordinator for the Multiuniversity RGC Theme- 2013, he worked as a postdoctoral fellow at the
Based Research Scheme Project on Sustainable Power Delivery Structures University of Groningen, the Netherlands. He moved to the University of
for High Renewables. He is also a part-time Professor and the Director of the Hong Kong in September 2013, and worked also as a postdoctoral fellow
Centre for Future Energy Networks, University of Sydney, Australia. From until June 2015, where he is currently a Research Assistant Professor in the
2005 to 2010, he was an Australian Research Council Federation Fellow with Department of Electrical and Electronic Engineering.
the Australian National University, and since 2006, he has been a Theme His research interests include power systems, dynamical networks, dis-
Leader of Complex Networks and the Deputy Director with the ARC Centre tributed control, event-triggered control and switched systems.
of Excellence for Mathematics and Statistics of Complex Systems.
He has held various positions at the University of Sydney since 1994,
including the Chair of Electrical Engineering until 2002 and again from
2010 to 2013, along with an ARC Professorial Fellowship. He has also held
academic and substantial visiting positions at the University of Melbourne; the
University of California at Berkeley; the University of Newcastle, Australia;
the University of Lund, Sweden; the University of Munich; the City University
of Hong Kong; and Hong Kong Polytechnic University. From 1996 to 1999
and 2001 to 2004, he served as the Head of the respective departments in
Sydney and Hong Kong.
Prof. Hill’s general research interests are in control systems, complex
networks, power systems, and stability analysis. His work is now mainly
on control and planning of future energy networks, and basic stability and
control questions for dynamic networks. Prof. Hill is a fellow of the Society
for Industrial and Applied Mathematics, USA; the Australian Academy
of Science; and the Australian Academy of Technological Sciences and
Engineering. He is also a Foreign Member of the Royal Swedish Academy
of Engineering Sciences.

View publication stats

You might also like