3774 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO.

5, SEPTEMBER 2022

Optimal Communication Network Design of

Microgrids Considering Cyber-Attacks
and Time-Delays
Lina Sheng, Graduate Student Member, IEEE, Guannan Lou , Member, IEEE, Wei Gu , Senior Member, IEEE,
Shuai Lu , Member, IEEE, Shixing Ding, Graduate Student Member, IEEE, and Zhigang Ye , Member, IEEE

Abstract—Distributed secondary control stands out for its Index Terms—Communication network design, convex relax-
flexibility and expandability in microgrids (MGs) control, in ation, structural survivability, speed of convergence, robustness
where communication network plays a fundamental and critical to time-delay, the multiple deliberate cyber-attack, microgrid,
role. The communication topology and link-weights have signif- consensus-based secondary voltage control.
icant impact on the attack-resilience and dynamic performance
of MGs, which should be appropriately designed. However,
the joint optimization problem of communication topology and I. I NTRODUCTION
link-weights in the existing literature of consensus-based sec- ITH the intensification of global energy crisis and
ondary voltage control of MGs is usually neglected. To bridge
this gap, we propose a novel two-stage optimization approach
W environmental pollution, building a cleaner and lower-
carbon energy system is an inevitable choice for human
for the communication network design, which jointly optimizes
the topological structure to enhance the structural survivability beings. As an effective organization form of DGs, MG plays
and link weights to improve the dynamic performance (includ- an increasingly noteworthy role in enhancing the reliability
ing the speed of convergence and robustness to time-delay). and flexibility of power supply, especially in remote areas
The first stage problem is formulated into a mixed-integer or islands. MG with secondary control layer can be seen
semi-define programming (MISDP) model based on convex relax-
ation technique, which is then converted equivalently into an as a typical cyber-physical system (CPS). DGS , local con-
integer quadratic programming (IQP) problem and then a trollers, sensors, loads, power lines, etc., constitute the physical
MISDP feasibility problem to facilitate the solution. The second layer [1], while terminal agent modules, communication links,
stage problem is formulated into a bi-objective SDP model to and so on build up the cyber layer [2]. The existing sec-
compromise between the convergence performance and robust- ondary control modes of MG mainly include centralized con-
ness to time-delay. Simulations based on a microgrid with
10 distributed generation (DG) units under different scenar- trol, decentralized control, and distributed control [3], among
ios are implemented to verify the effectiveness of the proposed which the distributed control is increasing popular for its flexi-
method. bility and scalability. In the distributed control architecture, the
information exchange between adjacent DGs is usually real-
ized through a sparse communication network, which directly
affects both the stability and performance of MG systems.
Manuscript received 23 November 2021; revised 4 March 2022; accepted On the one hand, with the increasing application of wire-
16 April 2022. Date of publication 27 April 2022; date of current version less communication technologies such as 5G and Wi-Fi in
23 August 2022. This work was supported in part by the National Science the communication of MG, the issue caused by cyber-attacks
Foundation of China under Grant 52007031; in part by the National Science
Foundation of Jiangsu Province under Grant BK20200404; in part by the of MGs becomes more and more critical. The communica-
“Zhishan” Young Scholars Programs of Southeast University; and in part by tion network as the key of distributed secondary control of
the Project Funding for Jiangsu Provincial Key Laboratory of Smart Grid MGs is very vulnerable to malicious cyber-attacks like false
Technology and Equipment, Southeast University. Paper no. TSG-01869-2021.
(Corresponding author: Guannan Lou.) data injection (FDI) attack, denial of service (DoS) attack,
Lina Sheng and Shixing Ding are with the School of Cyber spoofing attack, replay attack, etc. Many works have been
Science Engineering, Southeast University, Nanjing 211189, China (e-mail: implemented on the cyber-attacks in MGs, and most of them
[email protected]; [email protected]).
Guannan Lou is with the School of Electrical Engineering and focus on the risk assessment metric, attack detection and iso-
the Jiangsu Provincial Key Laboratory of Smart Grid Technology lation technology, and resilient control strategy design [4]–[6].
and Equipment, Southeast University, Nanjing 210096, China (e-mail: In the resilient control strategy, many existing works are
[email protected]).
Wei Gu and Shuai Lu are with the School of Electrical Engineering, carried out based on the specific characteristics and impact
Southeast University, Nanjing 210096, China (e-mail: [email protected]; of cyber-attacks, aiming at mitigating the negative effect of
[email protected]). cyber-attacks on MGs [7]–[10]. A high-order sliding-mode
Zhigang Ye is with the Electric Power Research Institute, State Grid Jiangsu
Electric Power Company Ltd., Nanjing 211103, Jiangsu, China, and also observer is implemented to estimate system states in [7], and
with Southeast University, Nanjing 210096, China (e-mail: yzhggoodluck@ the state is used to reconstruct the attack signal to mitigate
hotmail.com). the impact from FDI attack. In [8], an attack-resilient event-
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TSG.2022.3169343. triggered control strategy is proposed to stabilize a DC-MG
Digital Object Identifier 10.1109/TSG.2022.3169343 system in the presence of DoS attack. A resilient controller
1949-3053 
c 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3775

of DC microgrid is presented to achieve current sharing and consensus-based secondary control in MGs can be improved
voltage restoration under discrete-time FDI and DoS attack by optimizing the link weights [18], [19].
in [9]. In [10], a distributed discrete-time event-triggered con- Moreover, the convergence speed and the robustness to time-
sensus control strategy is designed, and a switching framework delay are equally important for consensus-based secondary
between DoS attack intervals and communication is applied to controllers in MGs, both of which greatly affect the dynamic
determine sufficient conditions for safe current sharing. In the performance of microgrid. Many works have been done to
above research, though the resiliency to cyber-attacks has been make the voltage and frequency of MGs converge to ref-
improved, they ignore the evitable impact of communication erence values within a finite time [20], [21]. However, few
network on system performance. papers focus on simultaneously improving the convergence
In addition to the above-mentioned cyber-attacks aimed speed and robustness to communication delays in microgrid
at manipulating communication information (such as FDI control. In the optimization of link weights, there is a con-
attacks, replay attacks, etc.) and changing the communication tradiction between the convergence speed and the robustness
state (such as DoS attacks, etc.), some more serious destructive to communication delay [18], [19]. Although the secondary
cyber-attacks will directly lead to the failure of communication controller parameters are designed considering the trade-off
nodes, which should be paid more attention. The communica- between the convergence speed and the robustness to commu-
tion network with high attack-resilience can keep connected nication delay in [19], the influence of communication link
after lots of nodes and corresponding links are destroyed, weights is ignored. The link weights design of the communi-
i.e., high structural survivability under cyber-attacks, meaning cation network is in fact a complicated nonconvex optimization
that it is necessary to improve the communication network model, for which both the optimality and the computational
to enhance the attack-resilience of MGs. However, it has not efficiency are troublesome, which has not been fully addressed
obtained enough attention to improving the attack-resilience up to now. Besides, as far as the authors know, in the exist-
of MGs by the optimizing communication topology, leaving a ing research, there is no effective optimization scheme to well
research gap. balance the robustness to time-delay and convergence speed
Besides, from the perspective of attack probability, cyber- when optimizing the communication link-weights of MGs.
attacks can be divided into the random attack and deliberate To bridge the research gaps mentioned above, this paper pro-
attack [11]. This paper mainly focuses on the deliberate poses a novel two-stage communication network optimization
attack, since it is more destructive to the system stability approach for the consensus-based secondary voltage control of
than the random attack. At present, the study on improv- MGs. The first stage aims to enhance the attack-resilience of
ing the attack- resilience of MGs under random or deliberate the MGs under the multiple deliberate cyber-attack. Based on
cyber-attacks is rare. In [12], a corresponding resilient con- the optimal topology, the second stage aims to optimize the
troller of MG is designed against the random DoS attack. link weights to improve the dynamic performance of MGs.
However, the attack-resilient scheme against the deliberate The main contributions are summarized as follows.
cyber-attack in MGs is neglected. Besides, few studies focused (1) Both the improvement of attack-resilience and dynamic
on how to mitigate the impact of deliberate cyber-attacks performance (including the robustness to communication time-
on the MGs from the perspective of communication network delay and convergence speed) in MGs are considered in the
optimization. process of communication network optimization.
On the other hand, there are some inherent issues in the (2) A structural survivability index based on the variance
communication network, such as data packet loss [13], com- of node degree is first proposed for the communication topol-
munication time-delay [14]–[16], communication noise [17], ogy design of MGs to enhance the attack-resilience of system
channel congestion, etc., which will negatively affect the under the multiple deliberate cyber-attack, which is ignored in
dynamic control performance of MGs. Among them, the existing literatures.
communication delay is a tricky problem in engineering appli- (3) A mixed-integer semidefinite programming (MISDP)
cations, which has attracted a lot of attention in the field of model is established for the first-stage topology optimization
microgrid distributed control. The relevant research mainly based on the convex relaxation technique, and then it is equiva-
includes analyzing stability robustness of MGs with commu- lently converted into an integer quadratic programming (IQP)
nication time delays, designing appropriate control strategies main problem and an auxiliary SDP feasibility sub-problem
to alleviate the negative impact of time-delay, and so on. The for efficient solution.
influence of the communication delay on the consensus-based (4) A bi-objective SDP model is established for the second-
secondary controller in DC MGs is analyzed in [14], and the stage link weight optimization, in which the convergence speed
sensitivity expression of the maximum time-delay to system and robustness to communication delay of consensus-based
parameters during the steady-state operation of the MG is secondary control in MGs can be well balanced.
proposed. A H ∞ resilient distributed consensus-based control The remainder of this paper is organized as follows:
scheme is presented in [15], considering the uncertain commu- Section II describes the preliminary of graph theory and
nication time-delays. In order to eliminate the adverse effects consensus-based secondary voltage control of MGs. The
of time delay, a controller based on sliding mode estimation is multiple deliberate cyber-attack and three indicators involved
designed in [16] to predict time delay and MG states. Besides, in the optimization of communication topology are intro-
apart from the above methods, from the perspective of the duced in Section III. The two-stage communication network
communication network design, the time delay margin of the optimization scheme is put forward in Section IV. Simulations

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 3776 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO. 5, SEPTEMBER 2022

under different scenarios are presented in Section V to ver-

ify the effectiveness of the proposed solution. In the end,
conclusions are given in Section VI.

II. P RELIMINARIES OF G RAPH T HEORY AND

C ONSENSUS -BASED VOLTAGE C ONTROL OF M ICROGRIDS
A. Graph Theory
According to the graph theory, we know that each DG unit
can be seen as a node, and communication links between
DGs can be regarded as edges. Communication links in MGs
usually are considered to be bi-directional, that is, all edges
are undirected. The communication link between DGi and
DGj is regarded as eij . Therefore, the communication network
in a MG can be depicted as a weighed undirected graph
G = (V, W). V = {v1 , v2 , . . . , vn } is the finite set of nodes and
I = {1, 2, . . . , n} is the set of node labels. W = [aij ] ∈ Rn×n
represents the weighted adjacency matrix of G and aij is the Fig. 1. The CPS structure of the microgrid under cyber-attacks.
weight of edge between vj and vi . Since self-loops are not con-
sidered, there is aij > 0 for all i, j ∈ I and i = j. The degree of weighted average voltage estimator is designed to obtain the
vi refers to the number of edges connected to it, which can be average voltage of DGs as:
denoted by ki , reflecting the local characteristics of the node.  
 
D = diag{di } ∈ Rn×n represents the in-degree  matrix of vi in a Ūi = Ui (t) + CE ij Ūj − Ūi dt
aU (4)
weighted undirected graph, where di = nj=1 aij . In addition, j∈Ni
the Laplacian matrix of G satisfies the relationship L = D−W.
Q
wherein aU ij is the adjacency matrix element, and aij = aij
U

B. Consensus-Based Secondary Voltage Control of MGs since the reactive power sharing and the weighted average
voltage observation in this paper are implemented in the same
According to the Consortium for Electric Reliability
communication network. CE represents the voltage coupling
Technology Solutions (CERTS) [22], [23], the droop control
coefficient, which is required for stability. Ūi and Ūj are
can be formulated as:
 the estimated average voltage of DGi and DGj , respectively.
ωi = ωni − mi Pi The cooperative secondary voltage control strategy can be
(1)
kVi U̇i = Uni − Ui − ni Qi described as follows [19]:
wherein ωni is the desired value of angular frequency. Qi and KVi U̇i = Uni− Ui − ni Qi + ui (5)
Pi represent the reactive power and active power, respectively.   
ni is the frequency droop coefficient, mi is the voltage droop ui = Ki uQi + βi Uref − Ūi dt (6)
coefficient, Uni is the nominal voltage amplitude, and KVi is
the voltage control coefficient and there is KVi = 1 [19]. Two In (5), ui represents the secondary voltage adjustment term; Ki
control goals are required to be met to realize the consensus- is the coefficient of controller integral term; β i is the weight
based cooperative voltage control of MGs. One is adjusting the coefficient of reactive power sharing and voltage restoration;
terminal voltage of the inverter to the rated value. Another is Uref indicates the rated voltage reference value of the system.
achieving accurate reactive power sharing. However, according
to the characteristic curve of voltage droop, it is difficult to III. I NDEXES FOR C OMMUNICATION N ETWORKDESIGN
achieve these two control objectives at the same time. Thus, IN M ICROGRIDS
a compromise between them should be considered. Reactive The communication network is key for realizing the dis-
power sharing refers to that each DG provides corresponding tributed secondary control, which greatly affects the attack-
reactive power proportionally according to ni . The relationship resilience and dynamic performance of MGs. In this section,
between ni and Qi is shown as follows. we mainly establish three optimization indicators for the
communication network design of MGs from the perspec-
n1 Q1 = n2 Q2 = · · · = ni Qi (2)
tive of cyber-security and dynamic performance, including
Based on a consensus algorithm, the deviation of reactive communication topology and link-weights.
power sharing can be expressed as:
 Q  A. Invulnerability of Communication Network in MGs
uQi = CQ aij nj Qj − ni Qi (3)
j∈Ni
1) The Multiple Deliberate Cyber-Attacks in MGs : Based
on Section II-B, the CPS structure of MG with the consensus-
Q
wherein aij represents the communication link-weight between based secondary voltage controller under cyber-attacks is
DGi and DGj . CQ refers to the reactive power coupling shown as Fig. 1. The communication network in Fig. 1 is vul-
coefficient which is required for stability [24]. A distributed nerable to cyber-attacks. According to the probability of each

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3777

cyber-attack, we should try to ensure that the remaining nodes

in the original communication network are still connected. The
ability of the communication topology to maintain connectivity
under the multiple deliberate cyber-attack is called structural
survivability in this paper.
Node degree is an important index of the topology, which
reflects the importance of the node in the communication
network. Hence, in order to improve the attack-resilience of
Fig. 2. Examples of random cyber-attack and deliberate cyber-attack. communication network under the multiple deliberate cyber-
attack, node-degree of each DG should be as homogeneous as
possible. In graph theory, the entropy of degree distribution is
often used to measure the homogeneity of the nodal degrees,
communication node being attacked, there are random cyber- as follows.
attack and deliberate cyber-attack. Random attack means that
the hacker launches an indiscriminate attack on each node. 
N
Hentropy = − p(ki ) ln p(ki ) (7)
That is, the probability of each DG being attacked is equal. i=1
Hackers always launch random attacks when they are unaware
of the communication topology. However, once hackers have Wherein N is the total number of nodes in the network; p(ki )
access to relevant information about the communication topol- is the distribution of node degrees, which is the ratio of nodes
ogy through some special means, they will carry out precise with degree di in the network and can be calculated as follows.
strikes on some communication nodes selectively. θi
Specifically, the difference between random cyber-attack p(ki ) = (8)
N
and deliberate cyber-attack occur in an island MG with five
DGs can be illustrated based on Fig. 2. Under the random where θ i is the number of nodes with degree ki , and di ∈ {1,
cyber-attack, each DG in the microgrid will be attacked with 2, . . . , N-1}. However, if we use Hentropy as the objective of
the same probability. When DG1 is attacked, e15 and e25 are the network design problem, extra binary variables need to be
interrupted and DG1 enters the local control state, and the introduced to link p(ki ), θ i , and the decision variable D, and the
remaining four DGs can continue to operate in the consensus- objective of the optimization model is nonlinear. Besides, the
based secondary control mode. While under the deliberate function ln(x) is defined on (0, +∞), meaning that numerical
cyber-attack, DG5 with the largest degree (k5 = 4) is the first problem will occur when p(ki ) = 0.
to be attacked and e15 , e25 , e35 and e45 are interrupted. The Inspired by the definition of the degree distribution, a novel
remaining communication nodes in Fig. 2(b) are divided into structural survivability index H 1 based on the degree devia-
two isolated parts: DG1 and DG2 are connected, DG4 and DG5 tion among communication nodes is proposed in this paper for
are connected. Thus, the consensus-based distributed voltage communication topology optimization. In contrast to Hentropy ,
control strategy of MG fails under the deliberate cyber-attack, the index H 1 is the quadratic function of the decision vari-
which is shown in Fig. 2(b). able D, which will greatly decrease computational cost of the
What’s more, the implementation process of the multiple network design problem. Therefore, we use the variance of the
deliberate attack is as follows. Hackers usually master certain nodal degree to measure the homogeneity of nodal degrees,
topological information of the MG communication network which is
in advance by exploiting some communication loopholes, 2
H1 = D − D̄ · I 2
(9)
and then penetrate the entire communication system or
related applications through some special means [25]–[27]. The smaller the value of H 1 is, the stronger the struc-
The multiple deliberate cyber-attack is further launched in tural survivability of the communication topology under the
order of the node-degrees from large to small until the com- multiple deliberate cyber-attack will be.
munication topology composed of the remaining nodes is no
longer connected. The attacked communication node becomes B. Robustness to Communication Time-Delay
invalid, that is, unable to receive and send system interaction
Assume that the transmission time-delay of communication
information.
links is equal and τ ij > 0. The dynamic of linear consensus
2) Structural Survivability of Communication Topology: As protocol with time-delay can be described as:
mentioned above, the hacker usually launches multiple attacks      
in the order of node-degree from largest to smallest under ẋi (t) = aij xj t−τij − xi t−τij (10)
the multiple deliberate cyber-attack. Here, we assume that the j∈Ni
set of attacked communication nodes is N∗ and N∗ ∈ V, and
The Laplace transform of the above formula is:
the remaining communication nodes constitute the set S =
V − N ∗ = {vs } and s ∈ {1, 2, . . . ,n}. It is worth noting that  −1
X(s) = sI + e−τij s L x(0) (11)
whether all nodes in S can form a connected topology is the
key to reaching a consensus. Therefore, when some nodes in L is the Laplacian matrix of the undirected and connected
the communication network fail under the multiple deliberate graph G. When the following condition is met, (10) can realize

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 3778 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO. 5, SEPTEMBER 2022

global asymptotic stability with time-delay τ ij and the state of

vi converges to the average value [18].
  π
τij ∈ 0, τ ∗ , H2 = τ ∗ = (12)
2λmax (L)
wherein λmax (L) is the greatest eigenvalue of L. τ ∗ is the
maximum communication time-delay that the consensus-based
secondary control of the system can tolerate, which is defined
as the optimization indicator H 2 . According to the Gerschgorin
theorem, the following inequality holds:
⎧ ⎫
⎨ ⎬
λmax (L) ≤ 2dmax (G) = 2 max degout (vi ) = 2 max aij
⎩ ⎭
j∈Ni
(13)

In order to maximize the transmission time-delay that the

system can tolerate, we should maximize H 2 when optimizing
weights of communication links.

Fig. 3. Flow chart of the two-stage optimal communication topology design

C. Convergence Speed approach.
In the analysis of the convergence speed of the lin-
ear consensus algorithm, the spectral properties of the
Laplacian matrix L are very important. For an undirected IV. T WO -S TAGE C OMMUNICATION
graph, the Laplacian disagreement function of the graph is N ETWORKDESIGN A PPROACH
defined as: A. Design Procedure
1   2 The analysis above reveals that the parameters of commu-
G (x) = xT Lx = aij xj − xi (14)
2 nication topology, including the connection condition and link
vj ∈Ni
weights, are key factors for the performance of a linear con-
wherein x = [x1, x2, . . . , xn ]T . In addition, the Laplacian matrix sensus protocol. The cyber layer and physical layer of MGs
of the graph satisfies the following property [18], [19]: are closely coupled. Once the cyber system is paralyzed due to
a malicious cyber-attack, the operation of the physical system
xT Lx will be seriously affected. Therefore, when designing the com-
min = λ2 (L) (15)
x = 0 ||x||
2 munication topology of MGs, two critical problems should be
1T x = 0 focused on. The first problem is to enhance the structural sur-
vivability of the communication topology, which is the key to
According to the Gerschgorin Theorem, we know that all the the cyber-security of MGs. The second problem is to improve
eigenvalues of L are distributed in the complex plane closed the dynamic performance of MGs, including both the robust-
loop with diagonal elements as the center and λmax (L) as the ness to communication time-delay and the speed of uniform
radius. Sort all eigenvalues of L in ascending order: convergence.
The flowchart of the proposed two-stage optimal communi-
0 = λ1 (L) ≤ λ2 (L) ≤ · · · ≤ λn (L) ≤ 2dmax (G) (16) cation topology design approach is given in Fig. 3. The first
stage is to optimize the network topology to improve the struc-
wherein λ2 is the algebraic connectivity of the graph G and it tural survivability against cyber-attacks. The second stage is
satisfies the Fiedler inequality: to optimize the link weights of the network to improve the
dynamic performance. The detailed models are presented in
λ2 (L) ≤ γ (G) ≤ ξ(G) (17)
the following.
wherein γ (G) refers to the node connectivity of the
graph and ξ(G) represents the edge connectivity of the B. Stage 1: Topology Optimization
graph. If λ2 (L) increases, γ (G) and ξ(G) also increase, The proposed topology optimization model is as follows.
which means that the connectivity of G becomes better.
Therefore, another dynamic performance index is expressed as (P1) F1 (E) = min D − D̄ · I
2
(19)
2
follows: D,A,L
A = AT , Aii = 0 ∀i, Aij ∈ {0, 1} ∀i, j (20)
H3 = λ2 (L) (18)
1T D · 1 = 2E (21)
 
The larger H 3 is, the faster the convergence speed is. D = diag 1T A (22)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3779

L=D−A (23) 
k 
N
di ≤ k(k − 1) + vk,i ∀k = 1, . . . , N − 1
λ
⎧2 (L) ≥σ (24) i=1 i=k+1
T =L
⎪
⎪ L vk,i ≤ k ∀k = 1, . . . , N − 1, i = k + 1, . . . , N
⎪
⎪
⎨L 0 vk,i ≤ di ∀k = 1, . . . , N − 1, i = k + 1, . . . , N (27)
Lii ≥ 0 ∀i (25)
⎪
⎪
⎪
⎪ L·1=0 wherein vk,i are auxiliary variables to represent the minimum
⎩
L − diag(L) ≤ 0 value of k and di .
Although there exists a connected graph realization of d
The objective function in (19) is to minimize the variance of
(or D) obtained by the model (27), the solution of A is not
the nodal degree, wherein D̄ = 2E/N. Constraints (20) come
unique in this model and it could not yield a non-connected
from the properties of the adjacent matrix A. Constraints (21)
graph. Hence, the auxiliary feasibility problem is proposed in
define the relationship between the nodal degree and the num-
the following to find a connected graph based on the given D,
ber of edges. Constraints (22) define the relationship between
as follows.
the degree matrix D and the adjacent matrix A. Constraints (23)
define the Laplace matrix L by D and A. Constraints (24) (P1 − A) f (D) = min 0
define that the eigenvalue λ2 of L should be greater than 0 to L,A,γ ,β

ensure that the graph is connected, wherein σ is a small con- s.t. (20), (22), (23), (25), (26) (28)
stant. Constraints (25) define the constraints of the Laplace The model (28) is also a MISDP problem. However, com-
matrix L based on the spectral graph theory [28]. pared with the original problem (P1), it can be solved more
The (P1) problem consisting of (19)-(25) is a mixed- efficiently because of two reasons: (1) The integer variables are
integer nonlinear programming model because the eigenvalue greatly reduced; and (2) it only needs to find a feasible solu-
λ2 cannot be represented as an analytical expression of L. tion rather than the optimal solution. Numerical simulations
Alternatively, the constraint (24) can be relaxed into the also verify that it can be solved efficiently.
following semi-positive defined constraints [29].
γ I L + ξ 11T , γ ≥σ (26) C. Stage 2: Link Weight Optimization
The analysis in Section III indicates that the convergence
wherein ξ is an auxiliary variable and γ is the upper bound
rate of reaching the average consistency and the robustness
of λ2 .
of communication time-delay are two important indexes of
Replacing (24) by (26), the topology design model turns into
dynamic performance in MG control. Therefore, the objective
a MISDP model, which is still hard to solve. Therefore, in the
functions of the second-stage problem are given as follows.
following, we decompose the model (P1) into an IQP problem  
to optimize D and A and an auxiliary feasibility problem to π
F2 = min −λ2 (L), − (29)
find a connected graph, i.e., optimize A and L based on the L,λ2 ,λmax 2λmax (L)
given value of D. Based on the spectral graph theory, the Laplace matrix L of
Theorem 1 [28], [30]: A degree sequence d = the undirected graph should satisfy the constraints (25). Since
(d1 , 
. . . , dN ) is potentially connected if and only if di ≥ 1, ∀i the topology has been determined by the stage 1 problem, the
and N i=1 di ≥ 2(N − 1) , i.e., the number of edges is at least following constraint is added to ensure that L is consistent
N−1 . with the given topology.
Theorem 2 (Erdős-Gallai Theorem) [31]: A non-increasing
sequence of non-negative integers d = (d1 , d2 , . . . , dN ) is the − M · A ≤ L − diag(L) (30)
n
representation of a graph on N vertices if and only if i=1 di
k n wherein M is a constant with a large value (setting to 1000 in
is even and i=1 di ≤ k(k − 1) + i=k+1 min{k, di } for each
numerical simulations).
k ∈ {1, 2, . . . , N}.
Similar to the constraints (26), the following semi-positive
Based on Theorem 1 and 2, we can decompose the problem
define constraints are be used to provide an upper bound
(P1) into a degree optimization problem and a graph repre-
for λ2 .
sentation problem. The former aims to optimize the degree 
sequence (or degree matrix) on the premise that it is the rep- λ2 ≤ γ
(31)
resentation of a potentially connected graph. The latter aims to γ I L + ξ 11T
find a connected graph based on the given degree sequence (or Besides, since L is a symmetric matrix, λmax can be
degree matrix). Based on this, the degree matrix optimization calculated as follows.
problem can be formulated into the following IQP problem.
2
λmax = L 2 (32)
(P1 − D) F1 (E) = min D − D̄ · I 2
D,A,d,v Then, the problem consisting of (29)-(32) and (25) is a non-
s.t. (20) − (22) convex optimization since the existence of (32). To resolve this
d = diag(D) problem, we relax (32) into a second-order conic constraint,
1T d ≥ 2(N − 1) as follows.
di ≥ di−1 ∀i = 2, . . . , N λmax ≥ L 2 (33)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 3780 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO. 5, SEPTEMBER 2022

Finally, the second-stage problem can be reformulated into

an SDP problem, as follows.
F2 = min {−λ2 , λmax }
L,λ2 ,λmax ,γ ,β
s.t. (25), (30), (31), (33) (34)
Note that for a bi-objective optimization problem, there is
no essential difference between the objective functions in (34)
and those in (29). Also, since the objective functions in (34)
aim to maximize λ2 and minimize λmax , the constraints in (31)
and (33) are always tight, meaning that the problem (34) is
equivalent to the original problem.
To solve the bi-objective optimization problem in (34), a
commonly used method is the weighted sum method, based
on which the problem (34) can be converted into a single-
objective optimization problem, as follows.
F2 (α) = min α · (−λ2 ) + (1 − α) · λmax
L,λ2 ,λmax ,γ ,β
Fig. 4. The tested island microgrid with 10 DGs.
s.t. (25), (30), (31), (33) (35)
TABLE I
By choosing different values of α ∈ (0, 1), we can get E LECTRICAL AND VOLTAGE C ONTROL PARAMETERS OF THE T ESTED MG
the Pareto frontier of the second-stage optimization problem.
However, an interesting problem is that the problem (35)
always degenerates. To clarify this problem, we assume that
the solution of the problem (35) is L∗ , λ∗2 , and λ∗max . Then, if
0 < α < λ∗max /(λ∗max + λ∗2 ), L = 0 is always a better solution
than L∗ ; if λ∗max /(λ∗max + λ∗2 ) < α < 1, L = ML∗ , M > 1 is
always a better solution than L∗ . This means that the problem
(35) is numerically bad. This problem originates from the lin-
ear dependence of the two objective functions in (34). To
resolve this problem, we use (λmax )2 to replace the objective
function λmax in (34), which does not affect the solutions of the
bi-objective optimization problem (34). Hence, the problem
(34) is finally converted into a SDP problem as follows.
(P2) F2 (α) = min α · (−λ2 ) + (1 − α) · (λmax )2
L,λ2 ,λmax ,γ ,β
s.t. (25), (30), (31), (33) (36)
The SDP problem (36) can be solved by off-the-shelf solvers
such as Mosek and SeDuMi.

V. S IMULATION R ESULTS
To verify the effectiveness of the proposed method,
a 380V/50Hz islanded MG model is built in MATLAB
R2014b/Simulink platform. In the communication network
design stage, the IQP model is solved by Gurobi 9.5, the
MISDP feasibility problem is solved by cutsdp solver pro-
vided by Yalmip with Mosek 9.3 as the lower solver, and the
SDP model is solved by Mosek 9.3. All the simulations are
performed on a desktop with an Intel i7 CPU @3.60GHz and A. Analysis of Topology Optimization Results
8GB RAM. 1) Topology Optimization Results: Based on the description
The tested MG is composed of 10 DGs with different in Section III and Section IV, we optimize the communication
rated capacities, the structure of which is shown in Fig. 4. topology of the tested MG with 10 DGs. For the network
Parameters of DGs and consensus-based secondary voltage with 10 nodes, the number of edges should be between 9 and
control are given in Table I. Within 0 to 1s, only primary con- 45. Based on this, 37 optimization results and correspond-
trollers participate in the adjustment of voltage and frequency. ing optimal values of H1 under different edge numbers are
The cooperative secondary voltage control, as given in (5), is obtained as Fig. 5. It should be pointed out that when the num-
applied at 1s. ber of DGs is the same, the corresponding optimal values of H1

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3781

Fig. 5. The minimum value of index H 1 under different numbers of edges

(first-stage optimization).

Fig. 6. Optimal communication topology satisfying H 1 = 0.

Fig. 8. Dynamic performance of the tested microgrid in scenario 1.

equal to 6. In Fig. 7, attacked DGs are marked in red, and the

Fig. 7. Topological structure of communication network with 30 edges. remaining black DGs are safe and still participate in normal
communication interactions. Besides, the red dashed line indi-
cates that the communication link has been interrupted, and
under different edge-numbers may be equal, but this does not the black line indicates that information can be transmitted
mean that their communication topologies are the same. Each normally.
point in Fig. 5 represents a specific structure of communica- Scenario 1: According to the definition of the deliberate
tion topology. In addition, when H 1 = 0, the corresponding attack in Section III, DG with higher importance (i.e., greater
communication topology has the strongest structural surviv- node-degree) is more likely to be attacked. Here, we assume
ability under the multiple deliberate cyber-attack. In this case, that under the multiple deliberate cyber-attack, the correspond-
the degree of nodes in this communication topology is evenly ing communication nodes of DG1 , DG3 , DG6, and DG7 are
distributed, and all nodes have the same degree. For exam- attacked and quit normal communication at 2.5s, 4s, 5.5s
ple, the topologies that satisfy H 1 = 0 and contain different and 7s, sequentially. An external load with P = 50kw and
numbers of edges are shown in Fig. 6. Q = 80kvar is connected to the system at 9s. The link-
2) Structural Survivability Analysis: To demonstrate that weight from DGi to DGj in GQ and GU is set as 1 / di , i.e.,
Q
the optimized communication topology has good structural aij = aUij = 1/di .
survivability under the multiple deliberate cyber-attack, we Simulation results of the tested MG with the random com-
compare it with a random communication topology through munication topology in scenario 1 is presented in Fig. 8. The
simulation experiments. Fig. 7(a) presents a connected ran- proportional reactive power sharing consensus among DGs has
dom communication topology with 10 DGs and 30 links. The been reached at 2s as shown in Fig. 8(a). Fig. 8(b) shows
node degrees from DG1 to DG10 are 9, 6, 8, 5, 4, 8, 5, 4, 5, that the estimated average voltage consensus (in GU ) has con-
and 6, respectively. The optimal communication topology with verged to the reference value 311V at 2s. During 2.5s to 7s,
10 DGs and 30 communication links proposed in this paper is when DG1 , DG3 , and DG6 quit the original communication
shown in Fig. 7(b), and the node degrees of each DG are all topology successively due to cyber-attacks, ni Qi and Ūi of

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 3782 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO. 5, SEPTEMBER 2022

remaining DGs can still converge to a consistent value. It

should be noted that DG1 , DG3, and DG6 only rely on the
primary controller to adjust voltage after being attacked, and
the corresponding variables ni Qi , Ūi , and Ui are significantly
reduced, as shown in Fig. 8(a)-(c).
However, when DG7 exits the secondary control operation
under the malicious cyber-attack at 7s, the remaining DGs
can no longer constitute a connected communication topol-
ogy. Therefore, the consensus-based secondary voltage control
strategy cannot achieve the expected control goal, which will
have a negative impact on the voltage control of microgrid.
Obviously, the remaining DGs are divided into two parts from
7s, namely DG2 , DG8 , DG9 , DG10 are connected, DG4 and
DG5 are connected. ni Qi of these two groups converge to dif-
ferent values and Ūi converge to the reference value 311V
separately according to Fig. 8(a)-(b). Fig. 8(c) shows that the
output voltage of each DG deviates significantly from 311V
after the failure of consensus-based secondary voltage con-
trol, and the output voltage difference between DG5 and DG2
is about 10V. Figure 8(d) shows the average output voltage
of DGs during operation. This value finally stables at about
309.6V, and the voltage performance of the entire system
decreases.
Scenario 2: Similar to scenario 1, DGs with higher node-
degree in the optimized communication topology are the first
to be attacked. As shown in Fig. 7(b), DG1 , DG6 , DG9 , DG5 ,
DG4 and DG2 quit the original communication network at
2.5s, 4s, 5.5s, 7s, 8.5s and 10s, sequentially. Additionally, an
external load (P = 20kw, Q = 40kvar) is connected to the
system at 12s. Within 2-10s, although half of the information
nodes quit the communication topology due to the multiple
Fig. 9. Dynamic performance of the tested microgrid in scenario 2.
deliberate cyber-attack, the remaining DGs can still construct
a connected topology. The performance of the tested MG can
be seen in Fig. 9. The proportional reactive power sharing
and estimated average voltage consensus are reached during
2-10s in Fig. 9(a) and Fig. 9(b). However, the remaining
nodes are not connected anymore after the communication
node of DG2 is attacked at 10s. Among them, DG3 and DG7
are connected to DG8 while DG10 is isolated. As shown in
Fig. 9(a), the proportional reactive power sharing among the
remaining DGs fails after 10s. It is worth noting that since
the secondary controller of the attacked DG is still in oper-
ation, the estimated average voltage of remaining DGs can
finally stabilize at 311V in Fig.9(b), which does not mean
the consistency convergence of voltage. The output voltage Fig. 10. Pareto frontiers under different optimal connections obtained in
first-stage optimization.
of each DG in Fig. 9(c) deviates from the reference voltage
311V significantly after 12s. In addition, the average output
voltage of DGs finally stabilizes around 309V as shown in
Fig. 9(d). B. Analysis of Link Weight Optimization Results
From the simulation results of scenario 1, it is known 1) Pareto Frontier Analysis: Based on the connection
that the consensus-based secondary voltage control of of communication network obtained from the first-stage
the tested MG is invalid after 4 DGs are deliberately optimization, Pareto frontiers of second-stage optimization
attacked in the random communication network. However, with different numbers of edges are presented in the Fig. 10.
the optimized communication topology proposed in Specifically, the topological structure corresponding to dif-
this paper can tolerate the 6 attacked DGs. Therefore, ferent edge-numbers in Fig. 10 are depicted in Fig. 6. To
the optimized communication topology has stronger compromise the convergence speed and the communication
structural survivability against the multiple deliberate delay robustness of the consensus-based secondary voltage
cyber-attack. control strategy as well as select the optimal performance

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3783

Fig. 11. Evaluation index results of Pareto frontiers.

TABLE II
S IMULATION C ONDITIONS IN D IFFERENT RUNNING P ERIODS

point, this paper adopts the normalization method for the

optimization results of 35 edges. The horizontal and verti-
cal coordinates of each point are respectively normalized, and
then their values are multiplied as a performance evaluation
index. The index values of all the points on the Pareto frontier
are shown in Fig. 11. Among them, the calculated index value Fig. 12. Dynamic performance of the tested microgrid in scenario 3.
of point A in the subgraph of Fig. 10 is the largest, so the
performance of the corresponding communication network is
the best. average voltage consensus can be reached in Fig. 13(a). From
2) Dynamic Performance Analysis: In order to demonstrate Fig. 13(b) and Fig. 13(c), it can be seen that the maximum
the effectiveness of the optimization results, some simulation communication time-delay that the tested MG can tolerate
experiments are carried out based on the optimal commu- in the communication network corresponding to point A is
nication network, which corresponds to point A, B, and C, greater than that in point B. Besides, the maximum communi-
respectively in Fig. 11. In the following simulations, the cation time delay of the consensus-based secondary control
parameters of secondary controllers satisfy CQi = 40 and in MGs can be calculated according to the formula (12),
CEi = 15. and the time-delay margin of entire system (including the
Scenario 3: Different simulation conditions are set in dif- physical-layer and cyber-layer) can be calculated according to
ferent running periods of the tested MG, which is presented as the small-signal model of MGs [32]. Above simulation results
Table II. At 9s and 13s, an external load with p = 42kw and prove that the robustness to communication delay of the tested
Q = 24kvar is accessed to the system successively. Different MG increases as the proportion of H 2 in the optimization
optimal communication networks are applied to the tested MG objective function increases.
in different time periods.
The dynamic performance of reactive power, estimated aver- C. Computational Performance Analysis
age voltage, and output voltage are depicted in Fig. 12. The For the network with N DGs, the number of links varies
reactive power sharing and estimated average voltage con- from N − 1 to N(N − 1)/2, and thus (N 2 − 3N + 4)/2 models
sensus are achieved with 4s, 2s, and 1s, respectively. It is are solved, including both the IQP model and the possible
obvious that the greater the proportion of H 3 for the second- MISDP feasibility problem. Accordingly, in the second stage,
stage optimization is, the faster the convergence speed of the number of the SDP models solved is 19(N 2 − 3N + 4)/2.
consensus-based secondary voltage control in the tested MG Specifically, there are 45, 190, 435, 780, and 1225 models
will be. solved in turn in the first stage, and 855, 3610, 8265, and
Scenario 4: The robustness to communication time-delay 14820 models are solved in turn in the second stage. The box
of the consensus-based secondary voltage control with differ- plot in Fig. 14 gives the statistical results of the solver time.
ent optimal communication networks (corresponding to Pareto In the first stage, the average solver time is less than 1s, and
points B, A and C) is verified. When the communication the solver time increase as the number of agents increases.
delay is 180ms, both the reactive power sharing and estimated Although there are some prominent outliers, the solver time in

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 3784 IEEE TRANSACTIONS ON SMART GRID, VOL. 13, NO. 5, SEPTEMBER 2022

cyber-attack. In the first stage, a mixed-integer semi-define

programming (MISDP) model is proposed to optimize the
communication topology, which is decomposed into an inte-
ger quadratic programming problem and a MISDP feasibility
problem for efficiently solving. In the second stage, consider-
ing the balance of convergence speed and robustness to com-
munication time-delay, a bi-objective SDP model is proposed
to optimize the link weights. Simulations under different sce-
narios have been implemented to verify the effectiveness of
the proposed method.

R EFERENCES
[1] C. Deng, Y. Wang, C. Wen, Y. Xu, and P. Lin, “Distributed resilient
control for energy storage systems in cyber–physical microgrids,” IEEE
Trans. Ind. Informat., vol. 17, no. 2, pp. 1331–1341, Feb. 2021.
[2] M. Liu, C. Zhao, R. Deng, P. Cheng, W. Wang, and J. Chen, “Nonzero-
dynamics stealthy attack and its impacts analysis in DC microgrids,”
in Proc. Amer. Control Conf. (ACC), Aug. 2019, pp. 3922–3927,
doi: 10.23919/ACC.2019.8814631.
[3] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. de Vicuna, and M. Castilla,
“Hierarchical control of droop-controlled AC and DC microgrids–A
general approach toward standardization,” IEEE Trans. Ind. Electron.,
vol. 58, no. 1, pp. 158–172, Jan. 2011.
[4] O. A. Beg, T. T. Johnson, and A. Davoudi, “Detection of false-data
injection attacks in cyber-physical DC microgrids,” IEEE Trans. Ind.
Informat., vol. 13, no. 5, pp. 2693–2703, Oct. 2017.
[5] S. Sahoo, S. Mishra, J. C. Peng, and T. Dragičević, “A stealth cyber-
attack detection strategy for DC microgrids,” IEEE Trans. Power
Fig. 13. Dynamic performance of the tested microgrid in scenario 4. Electron., vol. 34, no. 8, pp. 8162–8174, Aug. 2019.
[6] M. Liu, C. Zhao, Z. Zhang, R. Deng, P. Cheng, and J. Chen,
“Converter-based moving target defense against deception attacks in
DC microgrids,” IEEE Trans. Smart Grid, early access, Nov. 19, 2021,
doi: 10.1109/TSG.2021.3129195.
[7] A. Cecilia, S. Sahoo, T. Dragičević, R. Costa-Castelló, and F. Blaabjerg,
“Detection and mitigation of false data in cooperative DC microgrids
with unknown constant power loads,” IEEE Trans. Power Electron.,
vol. 36, no. 8, pp. 9565–9577, Aug. 2021.
[8] S. Hu, P. Yuan, D. Yue, C. Dou, Z. Cheng, and Y. Zhang, “Attack-
resilient event-triggered controller design of DC microgrids under DoS
attacks,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 67, no. 2,
pp. 699–710, Feb. 2020.
[9] X. K. Liu, C. Wen, Q. Xu, and Y. W. Wang, “Resilient control and anal-
ysis for DC microgrid system under DoS and impulsive FDI attacks,”
IEEE Trans. Smart Grid, vol. 12, no. 5, pp. 3742–3754, Sep. 2021.
Fig. 14. Solver time of the models under different numbers of DGs. [10] M. Mola, N. Meskin, K. Khorasani, and A. Massoud, “Distributed event-
triggered consensus-based control of DC microgrids in presence of DoS
cyber attacks,” IEEE Access, vol. 9, pp. 54009–54021, Apr. 2021.
[11] Y. Nie, Z. Fang, and S. Gao, “Survivability analysis of LEO
the worst case is less than 8s. In the second stage, the average satellite networks based on network utility,” IEEE Access, vol. 7,
solver time is less than 4s, and the solver time also shows pp. 123182–123194, Aug. 2019.
[12] S. Liu, Z. Hu, X. Wang, and L. Wu, “Stochastic stability analysis and
an increasing trend as the number of agents increases. The control of secondary frequency regulation for islanded microgrids under
worst case is less than 7s. Therefore, the results reveal that the random denial of service attacks,” IEEE Trans. Ind. Informat, vol. 15,
proposed two-stage optimization method performs excellently no. 7, pp. 4066–4075, Jul. 2019.
at computational cost, indicating its great potential for practical [13] J. Peng, B. Fan, H. Xu, and W. Liu, “Discrete-time self-triggered control
of DC microgrids with data dropouts and communication delays,” IEEE
applications. Trans. Smart Grid, vol. 11, no. 6, pp. 4626–4636, Nov. 2020.
[14] A. B. Shyam, S. Anand, and S. R. Sahoo, “Effect of communication
delay on consensus-based secondary controllers in DC microgrid,” IEEE
VI. C ONCLUSION Trans. Ind. Electron., vol. 68, no. 4, pp. 3202–3212, Apr. 2021.
[15] M. Raeispour, H. Atrianfar, H. R. Baghaee, and G. B. Gharehpetian,
A complete two-stage communication network optimization “Resilient H∞ consensus-based control of autonomous AC microgrids
scheme for the consensus-based secondary voltage control with uncertain time-delayed communications,” IEEE Trans. Smart Grid,
of microgrid is first proposed in this paper. The attack- vol. 11, no. 5, pp. 3871–3884, Sep. 2020.
[16] H. Yan, X. Zhou, H. Zhang, F. Yang, and Z.-G. Wu, “A novel sliding
resilience and dynamic performance of microgrid can be well mode estimation for microgrid control with communication time delays,”
improved by the appropriate communication network design. IEEE Trans. Smart Grid, vol. 10, no. 2, pp. 1509–1520, Mar. 2019.
A novel optimization indicator is designed based on the devi- [17] N. M. Dehkordi, H. R. Baghaee, N. Sadati, and J. M. Guerrero,
“Distributed noise-resilient secondary voltage and frequency control
ation of node degrees to describe the structural survivability for islanded microgrids,” IEEE Trans. Smart Grid, vol. 10, no. 4,
of communication networks under the multiple deliberate pp. 3780–3790, Jul. 2019.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.
 SHENG et al.: OPTIMAL COMMUNICATION NETWORK DESIGN OF MGs 3785

[18] R. Olfati-Saber and R. M. Murray, “Consensus problems in networks of Wei Gu (Senior Member, IEEE) received the B.S.
agents with switching topology and time-delays,” IEEE Trans. Autom. and Ph.D. degrees in electrical engineering from
Control, vol. 49, no. 9, pp. 1520–1533, Sep. 2004. Southeast University, China, in 2001 and 2006,
[19] G. Lou, W. Gu, J. Wang, W. Sheng, and L. Sun, “Optimal design respectively.
for distributed secondary voltage control in islanded microgrids: From 2009 to 2010, he was a Visiting Scholar with
Communication topology and controller,” IEEE Trans. Power Syst., the Department of Electrical Engineering, Arizona
vol. 34, no. 2, pp. 968–981, Mar. 2019. State University. He is currently a Professor with
[20] Y. Xu and H. Sun, “Distributed finite-time convergence control of an the School of Electrical Engineering, Southeast
islanded low-voltage AC microgrid,” IEEE Trans. Power Syst., vol. 33, University. He is the Director of the Institute
no. 3, pp. 2339–2348, May 2018. of Distributed Generations and Active Distribution
[21] J. Choi, S. I. Habibi, and A. Bidram, “Distributed finite-time event- Networks. His research interests include distributed
triggered frequency and voltage control of AC microgrids,” IEEE Trans. generations and microgrids and integrated energy systems. He is an Editor
Power Syst., vol. 37, no. 3, pp. 1979–1994, May 2022. for the IEEE T RANSACTIONS ON P OWER S YSTEMS, the IET Energy Systems
[22] X. Yu, C. Cecati, T. Dillon, and M. Simões, “The new frontier of smart Integration, and the Automation of Electric Power Systems (China).
grids,” IEEE Ind. Electron. Mag., vol. 5, no. 3, pp. 49–63, Sep. 2011.
[23] F. Guo, C. Wen, J. Mao, and Y. Song, “Distributed secondary volt-
age and frequency restoration control of droop-controlled inverter-based
microgrids,” IEEE Trans. Ind. Electron., vol. 62, no. 7, pp. 4355–4364,
Jul. 2015.
[24] H. Zhang, F. L. Lewis, and A. Das, “Optimal design for synchronization
of cooperative systems: State feedback, observer and output feedback,”
IEEE Trans. Autom. Control, vol. 56, no. 8, pp. 1948–1952, Aug. 2011.
[25] E. Bou-Harb, C. Fachkha, M. Pourzandi, M. Debbabi, and C. Assi,
“Communication security for smart grid distribution networks,” IEEE Shuai Lu (Member, IEEE) received the B.S. degree
Commun. Mag, vol. 51, no. 1, pp. 42–49, Jan. 2013. in smart grid information engineering from the
[26] S. Pan, T. Morris, and U. Adhikari, “Developing a hybrid intrusion detec- Nanjing University of Science and Technology,
tion system using data mining for power systems,” IEEE Trans. Smart Nanjing, China, in 2016, and the Ph.D. degree
Grid, vol. 6, no. 6, pp. 3104–3113, Nov. 2015. in electrical engineering from Southeast University,
[27] A. S. Musleh, G. Chen, and Z. Y. Dong, “A survey on the detection Nanjing, in 2021.
algorithms for false data injection attacks in smart grids,” IEEE Trans. From 2018 to 2019, he was a Visiting Scholar
Smart Grid, vol. 11, no. 3, pp. 2218–2234, May 2020. with the University of New South Wales, Sydney,
[28] L. M. Surhone, M. T. Tennoe, S. F. Henssonow, Spectral Graph Theory. Australia. His research interests include multi-
Providence, RI, USA: Amer. Math. Soc., 2010. energy systems, operations research, and data-driven
[29] S. Boyd, “Convex optimization of graph Laplacian eigenvalues,” in Proc. modeling and optimization. He is currently a Lecture
Int. Congr. Math., vol. 3, Aug. 2006, pp. 1311–1320. with the School of Electrical Engineering, Southeast University. He was
[30] Y. C. Zhao, Y. S. Zhang, and L. Y. Miao, The Degree Sequence selected as an Outstanding Reviewer for IEEE T RANSACTIONS ON P OWER
of Connected Graphs and the Number of Lower Degree Vertices of S YSTEMS in 2020.
Connected Plannar Graphs. Washington, DC, USA: College Math.,
2009.
[31] P. Erdős and T. Gallai, “On maximal paths and circuits of graphs,” Acta
Mathematica Academiae Scientiarum Hungaricae, vol. 10, pp. 337–356,
Sep. 1959.
[32] G. Lou, W. Gu, Y. Xu, W. Jin, and X. Du, “Stability robustness for
secondary voltage control in autonomous microgrids with consideration
of communication delays,” IEEE Trans. Power Syst., vol. 33, no. 4,
pp. 4164–4178, Jul. 2018.
Shixing Ding (Graduate Student Member, IEEE)
received the M.S. degree in power engineering from
North China Electric Power University, Beijing,
China, in 2019. He is currently pursuing the
Lina Sheng (Graduate Student Member, IEEE)
Ph.D. degree in cyber science and engineering with
received the M.S. degree in electrical engineering
Southeast University, Nanjing, China.
from Jiangnan University, Wuxi, Jiangsu, China, in
His research interests include modeling, simula-
2019. She is currently pursuing the Ph.D. degree
tion, and optimization of integrated energy systems;
with the School of Cyber Science Engineering,
cyber-physical security for energy systems.
Southeast University, Nanjing, Jiangsu, China.
Her research interests include microgrid sec-
ondary control, distributed generation, and cyber–
physical security of power systems.

Guannan Lou (Member, IEEE) received the B.S.

and M.S. degrees in control science and engineering
from North China Electric Power University, China, Zhigang Ye (Member, IEEE) received the B.S. and Ph.D. degrees from
in 2008 and 2011, respectively, and the Ph.D. degree Xi’an Jiaotong University, Xi’an, China, in 2013 and 2020, respectively.
in electrical engineering from Southeast University, He was also a visiting Ph.D. student with Texas A&M University, College
China, in 2018. Station, TX, USA, from 2016 to 2017, and with Argonne National Laboratory,
From 2011 to 2015, she joined with Guodian Lemont, IL, USA, from 2017 to 2018. He is currently a Postdoctoral Fellow
Nanjing Automation Company Ltd., Nanjing. From with the Electric Power Research Institute of State Grid Jiangsu Electric
2017 to 2018, she was a joint Ph.D. student with Power Company Ltd., and Southeast University, Nanjing, China. His research
Argonne National Laboratory. She is currently an interests include power system reliability and resilience, distribution systems
Associate Professor with the School of Electrical and microgrids, demand side management, and cyber–physical systems. He
Engineering, Southeast University. Her research interests include distributed is the recipient of the “2020 Top 5 Reviewers for the IEEE Transactions on
generations integration, microgrid modeling and control. Smart Grid” Award in 2021.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on February 09,2025 at 14:20:34 UTC from IEEE Xplore. Restrictions apply.