IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO.

2, MARCH 2021 923

A Graph Theoretic Approach to Power System

Vulnerability Identification
Reetam Sen Biswas , Student Member, IEEE, Anamitra Pal , Senior Member, IEEE,
Trevor Werho , Member, IEEE, and Vijay Vittal , Fellow, IEEE

Abstract—During major power system disturbances, when mul- flA Flow in edge el for the network flow solution A.
tiple component outages occur in rapid succession, it becomes FP The flow injected along path P .
crucial to quickly identify the transmission interconnections that
have limited power transfer capability. Understanding the impact G A set containing the locations of generator buses.
of an outage on these critical interconnections (called saturated cut- Ig Active power injected at a source vertex vg ∈ G.
sets) is important for enhancing situational awareness and taking k Total number of edges in cut-set K.
correct actions. This paper proposes a new graph theoretic ap- K Any cut-set in the power network.
proach for analyzing whether a contingency will create a saturated Ki ith cut-set associated with edge el ∈ E.
cut-set in a meshed power network. A novel feature of the proposed
algorithm is that it lowers the solution time significantly making K crit Limiting critical cut-set for edge el ∈ E.
the approach viable for real-time operations. It also indicates the L A set containing all the load buses.
minimum amount by which the power transfer through the crit- n Total number of indirect paths for edge el .
ical interconnections should be reduced so that post-contingency P This is a path (sequence of edges) from a source vertex
saturation does not occur. Robustness of the proposed algorithm to a sink vertex in the graph G.
for enhanced situational awareness is demonstrated using the
IEEE-118 bus system as well as a 17,000+ bus model of the PG1 Total active power generation in cluster C 1 .
Western Interconnection (WI). Comparisons made with different PG2 Total active power generation in cluster C 2 .
approaches for power system vulnerability assessment prove the PL1 Total active power demand in cluster C 1 .
utility of the proposed scheme for aiding power system operations
during extreme exigencies. PL2 Total active power demand in cluster C 2 .
ΔP 1 Net active power injection in cluster C 1 .
Index Terms—Graph theory, network flow, power system
ΔP 2 Net active power injection in cluster C 2 .
disturbances, power system vulnerability, saturated cut-set.
PK Total active power to be transferred across cut-set K.
rl Rating of edge el ∈ E.
NOMENCLATURE RK Total active power transfer capacity of cut-set K, ex-
cluding edge el , examined by feasibility test (FT).
cF T
Directed weight associated with edge el from vertex
l Tli Transfer margin of the ith saturated cut-set, associated
vlF towards vlT in the latent capacity graph (C).
with edge el .
cTl F Directed weight associated with edge el from vertex
Tl Transfer margin of the limiting critical cut-set associ-
vlT towards vlF in the latent capacity graph (C).
ated with edge el .
C1 The set of vertices contained in cluster 1.
T Cl Total additional active power transfer capability of the
C2 The set of vertices contained in cluster 2.
indirect paths of edge el .
CP Maximum extra flow that can be transferred along path
vg A vertex that has a source (or generator).
P from a source vertex towards a sink vertex.
vl A vertex that has a sink (or load).
Dl Active power withdrawn at a sink vertex vl ∈ L.
vlF The “from vertex” of edge el .
el lth edge in the edge set E.
vlT The “to vertex” of edge el .
E A set containing all edges of the power network.
V A vertex set containing all buses of the power network.
fl A directed weight associated with edge el from vertex
x Total number of cut-sets associated with el .
vlF towards vertex vlT in the flow graph (F).
y Total number of saturated cut-sets associated with el .
Manuscript received September 4, 2019; revised January 27, 2020, May 16,
z A variable denoting impedance of a branch.
2020, and June 23, 2020; accepted July 11, 2020. Date of publication July 20, G An undirected weighted graph of the power network.
2020; date of current version February 19, 2021. This work was supported by the F A directional flow graph of the power network.
Power System Engineering Research Center (PSERC) Grants S-74 and S-87.
Paper no. TPWRS-01322-2019. (Corresponding author: Reetam Sen Biswas.)
C A bidirectional latent capacity graph of the network.
The authors are with the School of Electrical, Computer, and En-
ergy Engineering, Arizona State University (ASU), Tempe, AZ 85287
USA (e-mail: rsenbisw@[Link]; [Link]@[Link]; twerho@[Link];
I. INTRODUCTION
[Link]@[Link]). NALYSIS of major blackouts has indicated that they in-
Color versions of one or more of the figures in this article are available online
at [Link]
Digital Object Identifier 10.1109/TPWRS.2020.3010476
A volve successive outages of power system assets [1]. For
example, the 1977 New York City blackout was caused by the

0885-8950 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See [Link] for more information.

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
924 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

loss of 11 transmission lines in 52 minutes. The Federal Elec- and structure of the power system [18]–[27]. In [18], Albert et al.
tricity Regulatory Commission (FERC) reported that one of the studied the structural vulnerability of the North American power
causes of the blackout was “the failure to recognize that a critical grid using a metric called the node degree, which refers to the
interconnection to the west was effectively unavailable” [2]. number of lines connected to a bus. Use of betweenness indices,
More recently, the initiating event for the 2011 U.S. Southwest which refer to the number of shortest paths traversing a given
blackout was the loss of the 500 kV Hassayampa-North Gila element, were explored in [19], [20]. Such purely topological
(H-NG) line, which then triggered a sequence of events that indices do not consider the electrical properties of the power
resulted in the blackout of San Diego [3]. Werho et al. stated network.
that a critical interconnection does not necessarily refer to a Modified centrality indices were used in [21] and [22] to assess
single line whose status can be monitored [4]; i.e., a critical in- the risk of blackouts/brownouts and systemic vulnerabilities,
terconnection can consist of multiple lines. Therefore, real-time respectively. In [23] and [24], different statistical measures such
vulnerability assessment for enhanced situational awareness of as the betweenness indices, node-degree, and geodesic distance
a power system that is suffering from multiple outages is a were used as possible alternatives to power flow techniques to
challenging task [5], [6]. quantify power system vulnerability during N-1 contingencies
The traditional approaches for improving situational aware- and cascading failures. In [25], Zhu et al. proposed a metric
ness are based on steady-state contingency analyses techniques called risk graph to better capture the cascade failure vulner-
that solve AC or DC power flows [7]–[11]. These techniques ability of the power system. Recently, Beyza et al. in [26]
cannot detect transient/dynamic stability related violations but investigated the structural vulnerability of the power system
can identify branch overloads and voltage violations. However, when successive N-1 contingencies progressively alter the net-
power flow-based contingency analysis (CA) is not fast enough work structure. These global vulnerability metrics (node degree,
to perform an exhaustive N-1 real-time contingency analysis betweenness indices, modified centrality indices, geodesic dis-
(RTCA) [7]. Therefore, power utilities select a subset of the tance) describe the vulnerability of the system by a single num-
contingencies for evaluation based on some pre-defined criteria ber. However, such indices do not convey meaningful actionable
[9], [10]. In [11], Huang et al. stated that the size of this subset information to an operator who is trying to prevent the system
has considerable impact on RTCA solution: a large subset is from collapsing! This is because these metrics do not consider
computationally burdensome, while a small subset might miss the physical manifestation of a vulnerability – a key issue that
critical scenarios. This can be a problem for real-time operations the proposed research seeks to address.
during extreme exigencies when multiple outages occur in rapid In [4], Werho et al. used a graph theory-based network flow
succession [4]. algorithm to identify the cut-set of minimum size between a
For managing extreme event conditions, a variety of ap- source-sink pair. A cut-set denotes the set of edges which when
proaches that can identify vulnerabilities quickly have been removed separates the graph into two disjoint islands; the size
proposed; these include statistical analyses ([12]–[16]), graph of the cut-set refers to the number of edges present in it. If
theoretic analyses ([4], [17]–[27]), and linear sensitivity-based the number of edges contained in the minimum sized cut-set
analyses ([28]–[33]). These types of analyses are suitable for progressively decreases, it indicates a structural weakness be-
exhaustive N-1 and potentially N-X evaluations. The proposed tween the selected source-sink pair. In [27], Beiranvand et al.
graph theoretic approach also belongs to this category of analy- presented a novel topological sorting algorithm to screen out
ses as it enhances situational awareness for real-time operations. coherent cut-sets. Coherent cut-sets denote the set of edges that
A brief overview of these other techniques ([12]–[33]) that partition the network, such that the power flows in the same
belong to this category is provided below. direction through all the edges. However, coherent cut-sets may
Dobson et al. in [12], [13] obtained statistics of cascading not be the only bottlenecks in a power system, as there may be
line outages from utilities to understand how cascades initiate a cut-set in which the power flows are not unidirectional, but a
and propagate in the power system. In [14], Rezaei et al. es- single outage limits the power transfer through it.
timated the risk of cascading failure with an algorithm called Bompard et al. used power transfer distribution factors
random chemistry. In [15], Rahnamay-Naeini et al. performed (PTDFs) and transmission line capacities for screening out
probabilistic analysis to understand the dynamics of cascading critical contingencies [28], [29]. Line outage distribution factors
failures. In [16], Hines et al. proposed an influence graph model (LODFs) have been used for quickly detecting an island forma-
to capture patterns of cascading failures in power systems and tion due to a multiple element contingency [30]. Werho et al.
validated the model using historical data. Instead of relying on used DC power flow based linear sensitivity analysis to detect
prior historical data, which may or may not be relevant for the an island formation due to a contingency [31]. In [32], [33],
present scenario, the proposed approach exploits knowledge of contingency screening was done using LODFs. These sensitivity
the current network conditions to identify the system’s critical indices capture the topological as well as the electrical properties
interconnections, the loss of which might trigger a cascade. of the power system and are useful for comparison with the
Graph theoretic approaches have found applications in a proposed approach (e.g., see Table V).
variety of fields [34]–[36]. Ishizaki et al. summarized the appli- The goal of this paper is to investigate if cut-sets will become
cations of graph theory for power systems modeling, dynamics, saturated (i.e., cannot transfer the required amount of power)
coherency, and control [17]. With regards to vulnerability assess- due to a would-be outage, irrespective of the direction in which
ment, graph theoretic approaches have focused on the topology power flows through different edges of the cut-set. Such cut-sets

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 925

the direct path from vertex (bus) vlF to vertex (bus) vlT . There
could be many other electrical paths to transfer power from
vlF to vlT . Any path that contains multiple edges (transmission
lines or transformers) from vlF to vlT is an indirect path. Let
there be n indirect paths between vertices (buses) vlF and vlT .
If all the n indirect paths combined do not have the capacity
to reroute fl units of power that was flowing through the direct
path, it implies that the loss of edge el would inevitably result
Fig. 1. Network connectivity between two vertices (buses). in post-contingency overloads. Based on this inference, a graph
theory-based network analysis tool is developed in this paper to
quickly detect violations of the type where the set of indirect
(termed saturated cut-sets henceforth) are the system’s critical paths do not have the extra capacity to carry the power that was
interconnections as they have limited power transfer capability. originally flowing through the direct path.
Essentially, this paper attempts to answer the following question: Contrary to traditional CA studies that detect if an outage
How to quickly make operators aware if a new contingency causes an overload on the remaining assets of the system [37],
will create saturated cut-sets in a meshed power network, after the primary goal of this research is to quickly detect if an
multiple component failures have occurred in rapid succession? outage overloads any cut-set of the power system. An overloaded
or saturated cut-set is one which transfers power beyond its
II. THEORETICAL BACKGROUND maximum power transfer capability. Let edges e1 , e2 , …, ek
A. Graph Theoretic Terminologies belong to cut-set K. If the power flowing through the different
edges of cut-set K are f1 , f2 , …, fk , and the ratings of those
In graph theoretic terminology, a power system can be rep- edges are r1 , r2 , … rk , then cut-set K will be called a saturated
resented by an undirected graph G(V , E), such that the buses cut-set if the following equation holds true:
are contained in the vertex set V and the transmission lines and
transformers are contained in the edge set E. The generators and 
k 
k
loads are the sources and sinks, respectively. The set G consists fl > rl , ∀el ∈ K (2)
of all vertices where a source is present and the set L consists l=1 l=1
of all vertices where a sink is present. The power injected at 
a source vg ∈ G is denoted by Ig and the power demand at a where, kl=1 fl is the actual power flowing through cut-set K

sink vl ∈ L is denoted by Dl . Now, every transmission asset and kl=1 rl is the maximum power that can flow through cut-set
(line or transformer) has an associated capacity called the asset K (limited by the ratings of the edges).
rating. To account for the asset ratings in the undirected graph At a given time, let us assume that PK units of power must
G(V , E), every edge el ∈ E is associated with a weight rl , be transferred through a cut-set, K, of a power network. Upon
where rl denotes the maximum power that can be transferred the loss of edge, el , that belongs to cut-set K, if the total power
through edge el . From the original graph G(V , E), we now transfer capability of the remaining edges of cut-set K is RK
create two graphs: the flow graph, F(V , E), and the latent such that RK < PK , it implies that the loss of el saturates cut-set
capacity graph, C(V , E). The flow graph, F(V , E), contains K. In such a situation, edge el is termed a special asset and the
information about the power flow through different edges of the cut-set K is said to be saturated by a negative transfer margin
network. If fl units of power flows through edge el from vertex of RK − PK due to the loss of the special asset, el .
vlF towards vertex vlT , a directed weight of fl is assigned to edge Now, let edge el be associated with x cut-sets of the network,
el in a direction from vlF to vlT . On the other hand, for edge, of which y cut-sets (y ≤ x) become saturated by a negative trans-
el , the latent capacity graph, C(V , E), provides information fer margin when el is lost (implying that y cut-sets of the network
regarding the extra flow that could be transferred from vlF to are saturated). As the y cut-sets may be saturated by different
vlT , and vice-versa. The weights associated with the edges of negative transfer margins, Tli , 1 ≤ i ≤ y, the objective here is
C(V , E) that provide information regarding the bidirectional to identify the cut-set that becomes saturated by the numerically
latent capacities are given by, largest negative transfer margin (i.e., Tl = max(|Tli |); 1 ≤ i ≤
 y ); this cut-set is henceforth referred to as the limiting critical
cF T
= rl − fl
l
(1) cut-set, K crit . Quickly identifying the limiting critical cut-set is
cTl F = rl + fl important because if appropriate preventive control actions are
taken so that the limiting critical cut-set is no longer saturated,
where, cF
l
T
is the latent capacity in the direction from vlF to vlT ,
the proposed approach, which is very fast, can be repeated
and cl is the latent capacity in the direction from vlT to vlF .
TF
multiple times until no limiting critical cut-sets are identified.
Note that this paper identifies the limiting critical cut-sets based
B. Research Scope on the thermal ratings of the different assets and the active
Let an edge el (transmission line or transformer) connect power flowing through them (power factor is set to unity for
vertices (buses) vlF and vlT as shown in Fig. 1. Since edge el the studies done here). However, the proposed network analysis
is a single element that joins vertices vlF and vlT , it is called tool is generic enough to incorporate line ratings obtained from

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
926 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

Fig. 2. Three valid graph theory-based flow graphs for the same system. “G” and “L” denote generation and load respectively; dotted line denotes a cut-set.

other analyses as well (such as, proxy limits based on power

Algorithm 1: Graph Theory-Based Network Flow Algo-
system stability criteria).
rithm.
i. Randomly select a source vertex vg ∈ G and a sink
III. GRAPH THEORY BASED NETWORK FLOW vertex vl ∈ L.
ii. Search C(V , E) to traverse the shortest unsaturated
The graph theory-based network flow algorithm is based on path P from vg to vl using breadth first search
the following assumptions: (1) power injections are known, and (BFS) [37].
(2) losses are negligibly small. Subject to these assumptions, iii. Use C to find the maximum extra flow, CP , that could
the goal is to generate network flows that can help detect if be transferred from vg to vl through path P .
a contingency saturates a cut-set. The graph theoretic network iv. Obtain the flow FP to be injected in F(V , E) along
flow algorithm is based on the principle: utilize the available path P from vg to vl as FP = min(Ig , Dl , CP ).
generation of the sources (generators) to satisfy the total demand v. Update weights of edges in graph F as fl = fl + FP ,
of the sinks (loads), without violating the asset ratings. The and in graph C as per (1), for all the edges that belong
network flows are obtained using Algorithm 1 described below. to path P .
At the start of the algorithm, edges in F(V , E) do not have any vi. Update available generation and unsatisfied demand at
weight, while the bidirectional weights of edges in C(V , E) are vg and vl as Ig : = Ig − FP and Dl : = Dl − FP .
equal to the corresponding asset ratings. vii. Depending upon the values of Ig and Dl , update the
The graph theory-based network flow algorithm obeys the law source and sink vertices in accordance with the
of conservation of energy, but it relaxes Kirchhoff’s voltage law following logic:
as it does not use impedances directly while building the network a. if Ig = 0 & Dl = 0, the source and sink vertices
flows; the impedances are accounted for indirectly through the are not changed;
asset ratings. The flow solution is also non-unique because de- b. if Ig = 0 & Dl = 0, a new source vertex, vg , is
pending on the order in which the sources and sinks are selected, selected from G, keeping the sink vertex, vl ,
there could be multiple valid flow solutions. However, since the unchanged;
network boundary conditions do not change (i.e., instantaneous c. if Ig = 0 & Dl = 0, a new sink vertex, vl , is
power injections are constant), the power transfer across any selected from L, keeping the source vertex, vg ,
cut-set of the network is the same for all valid graph-based flow unchanged.
solutions. viii. Repeat Steps (ii) through (vii) until the total power
Let the network graph G(V , E) be split into two clusters C 1 generation satisfies the total power demand.
and C 2 such that C 1 ∪ C 2 = V and C 1 ∩ C 2 = ∅. If PG1 (PG2 )
and PL1 (PL2 ) be the total generation and total demand in C 1 (C 2 ),
then the net generation in C 1 is given by ΔP1 = PG1 − PL1 , of conservation of energy, total power transfer across cut-set
while the net generation in C 2 is given by ΔP2 = PG2 − PL2 . K for each of the flow solutions A and B must be equal to
Now, cut-set K between clusters C 1 and C 2 would include ΔP1 = −ΔP2 , i.e.,
only those edges whose one end belongs to C 1 and the other
k
 k

end belongs to C 2 ; let the number of edges in cut-set K be k.
Also, let f1A , f2A , …, fkA denote network flows through different flA = flB = ΔP1 = −ΔP2 , ∀el ∈ K (3)
edges of cut-set K for a valid graph-based flow solution A, l=1 l=1

and f1B , f2B , …, fkB denote the network flows through the same The validity of (3) is illustrated through three different base-
edges for a valid graph-based flow solution B. Then, by the law case network flow solutions shown in Fig. 2. Cases 1 and 2

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 927

TABLE I
POWER TRANSFER ACROSS CUT-SET K FOR THREE DIFFERENT FLOW
SOLUTIONS OF FIG. 2

denote two valid flow graphs obtained using graph theory while
Case 3 depicts a flow graph obtained using a DC power flow
solution. Table I shows that for the cut-set K ={4-1, 9-2, 9-3},
even though the flows through the individual edges of the cut-set
are different, the net power transfer across cut-set K is equal to
380.86 MW for all three cases. Note that the flow limit of each
edge in Fig. 2 is 300 MW.

IV. GRAPH THEORY BASED NETWORK ANALYSIS TOOL Fig. 3. (a) Power transfer across cut-set K crit for flow graph of Fig. 2(a), (b)
Power transfer across cut-set K crit for flow graph of Fig. 2(b), and (c) Power
As described in Section II, a transmission line or transformer transfer across cut-set K crit for flow graph of Fig. 2(c).
will be considered a special asset if the power flowing through
it cannot be rerouted via the set of its indirect paths. For each
such special asset, the graph theory-based network analysis tool Algorithm 2: Graph Theory-Based Feasibility Test (FT).
finds the limiting critical cut-set as described below.
i. Define C (V , E) = C(V , E). Remove edge el
from C . Initialize a variable T Cl to zero (i.e.,
A. Graph Theory-Based Feasibility Test (FT) T Cl := 0).
The graph theory-based feasibility test (FT) described in Al- ii. Search C to obtain the shortest unsaturated path P
gorithm 2 below examines all the transmission assets to quickly from vlF to vlT using breadth first search (BFS) [37];
identify the set of special assets and the limiting critical cut-set path P is considered unsaturated if it has capacity to
corresponding to each special asset. That is, if fl units of power reroute additional flow.
flows through edge el from vlF to vlT , Algorithm 2 will first iii. Find the maximum extra flow, CP , that can be
identify if el is a special asset. If el does turn out to be a special rerouted through path P from vlF to vlT .
asset, then Algorithm 2 will identify the associated limiting iv. Update T Cl as T Cl := T Cl + CP , and the weights of
critical cut-set, K crit , and the power transfer margin associated C as per (1); note that this step saturates path P in C .
with K crit , denoted by Tl . v. Repeat Steps (ii) through (iv) until there exists no
Although there can be multiple saturated cut-sets associated unsaturated path in C from vlF to vlT .
with a special asset, Algorithm 2 is able to identify the limiting vi. Due to outage of el , compute the transfer margin, Tl ,
critical cut-set because it is the first one to get saturated in Step as: Tl = T Cl − fl . If Tl for el is negative, el is a
(iv). Consider the system shown in Fig. 2 once more. When edge special asset.
4-1 is examined by the FT, with respect to any of the three flow vii. To identify K crit , traverse the saturated graph C from
graphs, following observation is made: edge 4-1 is a special vlF to vlT . All the vertices that can be reached from vlF
asset as it fails FT, and is associated with a limiting critical without traversing a saturated edge are grouped into
cut-set containing edges 4-1 and 6-7 (i.e., K crit = {4-1,6-7}) cluster C 1 . Similarly, the vertices that cannot be
and Tl = −35.86 MW. The implication of the above statement reached from vlF without traversing a saturated edge
is explained with the help of Fig. 3(a), Fig. 3(b), and Fig. 3(c) are grouped into cluster C 2 . Cut-set K crit contains
which depict the power transfer across cut-set K crit for the three the edges whose one end is in C 1 and the other end is
different flow graphs of Fig. 2. in C 2 .
From Fig. 3 it is clear that although the individual flows on
different edges of the cut-set are different, FT finds that, for all
three flow graphs, if the edge 4-1 is lost, the cut-set K crit will and (c) computes the power transfer margin across the identified
have a power transfer capability shortage of 35.86 MW from limiting critical cut-set.
cluster C 1 to cluster C 2 . For example, in Fig. 3(a), when edge
4-1 is lost, the flow in edge 6-7 becomes (208+127.86) MW B. Graph theory-Based Network Flow Update Scheme (UPS)
= 335.86 MW, which exceeds its rating (of 300 MW) by 35.86 During major power system disturbances, multiple outages
MW. In summary, the FT: (a) detects special assets, (b) identifies can occur in rapid succession. Therefore, the FT results would
the limiting critical cut-set associated with each special asset, also change following the outage of an edge. To identify the

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
928 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

Fig. 4. (a) A flow graph obtained from graph-based network flow algorithm,
(b) Update scheme (UPS) of network flow solution for the outage of edge 5-6.

Fig. 5. (a) Rerouting the flow on edge el does not involve any edge of the
Algorithm 3: Graph Theory-Based Update Scheme (UPS). indirect paths of edge em , and (b) Rerouting the flow on edge el involves some
edges of the indirect paths of edge em .
i. Let, the flow to be rerouted be given by F = fl , where
fl refers to the flow through edge el from vertex vlF to
vlT .
ii. Remove edge el from F(V , E) and C(V , E). Let it be known from the base-case FT that flow through edge
iii. Search C to obtain the shortest unsaturated path P em can be rerouted through path P 1 , while the loss of edge el
from vlF to vlT using breadth first search (BFS) [37]. alters flow through path P 2 . Then, in Fig. 5(a), when el goes
iv. Find the maximum extra power, CP , that can be out, the flow through el is rerouted through P 2 by UPS. Now,
rerouted through path P . as P 1 and P 2 do not involve common edges, the rerouting of
v. If F > CP , inject CP units of flow through path P and power through P 2 by UPS does not modify the flows through
update F as F := F − CP . If F ≤ CP , inject F units P 1 ; therefore, FT need not be repeated for em . However, if P 1
of flow through path P and set F := 0. Update the and P 2 have common edges, as seen in Fig. 5(b); i.e., rerouting
weights of F and C accordingly. of the flow of el affects the flow through P 1 , then em must be
vi. Repeat Steps (ii) through (v) until F = 0. examined by FT once again after the outage of el . This rationale
of screening the assets to be examined by FT in the event of
an outage is called the shortlisting asset (SA) scheme. It will be
set of special assets following an outage, it is important to first shown in Section V that the usage of the SA scheme significantly
update the graph theory-based network flows to account for the reduces the computation time.
outage of any edge. The advantage of graph theory-based flows
is that rerouting of the flow upon the loss of an edge can be D. Graph-Traversal Scheme
achieved extremely fast. The technique of updating the flow
graph F(V , E) and latent capacity graph C(V , E) when edge The proposed FT and UPS algorithms use BFS [37] scheme
el suffers an outage is done in accordance with Algorithm 3, to traverse the graph. BFS, in comparison to depth first search
which describes the graph theory-based update scheme (UPS). (DFS), has the advantage that it starts at a source vertex and
The UPS for the outage of an edge is explained with the help of explores all the neighboring vertices at present depth before
Fig. 4. A flow graph obtained from graph theory-based network moving on to the vertices at the next depth. Once the sink vertex
flow algorithm is depicted in Fig. 4(a). The update of the network is reached the algorithm stops. When BFS is used to traverse
flows when the edge 5-6 goes out is shown in Fig. 4(b). The UPS the graph to reach a sink from a given source, the path traced by
simply reroutes 25 MW of flow through path 5-4-1-6 to create BFS is already the shortest path (if there was a shorter path, BFS
an updated network flow solution. would have found it earlier). Both the graph theory-based FT and
UPS scan through the set of indirect paths associated with any
edge. However, there could be many indirect paths associated
C. Shortlisting Assets (SA) Scheme for Feasibility Test (FT)
with an edge. The unique search properties of FT and UPS that
In the base-case scenario when the flow graph is built for the facilitates real-time identification of limiting critical cut-sets and
first time all transmission assets would be investigated by the rerouting of the network flow, respectively, are discussed below:
FT. However, in the event of the outage of an edge, when UPS 1) Graph Traversal During Feasibility Test (FT): In each
gives an updated flow graph, it is not necessary to test all the iteration of the FT, saturation of an indirect path occurs as
assets by the FT once again to identify the special assets. By described in Step (iv) of Algorithm 2. This will occur within
intelligently exploiting the information provided by FT in the a small number of iterations for a power network because many
base-case scenario and using the UPS to reroute the flow for the of its edges are common to multiple indirect paths. Therefore,
edge that is out, the FT can be performed on only a subset of the limiting critical cut-set, K crit , can be identified without
the assets to evaluate the impact of a second contingency. This requiring the BFS scheme to scan through the set of all indirect
is explained through Fig. 5. paths.

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 929

Fig. 6. Real time identification of limiting critical cut-sets on the IEEE 118-bus test system.

2) Graph Theory-Based Update Scheme (UPS): Based on TABLE II

IDENTIFICATION OF LIMITING CRITICAL CUT-SETS
the same rationale explained above, for rerouting the flow
through any edge of the power network, Algorithm 3 will only
utilize a small subset of indirect paths.
Apart from BFS, other commonly used graph search methods
for finding the shortest path between a source-sink pair are
Bellman-Ford algorithm [38], and Dijkstra algorithm [39]. If |E|
denotes the total number of edges, and |V| denotes the total num-
ber vertices, the time-complexity of the Bellman-Ford algorithm
is O(|E||V|) [40]. The time-complexity of Dijkstra algorithm
implemented using binary heap is O(|E| + |V |log|V |) [41].
Lastly, the time-complexity of the BFS algorithm is O(|E| +
|V | [42], which is the best among the three shortest-path graph
traversal techniques. Therefore, we have used the BFS graph
traversal scheme to design the proposed algorithms (FT and
UPS) to determine if contingencies create saturated cut-sets.

3) 2nd Outage: When 19-34 is lost, no additional special

V. RESULTS assets are identified.
4) 3rd Outage: When 37-38 is lost, the asset 42-49 fails the FT
A. IEEE 118-Bus System and is classified as a special asset. The loss of 42-49 would
The utility of the proposed algorithm for enhanced situational saturate the limiting critical cut-set K 3crit by a margin of
awareness is explained with a case-study on IEEE-118 bus −186 MW, i.e., Tl3 = −186 MW.
system. Due to a hurricane, let the following transmission asset 5) 4th Outage: When 49-66 is lost, no additional special
outages occur one after another: 15-33, 19-34, 37-38, 49-66, assets are identified.
and 47-69 (marked O1 through O5 in Fig. 6). From Fig. 6 and 6) 5th Outage: When 47-69 is lost, the assets 59-56, 63-59,
Table II, following information is obtained when the algorithm 63-64, and 64-65 are classified as special assets. The loss
is applied as outages manifest: of these four assets would saturate the limiting critical
1) Base-case: In the base-case scenario, the asset 26-30 fails cut-sets, K 5a 5b 5c 5d
crit , K crit , K crit , and K crit , by margins of
the graph theory-based FT and is classified as a special −64, −191, −191, and −219 MW, respectively (i.e., Tl5a
asset. The loss of 26-30 would saturate the limiting critical = −64 MW, Tl5b = −191 MW, Tl5c = −191 MW, Tl5d =
cut-set K 0crit by a margin of −77 MW, i.e., Tl0 = −77 −219 MW).
MW. The value of the information obtained above can be realized
2) 1st Outage: When 15-33 is lost, no additional special assets by considering the following scenario: after the occurrence of
are identified. the fifth outage, the proposed algorithm would inform the power

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
930 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

system operator that if any of the four assets identified in the last
row, second column of Table II is lost next (as the 6th outage),
the corresponding cut-set identified in the third column would
be saturated by the margin mentioned in the fourth column. If
this anticipated overload is to be avoided, the operator must
preemptively reduce the power flowing through the identified
cut-set by at least the amount mentioned in the last column.
Thus, the proposed network analysis tool is an enhanced power
system connectivity monitoring scheme that improves the power
system operators’ situational awareness by augmenting their
visualization in real-time.

B. 17,941 Bus Model of Western Interconnection (WI).

This section presents the results obtained using a 17,941-bus
model of WI. Sub-section 1) presents some statistics of graph
theory-based FT and UPS which highlight the computational
advantage of the proposed technique. Sub-section 2) describes
how the proposed network analysis scheme provides useful
information when a sequence of outages occurs in this system. Fig. 7. (a) Histogram of number of indirect paths traversed by the graph theory-
1) Computational Efficiency of the Graph Theory-Based FT: based FT, and (b) Histogram of maximum number of edges contained in an
It takes 6 min to run N-1 FT for this system. Conversely, the indirect path.
time required to run a single DC power flow is 0.12 s. As there
are 22,091 transmission assets in the system, time required to
run a DC power flow for the outage of every asset would require was done for the outage of the events, it took approximately
solving 22,091 DC power flows, which would approximately 0.5 s to build the flow graph and the latent capacity graph
require 0.12 × 22,091 s ∼ = 44 min. Therefore, for this system, for the base-case. As mentioned earlier, it takes approximately
performing N-1 FT is approximately 7 times faster compared to 6 min to run FT on all transmission assets in the base-case.
N-1 DC CA. For a valid comparison, the simulations were done Whether events 1 and 2 resulted in any additional special asset
on the same computer (Core i7, 3.60 GHz CPU processor with was investigated as follows:
16 GB RAM). Event 1: Once the 500 kV H-NG transmission line was lost,
When any edge el is examined by the graph theory-based graph theory-based UPS took only 0.20 s to reroute the flow
FT, the indirect paths of el are traversed by BFS. However, the to obtain a new flow graph. The SA scheme took 0.06 s to
saturation of the set of indirect paths may occur after a small identify 271 edges that were to be examined by FT for this
number of indirect paths are traversed by the graph theory-based new graph. Time required by FT to examine all the 271 edges
FT. Moreover, since BFS always identifies the shortest path from for an outage was 32 s. Among the 271 edges, 4 edges failed
the source to the sink, the number of edges contained in an FT and wereclassified as special assets as shown in Table III.
indirect path would be relatively small. For every non-radial edge For the four special assets, the FT found the corresponding
of this system, the number of indirect paths required to saturate limiting critical cut-set, K crit ; |K crit | in Table III denotes the
the graph and the maximum number of edges contained in an number of edges contained in K crit . Moreover, FT provided
indirect path is computed. The statistics of FT is summarized in information regarding the impact of the loss of a special asset
Fig. 7(a) and Fig. 7(b). on the associated limiting critical cut-set. For example, if the
Fig. 7(a) plots the histogram for the number of indirect paths transmission corridor 936-1192 is lost next, the limiting critical
used by BFS to saturate the latent capacity graph. The largest cut-set would be saturated by a margin of 441 MW. The total
number of indirect paths required was 58. Fig. 7(b) plots the time required to perform this network analysis and identify all
histogram of maximum number of edges contained in an indirect the limiting critical cut-sets after the outage of H-NG was 32.26
path traced by BFS; the maximum was 111. Thus, the histogram s (i.e., total time taken by UPS, SA, and FT). On the other hand,
plots demonstrate that the graph theory-based FT essentially if FT were to be run on all transmission assets (as was done
uses a small subgraph to detect the saturation of a cut-set; this in the base-case), the time required would be 6 min. Therefore,
is the fundamental reason why the graph theory-based FT is intelligently performing FT on a shortlisted set of edges reduced
computationally so efficient. the computation time from 6 min to 32.26 s.
2) A Case Study on Western Interconnection (WI): This Event 2: When 230/92 kV Coachella Valley transformers
sub-section demonstrates the usefulness and scalability of the are tripped, the UPS took only 0.06 s to obtain the updated
proposed approach through a N-1-1 event analysis of this system. network flow solution. Time required by the SA scheme to
The loss of 500 kV Hassayampa-North Gila (H-NG) transmis- shortlist the edges to be examined by FT was 0.07 s; 82 new
sion line was the first event, while the second event was the loss of edges were shortlisted. Time required by FT to examine all the
230/92 kV Coachella Valley transformers. Before the analysis 82 shortlisted edges was 10 s. Among the 82 edges examined,

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 931

TABLE III
APPLICATION OF THE GRAPH THEORY-BASED NETWORK ANALYSIS IN WESTERN INTERCONNECTION (WI)

10 edges failed FT and were classified as special assets (see

Table III). Total time required to identify the set of special assets
after the outage of Coachella Valley transformers was 10.13 s.
Conversely, performing a DC CA for all transmission assets
would have required 44 min, or about 260 times longer than the
proposed graph theory-based approach.

VI. DISCUSSIONS
A. Practical Utility of the Proposed Algorithm
Fig. 8. Topology of a sample six-bus power system (branch impedances are
After the 2011 U.S. Southwest blackout, the FERC [3] re- represented in terms of a variable z).
ported the following finding: “Affected TOPs (transmission
operators) have limited visibility outside their systems, typi-
cally monitoring only one external bus. As a result, they lack
adequate situational awareness of external contingencies that
could impact their systems. They also may not fully understand
how internal contingencies could affect SOLs (system operating
limits) in their neighbors’ systems.” The recommendation of
FERC to TOPs was to “review their real-time monitoring tools,
such as state estimator and RTCA, to ensure that such tools
represent critical facilities needed for the reliable operation of
the BPS (bulk power system)”.
Now, modeling all “critical facilities” over a large area (across
different utilities) could significantly increase the number of
Fig. 9. Scenario 1. (a) A DC power flow solution in base-case, and (b) A DC
contingencies to be evaluated by RTCA, which would then power flow solution for the outage of edge 1-2.
increase the solution time considerably [4], [11]. In this regard,
the ability of the proposed algorithm to analyze the effects
of any outage on very large systems and provide meaningful
using the test system shown in Fig. 8, and the corresponding
quantifiable information in a matter of seconds gives it a distinct
flows shown in Fig. 9 and Fig. 10.
advantage. Moreover, the special assets detected by the FT can
Fig. 9(a) presents a DC power flow solution, when 100 MW
be suitable candidates for detailed analysis by a more precise
of power is injected at bus 1, and 100 MW is withdrawn at bus
CA tool. Thus, the proposed research can complement real-
2 (Scenario 1). The numbers in non-bold fonts indicate flows,
time operations by extending an operator’s visibility to external
while the numbers in bold font denote line ratings. The proposed
contingencies, while alleviating the associated computational
FT algorithm identifies edge 1-2 as a special asset because the
burdens.
indirect paths of edge 1-2 do not have sufficient capacity to
reroute the flow through the direct path, namely, edge 1-2. A
B. Proposed Method is not Guaranteed to Detect all post-contingency DC power flow shown in Fig. 9(b) validates
Contingencies that Result in Post-Contingency that such an outage results in overloads along Indirect path 1.
Branch Overloads
Fig. 10(a) presents a DC power flow solution, when 85 MW of
As per the FT when all the indirect paths do not have suffi- power is injected at bus 1, and 85 MW is withdrawn at bus
cient capacity to reroute the power flowing through an edge, it 2 (Scenario 2). In this scenario, the proposed FT algorithm
implies that it would inevitably result in post-contingency branch does not identify edge 1-2 as a special asset because the set
overloads. However, the converse is not true. This is illustrated of indirect paths have sufficient capacity to reroute the flow of

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
932 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

Fig. 10. Scenario 2. (a) A DC power flow solution in base-case, and (b) A DC
power flow solution for the outage of edge 1-2.

Fig. 12. Topology of a sample five-bus power system (branch impedances are
represented in terms of a variable z).

Fig. 11. K i is the ith cut-set (among x cut-sets) associated with edge el that
separates the network into two disjoint clusters.

the direct path. However, a post-contingency DC power flow

solution shown in Fig. 10(b) indicates that the Indirect path 1
is still overloaded, due to lower impedance of Indirect path 1
compared to Indirect path 2.
From this illustration, the following conclusions can be
drawn: when the set of indirect paths do not have the capacity to
reroute the power flowing through the direct path (see Fig. 9), no
additional information is required to conclude that there would
be a post-contingency overload. The proposed graph theoretic
power flow model takes advantage of this observation to identify Fig. 13. Power transfer across four different cut-sets (K 1 , K 2 ,K 3 , K 4 )
violations quickly. At the same time, the proposed approach is associated with edge 3-4 for Case 1.
not able to capture the overload occurring in Fig. 10. This is
because the graph theory-based network flow algorithm ignores
the effects of impedances when creating the flows. Thus, the with edge el is denoted by
proposed approach may not detect all possible post-contingency  
branch overloads. K i = el , el1 , el2 , . . . , el(k−1) for 1 ≤ i ≤ x (4)

where, k is the total number of edges in cut-set K i , and x is

C. Proposed Method is Guaranteed to Detect all
the total number of cut-sets associated with edge el . When the
Contingencies that Saturate a Cut-Set transfer margin, Tl , computed by the graph theory-based FT
The discussion presented in Section VI-B reveals that the (proposed in Section IV-A) is negative it implies that the outage
graph theory-based FT is not guaranteed to identify all contin- of edge el saturates at least one cut-set, among the x cut-sets
gencies that create post-contingency branch overloads. How- that edge el is associated with. On the other hand, if the transfer
ever, the proposed algorithm does guarantee detection of all margin, Tl , computed by the FT is positive, it implies that the
contingencies that create a saturated cut-set in the network. This outage of edge el does not saturate any of the x cut-sets that it is
is explained as follows. Let us examine if the outage of edge el of associated with. Therefore, the graph theory-based FT will not
Fig. 11 would create a saturated cut-set in the system using the miss a single contingency that would create a saturated cut-set.
proposed FT. Edge el could be associated with multiple cut-sets This is illustrated using the test system shown in Fig. 12, and
in the system. With reference to Fig. 11 the ith cut-set associated the corresponding flows shown in Fig. 13 and Fig. 14.

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 933

TABLE V
RANKING OF CONTINGENCIES AND CASCADING ANALYSIS ON
IEEE 118-BUS TEST SYSTEM

solution (see Table IV). It is observed that PK is less than RK

for K 1 , K 2 , K 3 , K 4 . This proves that for Case 2, outage of
edge 3-4 does not saturate any cut-set that is associated with it.
Furthermore, note that in Fig. 13, the power flowing through
different edges of the limiting critical cut-set, K 2 = {3-4,3-
Fig. 14. Power transfer across four different cut-sets (K 1 , K 2 ,K 3 ,K 4 ) 5,1-5}, are not in the same direction. This implies that cut-set
associated with edge 3-4 for Case 2. K 2 is not a coherent cut-set (in a coherent cut-set power flows
in the same direction in all the edges of the cut-set [27]).
TABLE IV Therefore, such types of critical interconnections cannot be
POWER TRANSFER CAPACITY ACROSS DIFFERENT CUT-SETS ASSOCIATED detected by the algorithm presented in [27]. It is also important
WITH EDGE 3-4
to highlight here that enumerating the power transfer capability
across different cut-sets by a DC power flow solution requires
previously defining all the cut-sets. On the other hand, the graph
theory-based FT can investigate the power transfer capability of
different cut-sets without the cut-sets being pre-defined. This is
a unique advantage of the proposed network analysis, because
listing all possible cut-sets for a large power network containing
thousands of buses especially during extreme event scenarios is
not practically feasible.
Fig. 13 presents a DC power flow solution when the total
load and generation in the system is 594 MW (Case 1). The
FT algorithm finds that the outage of 3-4 saturates cut-set
D. Comparison of Proposed Method With Other Power System
K 2 ={3-4,3-5,1-5} by 31 MW. To validate this inference, the
power transfer capability across each of the cut-sets associated Vulnerability Assessment Techniques
with edge 3-4 is enumerated from the DC power flow solution. In this sub-section, the output of the proposed algorithm is
As shown in Fig. 13, edge 3-4 is associated with four cut-sets: compared with those obtained from two power system vulner-
K 1 ,K 2 , K 3 ,K 4 . The power transfer capabilities across the four ability assessment techniques, namely, the metrics developed
cut-sets of the test system when edge 3-4 is lost are summarized in [28] and [33], both of which can handle exhaustive N-1
in Table IV, where PK , denotes the total flow that is to be evaluation. In [28], an extended betweenness index (derived from
transferred across the cut-set, and RK denotes the total capacity PTDFs and transmission line limits) was used to rank different
of all the edges belonging to the cut-set (excluding edge 3-4 contingencies. In [33], a DC power flow based linear sensitivity
itself). It is observed that PK is greater than RK only for cut-set factor, called line outage impact factor, derived from LODFs
K 2 by 31 MW. This verifies that for Case 1, the outage of edge was used to screen out critical contingencies. The analysis was
3-4 would saturate cut-set K 2 by 31 MW. performed on the IEEE 118-bus system for the same sequence
Fig. 14 presents a DC power flow solution when the total load of outages that were described in Table II. In order to validate
and total generation of the system is 486 MW (Case 2). In this the severity of the different contingencies, a cascading failure
case, the FT algorithm detects that the indirect paths of edge simulation was run in MATCASC [43], a software package
3-4 have positive transfer margins indicating that they have the linked with MATPOWER that facilitates simulation of cascad-
capacity to carry additional power, if need be. To validate this ing failures for any initiating contingency. The amount of load
observation, the power transfer capability across each cut-set shed at the end of the cascade indicated the severity of the
associated with edge 3-4 is enumerated from the DC power flow contingency. The results of the comparison are shown in Table V.

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
934 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 36, NO. 2, MARCH 2021

Column 2 of Table V shows the contingencies identified by [2] FERC Staff, “The Con Edison power failure of July 13 and 14, 1977,” U.S.
MATCASC that result in load shed as the different events mani- Dept. of Energy Federal Energy Regulatory Commission, pp. 2–3, 19-20
Jun. 1978.
fest in the IEEE 118-bus system. The ranking of these load-shed- [3] FERC and NERC staff, “Arizona-Southern California outages on Septem-
causing-contingencies, obtained by the techniques developed in ber 8, 2011, causes and recommendations,” Federal Energy Regula-
[28] and [33] are provided in Columns 4 and 5, respectively. It tory Commission and North American Electric Reliability Corporation,
Apr. 2012. [Online] Available: [Link]
can be observed from Table V that the contingencies that actually [Link]
result in loss of load were not the top ranked contingencies [4] T. Werho, V. Vittal, S. Kolluri, and S. M. Wong, “Power system connectiv-
identified by the metrics developed in [28] and [33]. For instance, ity monitoring using a graph theory network flow algorithm,” IEEE Trans.
Power Syst., vol. 31, no. 6, pp. 4945–4952, Nov. 2016.
after the fifth outage, if any of the four new contingencies [5] M. Panteli and D. S. Kirschen, “Situational awareness in power systems:
identified in Column 2 were to occur (as the sixth outage), then it Theory, challenges, applications,” Electric Power Syst. Rese., vol. 122,
would result in load shedding in excess of 25%. However, none pp. 140–151, 2015.
[6] A. Abedi, L. Gaudard, and F. Romerio, “Review of major approaches to
of these four high load-shed-causing-contingencies appeared in analyze vulnerability in power systems,” Rel. Eng. Syst. Saf., vol. 183,
the top four ranked contingencies of [28] or [33]. On the other pp. 153–172, Nov. 2018.
hand, all the load-shed-causing-contingencies were detected as [7] X. Li, P. Balasubramanian, M. Sahraei-Ardakani, M. Abdi-Khorsand,
K. W. Hedman, and R. Podmore, “Real-time contingency analysis with
special assets by the proposed algorithm (compare Column 2 of corrective transmission switching,” IEEE Trans. Power Syst., vol. 32, no. 4,
Table V with Column 2 of Table II). This shows the usefulness pp. 2604–2617, Jul. 2017.
of the proposed algorithm in detecting critical contingencies. [8] Z. Li, J. Wang, H. Sun, and Q. Guo, “Transmission contingency anal-
ysis based on integrated transmission and distribution power flow in
smart grid,” IEEE Trans. Power Syst., vol. 30, no. 6, pp. 3356–3367,
VII. CONCLUSION Nov. 2015.
[9] J. Baranowski and D. J. French, “Operational use of contingency analysis
This paper presents a new graph theoretic approach for real- at PJM,” in Proc. IEEE Power Energy Soc. Gen. Meeting, San Diego, CA,
time vulnerability assessment resulting in enhanced situational USA, pp. 1–4, Jul. 2012.
[10] F. Garcia, N. D. R. Sarma, V. Kanduri, and G. Nissankala, “ERCOT control
awareness for power system operations. The major contributions center experience in using real-time contingency analysis in the new nodal
of this research are as follows: market,” in Proc. IEEE Power Energy Soc. Gen. Meeting, San Diego, CA,
1) The most important research finding is that a relaxed USA, pp. 1–8, Jul. 2012.
[11] S. Huang and V. Dinavahi, “Real-time contingency analysis on massively
graph theory-based network analysis tool can evaluate if parallel architectures with compensation method,” IEEE Access, vol. 6,
a contingency will create saturated cut-sets in a meshed pp. 44519–44530, 2018.
power system. The proposed algorithm finds the cut-set [12] I. Dobson, B. A. Carreras, D. E. Newman, and J. M. Reynolds-Barredo,
“Obtaining statistics of cascading line outages spreading in an electric
which becomes saturated by the largest transfer margin as transmission network from standard utility data,” IEEE Trans. Power Syst.,
an impact of the outage. The transfer margin indicates the vol. 31, no. 6, pp. 4831–4841, Nov. 2016.
minimum amount by which the power transfer through the [13] I. Dobson, “Estimating the propagation and extent of cascading line
outages from utility data with a branching process,” IEEE Trans. Power
cut-set must be reduced to alleviate the saturation of the Syst., vol. 27, no. 4, pp. 2146–2155, Nov. 2012.
cut-set and sustain the impact of the outage. [14] P. Rezaei, P. D. H. Hines, and M. J. Eppstein, “Estimating cascading failure
2) The proposed graph theoretic algorithms (feasibility test, risk with random chemistry,” IEEE Trans. Power Syst., vol. 30, no. 5,
pp. 2726–2735, Sep. 2015.
update scheme, and shortlisting asset) reduce the compu- [15] M. Rahnamay-Naeini, Z. Wang, N. Ghani, A. Mammoli, and M. M. Hayat,
tational time that is required for providing real-time sit- “Stochastic analysis of cascading-failure dynamics in power grids,” IEEE
uational awareness. Although the proposed analysis may Trans. Power Syst., vol. 29, no. 4, pp. 1767–1779, Jul. 2014.
[16] P. D. H. Hines, I. Dobson, and P. Rezaei, “Cascading power outages
not detect all types of branch overloads, it is guaranteed to propagate locally in an influence graph that is not the actual grid topology,”
detect all overloaded cut-sets. Fast detection of overloaded IEEE Trans. Power Syst., vol. 32, no. 2, pp. 958–967, Mar. 2017.
cut-sets is important because they represent the “seams or [17] T. Ishizaki, A. Chakrabortty, and J. Imura, “Graph-theoretic analysis of
power systems,” Proc. IEEE, vol. 106, no. 5, pp. 931–952, May 2018.
fault-lines across which islanding seems likely” [27]. [18] R. Albert, I. Albert, and G. L. Nakarado, “Structural vulnerability of the
Using the proposed approach, system operators will also have North American power grid,” Phys. Rev. E, vol. 69, pp. 1–10, Feb. 2004.
better preparedness to address the identified violations, long [19] H. Bai and S. Miao, “Hybrid flow betweenness approach for identification
of vulnerable line in power system,” IET Gener. Transmiss Distrib., vol. 9,
before the set of all possible violations are detected by a more no. 12, pp. 1324–1331, Jan. 2015.
detailed network analysis tool. Hence, the proposed method can [20] K. Wang et al., “An electrical betweenness approach for vulnerability
be a useful complementary tool to the existing methods of power assessment of power grids considering the capacity of generators and
loads,” Phys. A: Statist. Mech. Appl., vol. 390, no. 23/24, pp. 4692–4701,
system vulnerability assessment for real-time operations. Nov. 2011.
[21] P. Chopade and M. Bikdesh, “New centrality measures for assessing smart
grid vulnerabilities and predicting brownouts and blackouts,” Int. J. Crit.
ACKNOWLEDGMENT Infrastruct Prot, vol. 12, pp. 29–45, 2016.
The authors would like to acknowledge Dr. John Undrill from [22] J. Fang, C. Su, Z. Chen, H. Sun, and P. Lund, “Power system structural
vulnerability assessment based on an improved maximum flow approach,”
Arizona State University, and Dr. Yingchen Zhang from National IEEE Trans. Smart Grid, vol. 9, no. 2, pp. 777–785, Mar. 2018.
Renewable Energy Laboratory for their valuable discussions and [23] G. J. Correa-Henao and J. M. Yusta-Loyo, “Representation of electric
help during the course of this research. power systems by complex networks with applications to risk vulnerability
assessment,” DYNA, vol. 82, no. 192, pp. 68–77, Aug. 2015.
[24] J. Beyza, E. Garcia-Paricio, and J. M. Yusta, “Applying complex network
REFERENCES theory to the vulnerability assessment of interdependent energy infrastruc-
tures,” Energies, vol. 12, no. 3, pp. 421–436, Jan. 2019.
[1] B. A. Carreras, D. E. Newman, and I. Dobson, “North American blackout [25] Y. Zhu, J. Yan, Y. Sun, and H. He, “Revealing cascading failure vulnera-
time series statistics and implications for blackout risk,” IEEE Trans. bility in power grids using risk-graph,” IEEE Trans. Parallel Distrib. Syst.,
Power Syst., vol. 31, no. 6, pp. 4406–4414, Nov. 2016. vol. 25, no. 12, pp. 3274–3284, Dec. 2014.
Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
BISWAS et al.: GRAPH THEORETIC APPROACH TO POWER SYSTEM VULNERABILITY IDENTIFICATION 935

[26] J. Beyza, J. Yusta, M. G. Correa, and H. Ruiz, “Vulnerability assessment Anamitra Pal (Senior Member, IEEE) received the
of a large electrical grid by new graph theory approach,” IEEE Latin Amer. B.E. degree (summa) in electrical and electronics
Trans., vol. 16, no. 2, pp. 527–535, Feb. 2018. engineering from the Birla Institute of Technology,
[27] A. Beiranvand and P. Cuffe, “A topological sorting approach to identify Mesra, Ranchi (India) in 2008, the M.S. and Ph.D.
coherent cut-sets within power grids,” IEEE Trans. Power Syst., vol. 35, degrees in electrical engineering from Virginia Tech,
no. 1, pp. 721–730, Jan. 2020. Blacksburg in 2012 and 2014, respectively. He is now
[28] E. Bompard, E. Pons, and D. Wu, “Extended topological metrics for an Assistant Professor with the School of Electrical,
the analysis of power grid vulnerability,” IEEE Syst. J., vol. 6, no. 3, Computer, and Energy Engineering at Arizona State
pp. 481–487, Sep. 2012. University. Previously, from 2014-2016, he was an
[29] E. Bompard, R. Napoli, and F. Xue, “Extended topological approach for Applied Electrical and Computer Scientist in the Net-
the assessment of structural vulnerability in transmission networks,” IET work Dynamics and Simulation Science Laboratory
Gener. Transmiss Distrib., vol. 4, no. 6, pp. 716–724, Jun. 2010. at the Biocomplexity Institute of Virginia Tech. He is the recipient of the 2018
[30] C. M. Davis, and T. J. Overbye, “Multiple element contingency screening,” Young CRITIS Award for his contributions to critical infrastructure resiliency,
IEEE Trans. Power Syst., vol. 26, no. 3, pp. 1294–1301, Aug. 2011. as well as the 2019 Outstanding IEEE Young Professional Award from the IEEE
[31] T. Werho, V. Vittal, S. Kolluri, and S. M. Wong, “A potential island Phoenix Section. His current research interests include power system modeling,
formation identification scheme supported by PMU measurements,” IEEE transient and dynamic stability analysis, critical infrastructure resiliency assess-
Trans. Power Syst., vol. 31, no. 1, pp. 423–431, Jan. 2016. ment, and wide area measurements-based protection, monitoring, and control.
[32] P. Kaplunovich and K. Turitsyn, “Fast and reliable screening of N-2
contingencies,” IEEE Trans. Power Syst., vol. 31, no. 6, pp. 4243–4252,
Nov. 2016.
[33] A. K. Srivastava, T. A. Ernster, R. Liu, and V. G. Krishnan, “Graph-
theoretic algorithms for cyber-physical vulnerability analysis of power
grid with incomplete information,” J. Modern Power Syst. Clean Energy,
vol. 6, no. 5, pp. 887–899, Sep. 2018.
[34] W. Velásquez, M. S. Alvarez-Alvarado, and J. Salvachúa, “Body mass Trevor Werho (Member, IEEE) received the B.S.E.,
index in human walking on different types of soil using graph theory,” M.S., and Ph.D. degrees in electrical engineering
IEEE Access, vol. 6, pp. 47935–47942, 2018. from Arizona State University, Tempe, AZ, USA, in
[35] B. Chaudhuri, B. Demir, L. Bruzzone, and S. Chaudhuri, “Region-based 2011, 2013, and 2015, respectively. He is currently
retrieval of remote sensing images using an unsupervised graph-theoretic working as a Postdoctoral Scholar at Arizona State
approach,” IEEE Geosci. Remote Sensi. Lett., vol. 13, no. 7, pp. 987–991, University researching wind and solar forecasting.
Jul. 2016.
[36] M. S. Tootooni, P. K. Rao, C. Chou, and Z. J. Kong, “A spectral graph
theoretic approach for monitoring multivariate time series data from
complex dynamical processes,” IEEE Trans. Autom. Sci. Eng., vol. 15,
no. 1, pp. 127–144, Jan. 2018.
[37] D. Angel, “A breadth first search approach for minimum vertex cover of
grid graphs,” in Proc. IEEE 9th Int. Conf. Intell. Syst. Control (ISCO),
Coimbatore, India, pp. 1–4, 9-10 Jan. 2015. Vijay Vittal (Fellow, IEEE) received the B.E. degree
[38] A. Elmasry and A. Shokry, “A new algorithm for the shortest path in electrical engineering from the B.M.S. College of
problem,” Networks, vol. 74, pp. 16–39, Dec. 2018. Engineering, Bangalore, India, in 1977, the M. Tech.
[39] N. A. Ojekudo and N. P. Akpan, “An application of Dijkstra algorithm to degree from the Indian Institute of Technology, Kan-
shortest route problem,” IOSR J. Math., vol. 13, no. 3, pp. 20–32, Jun. 2017. pur, India, in 1979, and the Ph.D. degree from Iowa
[40] F. Ahmed, F. Anzum, M. N. Islam, W. Mohammad Abdullah, S. A. Ahsan, State University, Ames, in 1982. He is a Regents’
and M. Rana, “A new algorithm to compute single source shortest path Professor and the Ira A. Fulton Chair Professor in
in a real edge weighted graph to optimize time complexity,” in Proc. the Department of Electrical, Computer, and Energy
IEEE/ACIS 17th Intl. Conf. Comput. Inf. Sci. (ICIS), Singap., pp. 185–191, Engineering at Arizona State University, Tempe, AZ.
2018. He currently is the Director of the Power System
[41] M. Barbehenn, “A note on the complexity of Dijkstra’s algorithm for Engineering Research Center (PSERC). Dr. Vittal is
graphs with weighted vertices,” IEEE Trans. Comput., vol. 47, no. 2, a member of the National Academy of Engineering. He is a recipient of the IEEE
pp. 263-, Feb. 1998. PES Outstanding Power Engineering Educator award in 2000, the IEEE Herman
[42] L. Luo, M. Wong, and W. Hwu, “An effective GPU implementation of Halperin T&D Field Award in 2013, and the IEEE PES Prabha S. Kundur Power
breadth-first search,” in Proc. Des. Automat. Conf., Anaheim, CA, USA, System Dynamics and Control Award in 2018.
2010, pp. 52–55.
[43] Y. Koç, T. Verma, N. A. M. Araujo, and M. Warnier, “MATCASC: A
tool to analyse cascading line outages in power grids,” in Proc. IEEE Intl.
Workshop Intell. Energy Syst. (IWIES), Vienna, 2013, pp. 143–148.

Reetam Sen Biswas (Student Member, IEEE) re-

ceived the [Link]. degree in electrical engineer-
ing from the West Bengal University of Technol-
ogy, Kolkata, India, in 2016 and the M.S. degree in
electrical engineering from Arizona State University,
Tempe, AZ, USA in 2019, where he is currently work-
ing towards the Ph.D. degree. His current research
interests include power system contingency analy-
sis, cascading failure analysis, state estimation, and
power system planning studies with high penetration
of renewable energy.

Authorized licensed use limited to: Universidad Nacional de Ingeniería. Downloaded on April 20,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.