Sie sind auf Seite 1von 8

Page 1 of 8 2018-PSEC-0142

Graph Theory-based Ship Power System Expansion


Strategy for Enhanced Resilience
Kexing Lai Mahesh S. Illindala
The Ohio State University The Ohio State University
Columbus, OH 43210 Columbus, OH 43210
lai.323@osu.edu millindala@ieee.org

Abstract -- The next generation ships are geared toward power deficiency, which is the 95% of their daily electricity
electrification more than ever before. In an all-electric ship consumption [7]. To mitigate the negative effects caused by
(AES), the design and planning of ship power system (SPS) is attacks, this paper proposes a graph theory-based method for
critical for survivability against any physical attacks on power power line expansion strategy on SPS. The proposed strategy
lines. Therefore, enhancement of resilience against malicious aims at reducing load shedding under physical attacks on the
physical attacks is a pressing issue for SPS. This paper proposes
a graph theory-based power line expansion strategy. The
power lines.
proposed method simultaneously maximizes the algebraic The overall objective is to enhance the resilience of SPS by
connectivity and weighted max-flow from generator nodes to reducing the load shedding due to physical attacks on power
load nodes of the graph representative of the SPS. Further, in lines. Prior publications that investigated resilience
terms of load shedding, the proposed strategy is compared enhancement of a SPS can be classified into two categories.
against the planning strategy aimed at reducing the operational Those in the first category are focused on SPS design,
cost. Several tests are conducted to demonstrate the performance planning and operation, aimed at providing stronger support to
of proposed strategy on a ship DC power system. service restoration from contingencies. For instance, an
Keywords—graph theory, pareto optimality, power system energy storage system (ESS) design and planning strategy was
planning, resilience, shipbuilding industry. proposed in [8] to hedge against islanding the electrical zones
under fault conditions. In [9], different operation strategies for
I. INTRODUCTION
SPS considering load curtailments with failed generators were
Since the last few years, the naval ship architecture has covered. Reference [10] discussed the effects of the SPS
been transitioning toward the all-electric ship (AES). In 2013, topology design on its resilience. Alternatively, a graph
the newly launched destroyer, USS Zumwalt DG-1000, theory-based method was developed in [11] for SPS topology
became the US Navy’s first AES. A fully integrated power design to meet availability requirements. In the second
system (IPS) is the most distinctive feature, which includes publication category on resilience enhancement, the intelligent
generating and distributing electrical energy for the entire operation and reconfigurations are being studied for speeding
ship. Multiple missions can be carried out by the IPS such as up the restoration from contingencies. For instance, in [12], a
propulsion, battle, and support service [1]. The other two multiagent system was adopted for system recovery from
Zumwalt class destroyers, USS Zumwalt DG-1001 and DG- faults on SPS by system reconfiguration. Reference [13]
1002, with IPS are scheduled to be commissioned in 2018 and presented a graph theory-based search algorithm for SPS to
2019 [2]. Besides, the Royal Navy commissioned six Type-45 restore from system casualties by saving most of the loads
destroyers with IPS between 2009 and 2013 [3]. Due to the with least cost. A genetic algorithm was used in [14] to
numerous advantages offered by a direct current (DC) determine the actions of the switches for recovery considering
configuration over the AC configuration, such as higher both load level and priority. Unlike aforementioned literatures,
efficiency and better dynamic characteristics, the ship power this paper aims at enhancing the SPS resilience through the
system (SPS) design is moving toward the DC power power line expansion strategy using graph theory.
distribution system [4]. Graph theory is being widely applied nowadays in power
Warships are frequently confronted with malicious attacks, systems for system service restoration and vulnerability
which normally come from torpedoes, mines, or anti-ship analysis. The service restoration in a distribution system
cruise missiles [5]. However, since the SPS plays a critical comprising microgrids can be accelerated using spanning tree
role in the AES, the malicious attackers might target them to search [15]. In [16], the shortest path between sources and
cause significant damage to the ship operation. In the recent faulted point was identified by Dijkstra algorithm. Graph
decade, the utility grid had faced major attacks that affected theory and Choquet Integral were used to quantify the
economically and even led to life losses in some cases. For resilience of distribution system in [17]. Reference [18]
instance, in 2013, the physical attack on a substation in developed a risk-graph for investigating power grid
Coyote, California caused $15M loss [6]. The recent cyber- vulnerability under cascading failures. In [19], a weighted
attack on Ukraine power grid caused up to 73 MWh of electric entropy analysis method was proposed to identify the
vulnerable lines in the power grid with wind power. In this
paper, graph theory is applied for the power system planning
This work was supported by the Office of Naval Research under Award
N00014-16-1-2753

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


2018-PSEC-0142 Page 2 of 8

6 MW Gen1
G n
Ge 2 MW Pr
Propulsion
5 MW Gen3
G
Ge
N16
N3 load 1 N7

3.5 MW 3 MW 4 MW 3.5 MW 3.3 MW 3 MW 2 MW


N1 N5 N9 N11 N13
2 MW 3.7 MW
3.7 MW 3 MW 1.5 MW 2.5 MW
3.5 MW
4 MW N18 0.7 MW N19 N20 1 MW N21 0.7 MW N22 N15
0.3 MW
Control Hotel load 1 5 MW
Radar Hotel load 3 Hotel load 2
center
Weapon
4 MW 3.5 MW load
2.5 MW 3.5 MW
3 MW
N2 N6 N10 N12 N14 2 MW
2.1 MW 1.5 MW 2 MW 3.5 MW 3.5 MW 3 MW 4 MW

N17
4 MW G
Gen 4 N4 P
Propulsion N8 6 MW G
Gen 2
2.33 MW load 2

Fig. 1. Schematic of the ship DC zonal electrical distribution system adapted from [4]

to enhance the resilience of a DC SPS against physical attacks, generators, lines, and load demands are indicated in Fig. 1.
which is a critical issue that has been not been adequately Table I tabulates marginal costs of generators.
addressed in published literature. TABLE I. PARAMETERS OF GENERATORS
The major contributions of this paper are further explained Gen. Marginal cost ($/MWh) Gen. Marginal cost
as follows. A model is developed for SPS resilience ($/MWh)
1 40 3 50
enhancement using graph theory to determine power line 2 40 4 60
expansion. Specifically, the proposed model aims at
maximizing graph algebraic connectivity and weighted max- A. Ship Power System Description
flow from generator nodes to load nodes simultaneously. The Fig. 1 depicts the various components of SPS including
Pareto optimality set of such multi-objective optimization generators, loads, buses, and lines. Their characteristic
problem is obtained. Unlike [11], which assumes that lines features and mathematical models are described as following.
and loads are identical, the proposed method considers line The generators are all diesel generators. Their power
capacities and load priorities. Then, the proposed method and outputs are subject to power ratings as shown in (1), where the
the power line expansion method aimed at cost reduction notations, and , represent the power output and
(MCR) are evaluated for load shedding under physical attacks. capacity of the generator i, respectively.
Unlike [10], [20], which considered only the worst-case
scenario of a physical attack, this paper presents a 0≤ ≤ ,∀ (1)
comprehensive analysis on resilience against physical attacks
and illustrates the results as a weighted load shedding under Power flow on the line l is limited by its capacity as shown
various attack scenarios. Finally, the effects of various factors in (2). Besides, power flow on the line l depends on voltages
on the performance are studied. of the origin bus and the end bus as stated in (3) [10]. The
The reminder of the paper is organized as follows. Section notations, , , and stand for power flow, power
II describes the SPS based on a DC zonal electrical capacity, and conductance of the line l. Further, | ( ) and
distribution system topology. A method for cost reduction
| ( ) denote the voltages at the origin and end buses of the
model for power line expansion of SPS is also presented.
Section III presents the proposed method. Numerical studies line l. The line l in constraints (2) and (3) belongs to the set of
are presented in Section IV to demonstrate the advantages of existing power lines L.
proposed method toward resilience enhancement. Finally,
Section V presents the conclusion. − ≤ ≤ ,∀ ∈ (2)
= ∗ |( ) − | ( ) ,∀ ∈ (3)
II. SHIP POWER SYSTEM AND BASELINE POWER LINE
EXPANSION MODEL
The constraint (4) imposes limits on bus voltages, where
Fig. 1 displays the topology of a SPS, which is a DC zonal the notations of , and denote voltage, minimum
electrical distribution system, adapted from [4]. This is used voltage, and maximum voltage of the bus n.
as the test system in this paper. There are 22 buses and 29
lines. Four generators are installed on buses 3, 4, 16, and 17. (4)
Eight loads are installed on buses 7, 8, 15, 18, 19, 20, 21, and ≤ ≤ ,∀
22, which are prioritized from high to low as: weapons, radars,
control center, propulsion, and hotel loads. The capacities of In addition, for each bus the injected power flow is equal to
the exit power flow. The power balance constraint is shown in

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


Page 3 of 8 2018-PSEC-0142

(5). The set, denotes the set of generators connected to the N3 N7 N16
N16 N1
N1 l2 l7 N5 l8 l9 l16 l20 l22 N
N13
bus n. The indices o(l) and e(l) represent the origin and end
buses of line l. Load demand level on bus n is denoted by . N9 N11
l26
l3 l12 l14 l18 l24
N18
N 18 N19
N N20 N21 N22 N15
∑ +∑| () −∑| () = ,∀ (5)

B. Power Line Expansion Baseline Method for Cost l1 l15 l19 l25
l4 l13
Reduction (MCR) l27
N6 N10 l17 N12
N N14
The power line expansion method aimed at minimizing the l10 l11
l5 l6
operational cost after adding lines (MCR) is used as baseline N2 l21 l23
for comparison purpose [21]. The corresponding model is N4 N8 N17
N
described as follows: Fig. 2. Graph representation of the ship power system shown in Fig. 1

∑ ∗ (6) B. Overviews of Graph Theory Concepts


, , ,
s.t. = ∗ ∗ − ,∀ ∉ (7) Two concepts of graph theory, algebraic connectivity, and
|( ) | ( )
∑∉ (8) max-flow are applied here for the power line expansion in

SPS.
(1) ~ (5)
The algebraic connectivity of the graph G is indicated by
the second-smallest eigenvalue of the Laplacian matrix L;
The objective function, shown in (6), aims at minimizing this is also called as Fiedler eigenvalue in some references, for
the summation of generator fuel costs, which are determined example [23].
from the products of marginal costs and power outputs of Laplacian matrix L of graph G is obtained by (9) where the
generators. The binary variable decides whether the non- matrix A and D represent the adjacent matrix and degree
existing line l should be added or not. When is equal to 1, matrix of graph G, respectively. Adjacent matrix A is a binary
the corresponding line is added, otherwise, it is not built. The matrix, indicating whether there is an edge between two
set L denotes the set of existing lines in the original system. vertices. When the vertices i and j are connected, the
Equations (7) calculate power flow on the line l, which does corresponding elements Ai,j=1, otherwise Ai,j=0, as shown in
not exist in the original system ( ∉ ). If the non-existing (10). Degree matrix D is a diagonal matrix, which shows the
power line l is not added ( = 0), the power flow through it number of edges attached to each vertex. The number of
will be zero, as imposed by (7). Otherwise, the power flow attached edges to the vertex v is called the degree of the vertex
calculation method of the newly added line l ( = 1) is the v. Mathematical definition of degree matrix is given in (11).
same as existing lines in the original system. The constraint (8)
presents the budget for power line expansion. Specifically, the = − (9)
number of added lines is lower than or equal to . The 1; (10)
=
remaining constraints (1)  (5) were earlier explained in ,
0;
Section IIA. It should be noted that products of binary ( ); = (11)
, =
variables and continuous variables in the model make the 0; ℎ
problem difficult to solve. Reference [22] presents a
linearization method that is employed to equate the model to a The magnitude of indicates how well the graph is
mixed-integer linear programming (MILP) problem. This is to connected. The higher value of this number, the better
accelerate the solving process significantly and obtain the connected of the graph. Therefore, is one of the elements in
guaranteed global optimal solution. For brevity, the the resilience matrix defined in [24].
linearization method is not explained here. The max-flow problem is popular in graph theory and
optimization theory. It finds the maximum flow (MFs,t) from
III. GRAPH THEORY-BASED POWER LINE EXPANSION STRATEGY the source vertex s to the sink vertex t. The flow is a mapping
In this section, the proposed power line expansion strategy from the source vertex s to the sink vertex t. It is subjected to
is developed to enhance the system resiliency against physical two constraints [25]:
attacks. (i) The flow of an edge cannot exceed its capacity.
(ii) The sum of flows entering a vertex must equal the
A. Graph Visualization of the SPS flows exiting the vertex, except for the source and sink
The SPS topology shown in Fig. 1 can be visualized as a vertices.
graph. A graph G = {V, E} consists of a vertex set V and an
edge set E. The buses in the SPS are represented as vertices C. Proposed Method for Power Line Expansion
and lines become edges. Fig. 2 shows the graph visualization The power line expansion strategy is decided by filtering
of the SPS. the potential strategies. By comparing characteristics of the
resultant graphs after applying all potential strategies, the

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


2018-PSEC-0142 Page 4 of 8

strategies that lead to better graph algebraic connectivity and of corresponding loads. The higher the priority of a load node,
larger weighted sum of max-flow from the generator nodes to the larger is its weighting factor. The notation MFgn,ln denotes
load nodes are selected. The algebraic connectivity of the the maximum flow from the generator node (gn) to the load
graph after adding new lines is obtained by calculating the node (ln).
second-smallest eigenvalue of the Laplacian matrix L.
Define the set of gn and ln as nodes with generators and loads, =∑ ∗∑ , (12)
respectively. The weighted sum of max-flow (WSMF) from
generator nodes to load nodes is calculated by (12) where the The maximization of WSMF and second-smallest
notation wln is the weighting factor for the load node ln. eigenvalue of Laplacian matrix are two objectives for the
Weighting factors of load nodes are determined by priorities system resilience enhancement. For multi-objective
start optimization, the set of Pareto optimality is desired. Pareto
optimality is to determine if an allocation is optimal. If an
allocation is not optimal, an alternative allocation can improve
Sort all the potential power lines
expansion strategies (s1, s2, … sk)
at least one objective of the condition while not reducing the
utilities of any other objectives. In other words, a solution is
called Pareto optimality when it is not dominated by another
Calculate WSMF and second-smallest eigenvalues of solution for every objective. The definition of Pareto
Laplacian matrix (λ2) after applying ith expansion optimality is:
strategy (si)
The set of allocation X is Pareto optimal iff ∄ X’, such
that every utility function i (ui), ui(X) < ui(X’), ∀i. [26]
Compare resulting WSMF and λ2 by applying
Specifically, two objectives are important in this study:
strategy i (WSMFk and λ2k) with existing solutions in maximizing algebraic connectivity and WSMF. Therefore, the
Pareto optimality set goal of the proposed strategy is finding the Pareto optimality
set regarding these two objectives. Fig. 3 depicts the flowchart
of proposed algorithm. At first, the value of WSMF and
second-smallest eigenvalue of the Laplacian matrix L of
Is the strategy i (si)
the resultant graph representatives are obtained after applying
dominated by an alternative N all potential strategies. Then, the Pareto optimality set is
in existing Pareto optimality determined by comparing performance of potential strategies.
set for both objectives? The strategy dominated by an alternative for both objectives is
eliminated. The algorithm ends when no potential strategy is
left without comparing.
Y
IV. NUMERICAL STUDIES
Y
i=k? This section presents results of numerical studies tested on
the SPS shown in Fig. 1. The model of MCR, as described in
N Section II, has been implemented in General Algebraic
i=i+1 Modelling System (GAMS), which is a high-level modeling
system for mathematical optimization [27]. On the other hand,
Put si in Pareto the proposed algorithm is implemented in MATLAB. To
optimality set begin with, the outcomes of MCR and the proposed method
are obtained and analyzed. After that, both the methods are
N Does the strategy i compared in terms of load shedding with physical attacks.
dominate any Without loss of generality, it is reasonable to assume that
alternative in existing
Pareto optimality set for every power line can be attacked. To obtain the weighted load
both objectives? shedding under various attack scenarios, MATLAB was used
to generate all possible attack scenarios. An attack scenario,
for instance, can be expressed as “l1 and l2 are attacked”.
Y Then, all the potential attack scenarios are processed in
Put si in Pareto optimality set, remove the GAMS to determine values of weighted load shedding for
strategy dominated by si for both objectives corresponding attack scenarios. The values of weighted load
from Pareto optimality set
shedding are determined by solving an optimization problem
with the objective of minimizing weighted load shedding,
subject to the system constraints given in (1)(5). For brevity,
end
the model is not presented here; readers can check [10] for
Fig. 3. Flowchart of proposed method detailed model.

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


Page 5 of 8 2018-PSEC-0142

A. Expansion Strategies using Different Methods into two lines with each having a capacity of 1 MW. This is
Table II tabulates the power line expansion strategy using because with more lines to be added, there exists a higher
MCR for different added line capacities ( ) and degree of flexibility. For instance, adding a single line with
maximum number of added lines ( ). From the results, it the capacity of 2 MW is equivalent to adding two lines with
can be seen that expansion strategy may change with values of capacities of 1 MW each, but restricted to the same location.
which shows the effects of line capacities on With such a restriction, the capacities of added lines are more
expansion strategy. likely to be not fully used due to the congestion in other parts
As mentioned earlier, in the proposed method two of the system. However, when more edges are added, it is
objectives are being maximized simultaneously, viz., the more likely to explore their capacities by placing them
algebraic connectivity and weighted sum of max-flow from properly. Therefore, it can be concluded that adding lines with
generator nodes to load nodes. Fig. 4, Fig. 5, and Fig. 6 show higher capacities will increase weighted maximum power
the Pareto fronts of both objectives under different capacities flow from generators to loads. Moreover, adding more lines
of added lines for Amax=1, Amax=2, and Amax=3, respectively. It can be even more beneficial since it increases both algebraic
is assumed that weights of loads on bus (N22 N21 N20 N19 connectivity of graph and weighted max-flow from generator
N7 N8 N18 N15) in Fig. 1 are (1 8 1 1 5 5 10 10), respectively. nodes to load nodes.
Results show that, with higher values of , the resultant B. Comparison between the Different Methods
values of WSMF increase, which indicates that maximum TABLE II. RESULTS OF BASELINE MCR METHOD
power flow between generator nodes and load nodes is higher. =1 =2 =3
Comparing Fig. 4, Fig. 5, and Fig. 6, it can be observed that =1 N3-N15 N3-N15/N13-N11 N3-N15/N2-N4/N2-
the second-smallest eigenvalue (λ2) of Laplacian matrix N4
=2 N3-N15 N3-N15/N22-N14 N3-N15/N13-
increases with the value of Amax, which indicates better N15/N13-N15
connectivity. This is because higher edge redundancy leads to =3 N3-N15 N3-N15/N14-N15 N3-N15/N13-
better connected network. In addition, with more lines added, N15/N13-N15
the value of WSMF rises. For instance, comparing Fig. 4 (b) =4 N3-N15 N3-N15/N1-N18 N3-N15/N1-
N18/N1-N18
and Fig. 5 (a), it can be observed that the value of WSMF
increases by splitting a single line with a capacity of 2 MW
λ2
λ2

λ2

λ2

WSMF ((M) WSMF (M)( WSMF (M)


( WSMF (M)
(
((a)) ((b)) (c) (d)
( )
Fig. 4. Pareto Fronts if one line is added (Amax=1) for different capacities of added lines (PFmaxl), (a) PFmaxl = 1MW, (b) PFmaxl = 2MW, (c) PF
max
l = 3MW, (d)
PFmaxl = 4MW
λ2

λ2
λ2

λ2

WSMF (M) WSMF


SMF (M)
(M WSMF (M) WSMF (M)
(
(a) (b) (c) (d)
Fig. 5. Pareto Fronts if two lines are added (Amax=2) for different capacities of added lines (PFmaxl), (a) PFmaxl = 1MW, (b) PFmaxl = 2MW, (c) PFmaxl = 3MW, (d)
PFmaxl = 4MW
λ2
λ2

λ2

λ2

WSMF (M) WSMF (M)


( WSMF (M)
(
(a) WSMF (M)
( )
(b) (c) (d)
Fig. 6. Pareto Fronts if three lines are added (Amax=3) for different capacities of added lines (PFmaxl), (a) PFmaxl = 1MW, (b) PFmaxl = 2MW, (c) PFmaxl = 3MW,
(d) PFmaxl = 4MW

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


2018-PSEC-0142 Page 6 of 8

method, one solution from the Pareto optimality set is selected.


Third quartile Then, the selected solution and the solution shown in Table II
Maximum
are used to get the corresponding values of WLsh for all attack
Median
scenarios.
Outliers
ℎ=∑ ∗ ℎ (13)
Minimum

When two lines are added (Amax=2), distributions of values


First quartile of WLsh for various numbers of attacked lines, capacities of
Fig. 7. Illustration of box plot added lines, and expansion methods are shown in Fig. 8
through Fig. 10. By using the proposed method, the values of
To demonstrate the merits of proposed method over MCR WLsh reduce for all casesin terms of median, quartiles,
method in terms of enhanced resilience, this subsection maximum, and outliers. This demonstrates enhanced
presents comparison of total weighted load shedding (WLsh) resilience against physical attacks on lines. More attacked
under all attack scenarios, calculated by (13). The notations wn lines would lead to higher values of WLsh, which can be
and Lshn represent weighting factor of load on bus n and easily understood as more disabled lines lead to higher load
corresponding load shedding. It is assumed that weights of shedding. Besides, with higher capacity of added lines, the
loads on bus (N22 N21 N20 N19 N7 N8 N18 N15) in Fig. 1 load shedding under attack can be alleviated. This is because
are (1 8 1 1 5 5 10 10), respectively. In this study, box plots congestion will be less likely to occur with higher line
are used to present the distribution of values of WLsh under all capacities.
attack scenarios. The box plot, shown in Fig. 7, is a standard
way to illustrate the distribution of groups of data through
quartiles [28]. To compare the proposed method and the MCR
WLsh

WLsh

WLsh

Method Method Method


(a) (b) (c)
Fig. 8. Box plots of weighted load shedding under all attack scenarios with 1 line to be attacked for MCR method and proposed method with 2 added lines
max
(A =2), (a) =1MW, (b) =2MW, (c) =3MW
WLsh

WLsh
WLsh

Method Method Method


(a) (b) (c)
Fig. 9. Box plots of weighted load shedding under all attack scenarios with 2 lines to be attacked for MCR method and proposed method with 2 added lines
(Amax=2), (a) =1MW, (b) =2MW, (c) =3MW
WLsh
WLsh

WLsh

Method Method Method


(a) (b) (c)
Fig. 10. Box plots of weighted load shedding under all attack scenarios with 3 lines to be attacked for MCR method and proposed method with 2 added lines
(Amax=2), (a) =1MW, (b) =2MW, (c) =3MW

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


Page 7 of 8 2018-PSEC-0142

Figs. 11 – 13 present the box plots of WLsh for various Furthermore, the two methods are compared in terms of the
numbers of attacked lines, capacities of added lines, and number of attack scenarios that can lead to load shedding.
expansion methods when three lines are added (Amax=3). For Table III shows the results for different maximum numbers of
their effects on values of WLsh, the same conclusions can be attacked lines and expansion methods, assuming two lines are
drawn as the case with two added lines. A comparison added. It is observed that the number of attack scenarios
between two and three added lines highlights that load causing load shedding events in the SPS for the original
shedding is alleviated when more lines are added. Therefore, method is the highest. By contrast, the number of attack
by adding more lines, the resiliency can be enhanced against scenarios causing load shedding events is the lowest with the
physical attacks on lines. proposed method. These results have once again demonstrated
TABLE III. NUMBERS OF ATTACK SCENARIOS CAUSING LOAD the enhanced resilience offered by the proposed power line
SHEDDING expansion method.
1 attacked line (# of attack 2 attacked lines (# of attack
scenarios causing load scenarios causing load
shedding / total # of attack shedding / total # of attack
V. CONCLUSION
scenarios) scenarios) This paper presents a graph theory-based method for
Original 10/27 233/351 power line expansion on a ship power system (SPS) aimed at
system
MCR 6/29 167/406
enhancing resilience against physical attacks on the power
method lines. The proposed method simultaneously maximizes the
Proposed 5/29 138/406 graph algebraic connectivity and weighted max-flow from
method generator nodes to load nodes. This method considers line
capacities and load priorities of a SPS in making power line
expansion decisions.
WLsh

WLsh

WLsh

Method Method Method


(a) (b) (c)
Fig. 11. Box plots of weighted load shedding under all attack scenarios with 1 line to be attacked for MCR method and proposed method with 3 added lines
max
(A =3), (a) =1MW, (b) =2MW, (c) =3MW
WLsh

WLsh
WLsh

Method Method Method


(a) (b) (c)
Fig. 12. Box plots of weighted load shedding under all attack scenarios with 2 lines to be attacked for MCR method and proposed method with 3 added lines
(Amax=3), (a) =1MW, (b) =2MW, (c) =3MW
WLsh
WLsh

WLsh

Method Method Method


(a) (b) (c)
Fig. 13. Box plots of weighted load shedding under all attack scenarios with 3 lines to be attacked for MCR method and proposed method with 3 added lines
(Amax=3), (a) =1MW, (b) =2MW, (c) =3MW

978-1-5386-1053-4/18/$31.00 © 2018 IEEE


2018-PSEC-0142 Page 8 of 8

A comprehensive evaluation was conducted on the SPS [19] R. Fang, R. Shang, Y. Wang, X. Guo, “Identification of Vulnerable
Lines in Power Grids with Power Integration Based on a Weighted
resilience against various physical attack scenarios. The Entropy Analysis Method,” Internat. Jour. of Hydro Energy, vol. 42, no.
numerical studies demonstrated a whole range of advantages 31, Aug. 2017, pp. 20269-20276.
of using proposed method in reducing load shedding due to [20] X. Wu and A. J. Conejo, “An Efficient Tri-Level Optimization Model
attacked lines. In addition, the effects of the number of added for Electric Grid Defense Planning,” IEEE Trans. Power Syst., vol. 32,
lines and their capacities on the resilience are investigated. It no. 4, Jul. 2016, pp. 2984-2994.
is found that adding more lines with higher capacities is [21] X. Zhang, K. Tomsovic, A. Dimitrovski, “Security Constrained Multi-
Stage Transmission Expansion Planning Considering a Continuously
beneficial to enhancing SPS resilience against physical attacks Variable Series Reactor,” IEEE Trans. Power Syst., vol. 32, no. 6, pp.
on power lines. 4442-4450, Nov. 2017.
[22] J. Arroyo, “Bilevel programming applied to power system vulnerability
REFERENCES analysis under multiple contingencies”, IET Gener. Transm. Distrib.,
[1] Navy’s New “All-Electric” Destroyer is a Seagoning Microgrid [online]. vol. 4, no. 2, pp. 178-190, Feb. 2010.
Available: https://cleantechnica.com/2013/11/05/us-navy-launches-new- [23] F. Chung, “Spectral Graph Theory,” Providence, RI: Amer. Math. Soc.,
all-electric-zumwalt-destroyer/ 1997.
[2] Zumwalt-class Destroyer [online]. Available: https: //en.wikipedia.org [24] Z. Song, G. Ren, L. Mirabella, S. Srivastava, “A Resilient Metric and Its
/wiki/Zumwalt-class_destroyer Calculation for Ship Automation Systems,” Proc. Resilience Week, Aug.
[3] Type 45 Destroyer [online]. Available: https://en.wikipedia.org 16-18, 2016.
/wiki/Type_45_destroyer [25] Maximum flow problem [online]. Available:
[4] Z. Jin, G. Sulligoi, R. Cuzner, L. Meng, J. Vasquez, and J. Guerrero, https://en.wikipedia.org/wiki/Maximum_flow_problem.
“Next-Generation Ship DC Power System: Introduction Smart Grid and [26] Pareto Efficiency [online],
DC Microgrid Technologies into Maritime Electrical Networks,” IEEE https://en.wikipedia.org/wiki/Pareto_efficiency
Electr. Mag., vol.4, no. 2, pp. 45-57, Jun 2016. [27] GAMS [online], https://www.gams.com/
[5] M. G. Balchanos, “A probabilistic technique for the assessment of [28] Box Plot [online]. Available: https://en.wikipedia.org/wiki/Box_plot.
complex dynamic system resilience,” Ph.D. dissertation, Georgia
Institute of Technology, 2012.
[6] P. W. Parfomak, “Physical Security of the U.S. Power Grid: High-
voltage Transformer Substations,” Congressional Research Service,
June 17, 2014.
[7] December 2015 Ukraine Power Grid Cyberattack [online].
https://en.wikipedia.org/wiki/December_2015_Ukraine_power_grid_cy
berattack
[8] K. Lai, M. Illindala, “Design and Planning Strategy for Energy Storage
System in a Ship DC Hybrid Power System,” Proc. 53rd IEEE/IAS Ind.
Commerical Power Syst. Tech. Conf., May 6-11, 2017, pp. 1-9.
[9] K. Lai, M. Illindala, “Enhancing the Robustness of Ship DC Hybrid
Power System Against Generator Failures,” Proc. IEEE Electric Ship
Technologies Symposium, Aug. 14-17, 2017, pp. 340-344.
[10] S. Jothibasu, S. Santoso, “New Electric Shipboard Topologies for High
Resiliency,” IEEE Trans. Power Syst., in press, 2017.
[11] A. Dubey, S. Santoso, “Availability-Based Distribution Circuit Design
for Shipboard Power System,” IEEE Trans. Smart Grid, vol. 8, no. 4,
pp. 1599-1608, Jul. 2017.
[12] C. Su, C. Lan, T. Chou, C. Chen, “Performance Evaluation of
Multiagent Systems for Navy Shipboard Power System Restoration,”
IEEE Trans. Ind. Appl., vol. 51, no. 4, pp. 2769-2779, Jan. 2015.
[13] M. Nelson, P. Jordan, “Automatic Reconfiguration of a Ship’s Power
System Using Graph Theory Principles,” IEEE Trans. Ind. Appl., vol.
51, no. 3, pp. 2651-2656, Oct. 2014.
[14] F. Shariatzadeh, N. Kumar, A. Srivastava, “Optimal Control
Algorothms for Reconfiguration of Shipboard Microgrid Idstribution
System Using Intelligent Techniques,” IEEE Trans. Ind. Appl., vol. 53,
no. 1, pp. 474-482, Aug. 2016.
[15] J. Li, X. Ma, C. Liu, K. Schneider, “Distribution System Restoration
with Microgrids Using Spanning Tree Search,” IEEE Trans. Power
Syst., vol. 29, no. 6, pp. 3021-3029, Nov. 2014.
[16] O. Swathika, S. Hemamalini, “Prims-Aided Dijkstra Algorithm for
Adaptive Protection in Microgrids,” IEEE Jour. Emerg. And Selec.
Topics in Power Electron., vol. 4, no. 4, pp. 1279-1286, Dec. 2016.
[17] P. Bajpai, S. Chanda, A. Srivastava, “A Novel Metric to Quantify and
Enable Resilient Distribution System using Graph Theory and Choquet
Integral,” IEEE Trans. Smart Grid, in press, 2017.
[18] Y. Zhu, J. Yan, Y. Sun, H. He, “Revealing Cascading Failure
Vulnerability in Power Grids Using Risk-Graph,” IEEE Trans. Para.
and Dist. Syst., vol. 25, no. 12, Dec. 2014.

978-1-5386-1053-4/18/$31.00 © 2018 IEEE

Das könnte Ihnen auch gefallen