Professional Documents
Culture Documents
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
Abstract—This paper addresses the problem of distributed storage systems (ESSs) and loads, and it can be operated in a
resilient finite-time control of multiple heterogeneous battery grid-connected or islanded mode. In a micrgrid, introduction
energy storage systems (BESSs) in a microgrid subject to denial- of ESSs can improve power quality and network reliability
of-service (DoS) attacks. Note that DoS attacks may block
information transmission among BESSs by preventing the BESS by providing a balance between intermittent generations (for
from sending data, compromising the devices and jamming a example, wind and Photovolatic) and varying loads.
communication network. A distributed secure control framework For a microgrid, fundamental control issues to be addressed
is presented, where an ACK-based attack detection strategy are to ensure power quality and frequency/voltage stability,
and a communication recovery mechanism are introduced to and so on. With the integration of battery energy storage
mitigate the impact of DoS attacks by repairing the paralyzed
topology graphs caused by DoS attacks back into the initial systems (BESSs) into a microgrid, energy levels or the state
connected graph. Under this framework, a distributed resilient of charge (SoC) levels of BESSs should be coordinated
finite-time secondary control scheme is proposed such that properly to guarantee that the BESSs will not prematurely
frequency regulation, active power sharing and energy level run out of energy or be overloaded. In [3], the estimation
balancing of BESSs can be achieved simultaneously in a finite of SoC levels was carried out effectively by a Lebesgue-
time, meanwhile, operational constraints can be satisfied at any
control transient time. Moreover, based on theoretical analysis, sampling-based extended Kalman filter. Besides, traditionally,
the impact of the duration time of DoS attacks on the convergence a hierarchical control structure microgrids with BESSs is
time of the control algorithm can be explicitly revealed. Finally, utilized widely with three levels: a primary control level, a
validity and effectiveness of the proposed control scheme are secondary control level and a tertiary control level [4]. The
demonstrated by case studies on a modified IEEE 57-bus testing primary control level is to achieve power/load sharing and
system.
maintain frequency/voltage stability. In the primary control
Index Terms—Distributed resilient finite-time control, denial- level, one commonly used method is decentralized droop
of-service attack, active power sharing, frequency regulation,
control, which is based only on local information of BESSs
energy level balancing, battery energy storage system.
without requiring any communication [5]–[9]. Note that the
standard f − P and V − Q droop control is not suitable
I. I
for BESSs since it does not consider SoC levels of BESSs
Due to widespread penetration of renewable energy sources at all. Hence, SoC based droop control was used in [10]
such as wind generation and photovolatic panels, and rapid to adjust the nominal frequency/voltage of microgrids. It is
advances in communication and computation technology, a commonly acknowledged that the droop control introduces
traditional power network has been experiencing a huge rev- frequency/voltage deviations from their desired values, thereby
olution towards a smart grid, which can significantly enhance degrading accuracy of power sharing and frequency/voltage
reliability and efficiency of power networks, and alleviate regulation. To compensate for the deviations caused by the
environmental pollution [1], [2]. Meanwhile, a microgrid has primary control, a secondary control level is employed. The
been proposed as a solution of smart grids to overcome tertiary control level is responsible to perform economic and
intermittent characteristics of renewable energy sources. Gen- optimal management of power dispatch.
erally, a microgrid consists of distributed generations, energy In this paper, an issue of secondary control of micgrogrids
This research work was supported in part by the program of Jiangsu with multiple BESSs is addressed. In general, secondary
Specially-Appointed Professor under Grant RK043STP19001, the fund of control strategies for microgrids are categorized into three
high-level talents at NJUPT under Grant XK0430919039, the Australian types: centralized [11], decentralized [12], and distributed
Research Council under Grant DP160103567, the National Key R&D Program
of China under Grant 2018YFA0702202 and the key project of National [13]. A centralized secondary control requires a central station
Natural Science Foundation of China under Grant 61833008. (Corresponding which is responsible to collect all information of units through
author: Qing-Long Han, Dong Yue). communication networks and then to implement algorithm
L. Ding and D. Yue are with the Institute of Advanced Technology, Nanjing
University of Posts and Telecommunications, Nanjing 210023, P. R. China (e- computation. In this case, any single point of failure in the
mails: dl522@163.com (L. Ding); medongy@vip.163.com (D. Yue)). processing of information transmission and computation will
Q.-L. Han and B. Ning are with the School of Software and Electrical seriously cause migrogrid instability. In contrast to the central-
Engineering, Swinburne University of Technology, Melbourne, VIC 3122,
Australia (e-mails: qhan@swin.edu.au (Q.-L. Han); bdning.ecust@gmail.com ized secondary control scheme, a distributed secondary control
(B. Ning). scheme using only local neighborhood information is more
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
III. D S C S U D S
A
B. DoS Attack Modeling
In this section, we design distributed resilient secondary
controllers to achieve frequency regulation, active powering Generally, DoS attacks refer to a class of attacks which
sharing and energy balancing under DoS cyber attack. First, a make certain or all components of a control system inac-
description of DoS cyber attack is presented. Then, theoretical cessible. The adversaries attempt to jam the communication
analysis is conducted to prove the convergence of the proposed channels, compromise the devices, block data sending and
control algorithms. distort the routing protocols by imposing the DoS attacks on
both the system and communication network. As a result,
the occurrence of DoS attacks may seriously affect both
A. Communication Topology
system measurement and data transmission, causing loss of
To facilitate the design of distributed secondary controllers, data and failure of control actions [28]. According to different
suppose that each BESS can communicate with its neighbors attacked targets, DoS attacks on a networked system can be
via a communication network. The communication topology roughly categorized into: 1) node-based DoS attacks, which
is described as an undirected graph G = (V, E), where V = aim to prevent a node from sending its measurement data
{1, 2, · · · , N } is the set of nodes and E ⊆ V × V is the set to any of its neighbors; 2) link-based DoS attacks, which
of edges. Node j is a neighbor of node i if there is an edge aim to remove or damage the communication link between
defined as εij ≜ (i, j) ∈ E. Ni ≜ {j| εij ∈ E} represents the neighboring nodes. Note that, under the node-based attack, the
set of neighbors of node i. A = [aij ] ∈ RN ×N is a weighted connectivity of communication topology must be paralyzed
adjacency matrix, where aii = 0 for all i, and aij = 1 if since some or all of nodes are completely isolated from
εij ∈ E, otherwise, aij = 0. The degree matrix of graph G is the communication network. Unfortunately, it is devastating
D = diag{d1 , d2 , · · · P, dN }, where the diagonal elements are for distributed control algorithms requiring necessary support
represented as ϖi = j∈Ni wij . The Laplacian matrix L of of connected communication topology, probably resulting in
the graph is defined as L = D − A. A path is a connected degrading or even destroying the convergence of control
edge in a graph. The graph G is connected if there exists a algorithms. While, under the node-based attack where some
path between any nodes. The matrix L of a connected graph communication links among nodes are removed, it is possible
is symmetric and has a simple zero eigenvalue and all other that the communication topology is kept either connected
eigenvalues are real and positive [26]. or unconnected. It is clear that the former case only slows
Assumption 1: The communication topology graph G is down the convergence of distributed control algorithms, while
undirected and connected when DoS attacks are absent. the latter case is in fact regarded as a case of the node-
Remark 3: In this paper, for analysis brevity, suppose that based attack. In contrast to the link-based attack, the node-
the transmission of all the state information such as frequen- based attack may probably result in relatively more serious
cies, active power and energy levels of BESSs uses a common degradation or even damage of control performance. Thus,
communication topology graph G. In a practical microgrid, in this paper, we focus more on the case of node-based DoS
however, it is possible that the information is shared through attacks, where each battery is suffering from DoS attacks such
different communication networks (denoted by Gw , Gp and that its measurement is blocked to be transmitted to all of its
Gp , respectively) like [27]. It should be emphasized that, the neighbors through a communication network. As a result, the
control method proposed later in this paper can be extended battery under such DoS attacks gets isolated in communicating
to accommodate the case of different topologies. with others. It should be mentioned that the control framework
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
designed later can also accommodate link-based DoS attacks. making of network repairment. Similarly, with the cyberse-
1) Impacts of DoS attacks: Obviously, the DoS attacks curity framework [29], [30], a reliable and secure channel is
considered in the paper will break communication connection employed to transmit the attack detection report between each
among neighbors, leading to paralysis of graph connectivity battery and the SCMC. When detecting the event of attacks,
of communication topology. To explain this more clearly, batteries will send a report massage to the SCMC. Then, the
Fig. 2 illustrates the impacts of DoS attacks on changes of SCMC will promptly activate operation of network repairment
communication topologies. It is emphasized that, graph G0 is to reinitialize or recover the entire communication topology
the original communication topology among batteries when of batteries. It should be mentioned that there have been
attacks are absent, while some batteries in graphs G1 , G2 many well-studied approaches to reinitialization or recovery of
and G3 are under DoS attacks. For example, in graph G1 , broken communication topology in the existing literature such
the adversary launches a DoS attack on Battery 1 such that as [31]–[33]. The details on these communication recovery
its measurement data cannot be delivered to Battery 2 and mechanisms can be referred to [31]–[33] for interested readers,
Battery 4, or vice versa. In this case, the Laplacian matrix of which are beyond the scope of this paper. In practice, the
communication topology becomes processing of communication network repairement will take a
finite period of time, which closely depends on the numbers
0 0 0 0 0 0 0
0 3 −1 0 −1 −1 0 of attacked nodes and broken communication links.
4) Model of DoS attacks: Based on the observation above,
0 −1 2 0 0 0 −1
the DoS attacks can be modeled by a set of communica-
L= 0 0 0 1 −1 0 0
0 −1 0 −1 2 tion topologies which are swiching between the connectivity-
0 0
0 −1 0 paralyzed graphs and the initial connected graph. To be more
0 0 2 −1
specific, the communication topology without any attacks
0 0 −1 0 0 −1 2
and after repairment is set by G0 with its weights a0ij and
It is noted that the all the elements in both the first row its Laplacian matrix L0 . Here, it is assumed that, the kth
and column of L are zero, which implies that Battery 1 has attacks happening to the microgrid can be detected at the
no any communication with others. According to [26], it is time sequence tak (k = 1, 2, · · · ), and the time when the
theoretically not possible for Battery 1 to achieve cooperative communication topology is recovered from the kth attack is
control. More specially, all the communication links in graph denoted by tk (k = 0, 1, · · · ) with t0 = 0 being the initial time.
G3 are broken although only three batteries are subject to the Thus, it is easy to know that the duration intervals without and
DoS attacks. Therefore, the presence of DoS attacks signifi- with attacks are [tk−1 , tak ) and [tak , tk ), respectively. In fact, the
cantly paralyzes the connectivity of communication topology, attacked time length of [tak , tk ] is the mount of attack detection
leading to failure of control algorithms in terms of frequency time, message transmission time and network recovery time.
regulation, active power sharing and energy balancing. Let δij represent the status of communication links at node
2) Detection of DoS attacks: Due to the negative impact of i, where δij (t) = 0 for any j if node i is attacked during
DoS attacks on control performance, how to efficiently detect t ∈ [tak , tk ), otherwise, δij (t) = 1. Then, the communication
DoS attacks becomes critical in the microgrid, especially when topology under attack can be described as a time-varying graph
δ (t)
attackers can continuously put the jamming on nodes. In Gδ(t) with its weights aijij = δij (t)a0ij and its Laplacian
order to mitigate the impact of DoS attacks, an ACK-based matrix L . δ(t)
attack detection approach is introduced in this paper, which
is simple to implement in practice. More specifically, Battery
i sends its data information at time t to its neighbors j, and C. Distributed Finite-Time Secondary Frequency Controller
then its neighbors that have access to the data information Under DoS Attacks
will immediately send an acknowledgment message back to To simplify presentation, let P̂i (t) = mi Pi (t) and Êi (t) =
Battery i. Here, it is assumed that every ACK-message can be Ei (t)/∆Ei , and we drop t from the states. Then, regarding
successfully transmitted through secure channels with different the effects of DoS attacks, the following distributed secondary
protocols from communication channels among batteries. Such controllers are proposed:
implementation is practical and commonly used in the area of
No attacks i.e. t ∈ [tk−1 , tak )
cybersecurity framework for electrical grids [29], [30]. Thus, X
if Battery i does not receive the ACK-message in a period ab
uE
i (t) = Π βi
−k 1 ∆E i a0
ij Êi (t) − Êj (t) (13a)
[t, t + τ ack ] with τ ack being the maximum time for receiving j∈Ni
the ACK-message, then it implies that Battery i is suffering X
a
from the DoS attacks with time t being time instant of the upi (t) = ProjPi −k2 a0ij P̂i (t) − P̂j (t) b (13b)
kth DoS attack, denoted by taik . When there are multiple DoS
j∈Ni
attacks on different nodes before the network recovery, one X a
uw
i (t) = Projwi −k3 a0ij wi (t) − wj (t) b
has the kth attack instant tak = inf{t|taik , i ∈ N }.
j∈Ni
3) Communication recovery mechanism: To ensure security
ab
and reliability of network communication, a smart communi- − k3 gi wi (t) − w ref
(13c)
cation management center (SCMC) is employed to perform
the monitoring of communication environment and decision- Under attacks i.e. t ∈ [tak , tk )
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
X δ (t)
ab 2b
uE
i (t) = Π βi
−k1 ∆E i a ij
Ê i (t) − Ê j (t) the adjacency weights ã0ij = (a0ij ) a+b and its corresponding
ij
j∈Ni Laplacian matrix be L̃0 . Denote by λi (·) the ith smallest
(14a) eigenvalue of a matrix.
X
δ (t) a Now, we establish the following theorem.
upi (t) = ProjPi −k2 aijij P̂i (t) − P̂j (t) b (14b) Theorem 1: Under Assumption 1, for given scalars kl > 0
ηj
i j∈N
(l = 1, 2, 3) and parameters miη∆E i
= mj ∆E , ∀i, j ∈ V,
X δ (t) a i j
β if x = xmin or x = xmax TE =
b
, (20)
a
a+b
or v > β (discharging) (b − a)k1 2 b λ22b (L̃0 )
Πβ (v(x)) ≜ (15)
or v < β (charging) b∥ew (0)∥
b−a
Tw =
b
(21)
a+b
v otherwise a
(b − a)k3 2 b λ12b (M̃ 0 )
The projection operator Projx is used to constrain the magni-
with M̃ 0 = L̃0 + diag{gi }N .
tude of frequency and active power in control actions
max
Proof: Based on the definition of epi (t) and the con-
0 if v > 0 & x = x trollers (13b) and (14b), one has
Projx (v(x)) ≜ or v < 0 & x = xmin (16)
1 X ˙
N
v otherwise ėpi (t) = P̃˙ (t) − P̃ (t)
N i=1
where xmax and xmin are the upper and lower bounds of x, X
ab
respectively. According to the definition of Projx (·), it is easy
Proj −k a0
e (t) − e (t) for t ∈ [tk−1 , tak )
2 ij pi pj
to obtain its following characteristics j∈Ni
= X δ (t)
ab
(x − x∗ )Projα ∗
x (v(x)) ≤ (x − x )v(x) (17)
Proj −k a ij
e (t) − e (t) for t ∈ [tak , tk )
2 ij pi pj
j∈Ni
for any x∗ ∈ [xmin , xmax ]. It should be pointed out that,
introducing the projection operator Proj(·) in controllers (13) Choose a Lyapunov function as Vp (t) = 12 eTp (t)ep (t) =
PN 2
and (14) is to ensure that the hard constraints (6) can be 1
2 i=1 epi (t), and then taking a time derivative of Vp (t)
satisfied at any transient time. yields:
Remark 4: Different from the control algorithms in [14]– Case I: for t ∈ [tk−1 , tak )
[17], [19], the control algorithms (13) and (14) in this paper
include some new properties: i) the projection operators are X
N
introduced to ensure the constraints (6) can be satisfied at any V̇p (t) = epi (t)ėpi (t)
i=1
control transient time; and ii) the communication topologies
are switching among the initial connected graph and a set of XN X a
= epi (t)Proj −k2 a0ij epi (t) − epj (t) b
connectivity-paralyzed graphs caused by DoS attacks. Conse-
i=1 j∈Ni
quently, they lead to some difficulties in theoretical analysis X
and the estimate on the settling time of the algorithms, which X
N
ab
≤ −k2 epi (t) aij epi (t) − epj (t)
0
will be dealt with in the next subsection.
i=1 j∈Ni
k2 X
N X 2
a+b
2b
D. Stability Analysis =− ã0ij epi (t) − epj (t)
2 i=1 j∈Ni
In order to conduct convergence analysis of controllers
N a+b
(13) and (14), we define the error variables epi (t) = P̃ (t) −
PN k2 X X 0 2 2b
1
[e (t), ep2 (t), · · · , epN (t)]T , ≤− ãij epi (t) − epj (t) (22)
N i=1 P̃ (t) with ep (t) = PN p1 2 i=1
eEi (t) = Êi (t) − N j=1 Êj (t) with eE (t)
1
= j∈Ni
[eE1 (t), eE2 (t), · · · , eEN (t)]T , and ewi (t) = wi (t) − wref where the first inequality is based on the projection character-
with its vector ew (t) = [ew1 (t), ew2 (t), · · · , ewN (t)]T . Let istic (17).
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
0
Since Assumption 1 holds, one knowsPN that P L̃ has a simple In what follows, we will prove the finite-time convergence
eigenvalue of zero. Then, we have i=1 j∈Ni ã0ij epi (t) − of the energy balancing control algorithms (13a) and (14a) for
2
epj (t) = 2eTp (t)L̃0 ep (t) ≥ 2λ2 (L̃0 )eTp (t)ep (t). Thus, from BESSs. By solving (5a), one can obtain
(22), one can obtain Z t
ηi
a+b Êi (t) = Ēi (t) − P̂i (s)ds (30)
k2 2b 3600mi ∆Ei 0
V̇p (t) ≤ − 0 T
2λ2 (L̃ )ep (t)ep (t)
2 ˙
where Ēi (t) is the solution of Êi (t) = uE i /∆Ei with the
a+b
= −c̃1 Vp 2b
(t) (23) control algorithms PN (13a) and (14a).
a+b
Let VE = 12 i=1 e2Ei (t). Recalling the proof of active
a
where c̃1 = k2 2 b λ22b (L̃0 ). Based on (23), we have power sharing above, it is known that Ēi (t) = Ēj (t) =
PN ′
i=1 Ê(0), ∀i, j ∈ V for t ≥ tE = TE + Ta . Given
1
b−a ′ b − a − a+b N
Vp 2b (t) = Vp 2b (t)V̇p (t) ≤ −c1 (24) parameters miη∆E i ηj
= mj ∆E = κ, ∀i, j ∈ V, it is concluded
2b i j
∗ ∗ ′
c̃1 (b−a) that, for t ≥ tE = max{tp , tE }
where c1 = . Integrating
both sides of (24) on the inter-
2b
Rt ′ Rt PN
1 X
b−a N
val t ∈ a
Vp 2b (t) dt ≤ − tk−1 c1 dt,
[tk−1 , tk ) leads to tk−1 κ i=1 P̂ (0)
Êi (t) = Ê(0) − · t, ∀i ∈ V (31)
which implies that, for t ∈ [tk−1 , tak ), N i=1 N 3600
b−a b−a
Vp 2b (t) ≤ Vp 2b (tk−1 ) − c1 (t − tk−1 ) (25) which means that the energy level balancing can be achieved
within a prescribed time t∗E .
Case II: for t ∈ [tak , tk ) Next, we will prove that the frequencies of BESSs can
Let L̃δ(t) denote the Laplacian matrix with its adjacency be regulated to the reference value in a P finite time. Define
weights ã0ij δij . Since at least one battery is attacked at the N
a Lyapunov function candidate Vw = 21 i=1 e2wi . Follow-
time tak , the matrix L̃δ(t) has more than two eigenvalues of 0, ing the similar lines to the proofs above, one obtains that
which implies that λ2 (L̃δ(t) ) = 0. wi = wref , ∀i ∈ V for t ≥ t∗w = Tw + Ta , which means
Following the similar lines of (22) and (23), one can get that frequency regulation of BESSs can be achieved within a
a+b prescribed time t∗w .
k2 2b
V̇p (t) ≤ − δ(t) T
2λ2 (L̃ )ep (t)ep (t) =0 (26) Finally, according to Theorem 4.2 in [34], it is straightfor-
2 ward to conclude that the projection operator can ensure that
which implies that, all the trajectories of Pi (t), Ei (t) and wi (t) stay within the
b−a b−a hard constraints (6) if their initial conditions satisfy (6).
Vp 2b (t) ≤ Vp 2b (tak ) for t ∈ [tak , tk ) (27) From the observation above, it is concluded that the estima-
Based on Case I and Case II, we have: tion of total convergence time is T ∗ = max{t∗p , t∗E , t∗w }, which
1) if t ∈ [tk−1 , tak ), is rewritten as (18). The proof of Theorem 1 is complete.
b−a b−a
Remark 5: It should be pointed out that, Theorem 1 ex-
Vp 2b (t) ≤ Vp 2b (tk−1 ) − c1 (t − tk−1 ) plicitly reveals the relationship between the duration time Ta
b−a of DoS attacks and the convergence time T ∗ of the proposed
≤ Vp 2b (tak−1 ) − c1 (t − tk−1 ) control algorithms (13) and (14). More specifically, given the
b−a
≤ Vp 2b (tk−2 ) − c1 (tak−1 − tk−2 ) − c1 (t − tk−1 ) control gains, system parameters and initial conditions, the
b−a longer duration time of DoS attacks will result in the slower
= Vp 2b (tk−2 ) + c1 (tk−1 − tak−1 ) − c1 (t − tk−2 ) convergence rate. If an expected convergence time Texp ∗
is
≤ ··· preset, then the attack duration time Ta is required to be
∗
b−a less than Texp − max{Tp , TE , Tw }, which is meaningful for
≤ Vp 2b (0) + c1 Ta − c1 t (28) practical implementation.
2) if t ∈ [tak , tk )
b−a b−a IV. C S
Vp 2b (t) ≤ Vp 2b (tak )
b−a
In this section, case studies with a modified IEEE 57-
≤ Vp 2b (tk−1 ) − c1 (tak − tk−1 ) bus testing system are conducted to show effectiveness and
b−a efficiency of the proposed control scheme. As shown in Fig.
= Vp 2b (tk−1 ) + c1 (t − tak ) − c1 (t − tk−1 ) 3, the testing system includes 7 lithiumion BESSs and 42
≤ ··· loads which are connected to the micrigrid. The BESSs have
b−a
heterogeneous capacity, where the capacities of Batteries 1-3,
≤ Vp 2b (0) + c1 Ta − c1 t (29)
Batteries 4-6 and Battery 7 are 200kWh, 150kWh and 250
Obviously, when t ≥ t∗p
= Tp + Ta , Vp (t) will converge kWh, respectively. BESSs are connected to the main grid
to 0, which implies that ep (t) → 0 after t∗p . Therefore, it is through the Point of Common Coupling (PCC), which is used
concluded that the active power sharing can PN be achieved in a to determine the operation of the microgrid.
prescribed time t∗p , namely, P̂i = P̂j = N1 i=1 P̂ (0), ∀i, j ∈ The desired output frequency and voltage of BESSs
V for t ≥ t∗p . are f ref = 50Hz (wref = 2πf ref ≈ 314.15rad/s)
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
Battery 2 Battery 1
Battery 3
2 1
17
Battery 5
3 15 16
4 45 14
5 13
44 46 12
Battery 4
18 49
19
38
20 48
6 26 21 37 47
27 22 39
57
28
23 56
24
29 36
35 40 42
25
7 34 33 50
52 10
30 32
31
11 51
53 43
54
55 41
Fig. 5. Energy level balancing, active power sharing and frequency regulation
8 9
without any attack detection and communication recovery mechanisms
Battery 6 Battery 7
0.6
2.5
0.4
2 0.2
0
1.5 0 1 2 3 4 5 6 7 8 9 10
t(s)
1 50.5
0.5 50
0 49.5
49
-0.5 0 1 2 3 4 5 6 7 8 9 10
t(s)
-1
0 2 4 6 8 10
t(s)
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
Fig. 7. Energy level balancing, active power sharing and frequency Fig. 8. Distributed control scheme without operational constraints
restoration in a grid-connected mode
B. Grid-Connected Mode
In this case, it is assumed that the microgrid is working in
a grid-connected mode, which means that all the BESSs are
charging. Set the initial energy levels of batteries E(0) =
[60, 50, 55, 45, 30, 40, 95]T . Let k1 = 1.25 and all other
parameters be same as the case of Section IV-A. Applying
the control algorithms (13) and (14), we plot the evolution of
energy levels, active power sharing and frequency regulation
in Fig. 7. It is clear to see that the proposed control algorithms
(13) and (14) are effective for both the grid-connected and
islanded modes.
C. Comparison Results
Fig. 9. Comparison of convergence rates using different a/b
In this subsection, we will make a set of comparisons to
show the efficiency of the proposed control schemes.
1) Functions of projection operators: In order to guarantee
to achieve frequency regulation, active power sharing and
the operational constraints (6) in control actions, we introduce
energy level balancing in a finite time. In the control design,
the projection operators Π and Proj in the controllers (13) and
a communication recovery mechanism has been used to repair
(14). Fig. 8 shows the simulation results without projection
the connectivity-paralyzed communication topology so as to
operators. From Fig. 8, it is clear to see that the energy levels
ensure robustness of the proposed control scheme against DoS
of some batteries are increasing in the beginning period time,
attacks. Theoretical analysis has been conducted to verify
which is not reasonable in the discharging mode. Alternatively,
the convergence of the proposed control scheme. Finally,
the energy levels in Fig. 6 are non-increasing in the entire
simulation case studies on a modified IEEE 57-bus testing
control process. In this sense, it is clarified that the projection
system have been performed to demonstrate effectiveness of
operators are effective to ensure the operational constraints (6)
the proposed methods.
are satisfied in any control transient instants.
2) The effects of parameters a/b on convergence rates: In
this case, comparison results of convergence rate using dif-
R
ferent parameters a and b are given. To show the comparison
clearly, we set four cases with a/b = 1, a/b = 3/5, a/b = 5/9 [1] D. E. Olivares, A. Mehrizi-Sani, A. H. Etemadi, C. A. Canizares,
and a/b = 3/7, respectively, and plot the convergence of R. Iravani, M. Kazerani, A. H. Hajimiragha, O. Gomis-Bellmunt,
energy levels in Fig. 9. It is observed from Fig. 9 that, as the M. Saeedifard, R. Palma-Behnke et al., “Trends in microgrid control,”
IEEE Transactions on Smart Grid, vol. 5, no. 4, pp. 1905–1919, 2014.
value of a/b gets smaller, the convergence of BESS systems [2] C. Dou, D. Yue, X. Li, and Y. Xue, “MAS-based management and con-
will become faster. When a/b = 1, the distributed control trol strategies for integrated hybrid energy system,” IEEE Transactions
schemes (13) and (14) reduce to the similar form of [17], on Industrial Informatics, vol. 12, no. 4, pp. 1332–1349, 2016.
[3] W. Yan, B. Zhang, G. Zhao, S. Tang, G. Niu, and X. Wang, “A battery
which implies that the proposed distributed control scheme management system with a lebesgue-sampling-based extended kalman
has a faster convergence rate than that of [17]. filter,” IEEE Transactions on Industrial Electronics, vol. 66, no. 4, pp.
3227–3236, 2019.
V. C [4] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. De Vicuña, and
M. Castilla, “Hierarchical control of droop-controlled ac and dc micro-
A distributed resilient secondary control scheme that takes grids¡ªa general approach toward standardization,” IEEE Transactions
the effects of DoS attacks into account has been presented on Industrial Electronics, vol. 58, no. 1, pp. 158–172, 2011.
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
[5] A. Tuladhar, H. Jin, T. Unger, and K. Mauch, “Control of parallel [26] R. Olfati-Saber and R. M. Murray, “Consensus problems in networks
inverters in distributed AC power systems with consideration of line of agents with switching topology and time-delays,” IEEE Transactions
impedance effect,” IEEE Transactions on Industry Applications, vol. 36, on Automatic Control, vol. 49, no. 9, pp. 1520–1533, 2004.
no. 1, pp. 131–138, 2000. [27] X. Lu, X. Yu, J. Lai, J. M. Guerrero, and H. Zhou, “Distributed
[6] Q.-C. Zhong, “Robust droop controller for accurate proportional load secondary voltage and frequency control for islanded microgrids with
sharing among inverters operated in parallel,” IEEE Transactions on uncertain communication links,” IEEE Transactions on Industrial Infor-
Industrial Electronics, vol. 60, no. 4, pp. 1281–1290, 2013. matics, vol. 13, no. 2, pp. 448–460, 2017.
[7] S. Riverso, F. Sarzo, and G. Ferrari-Trecate, “Plug-and-play voltage and [28] C. Peng, H. Sun, M. Yang, and Y.-L. Wang, “A survey on security
frequency control of islanded microgrids with meshed topology,” IEEE communication and control for smart grids under malicious cyber
Transactions on Smart Grid, vol. 6, no. 3, pp. 1176–1184, 2015. attacks,” IEEE Transactions on Systems, Man, and Cybernetics: Systems,
[8] F. Chen, M. Chen, Q. Li, K. Meng, J. M. Guerrero, and D. Abbott, vol. 49, no. 8, pp. 1554 – 1569, 2019.
“Multiagent-based reactive power sharing and control model for islanded [29] J. Qi, A. Hahn, X. Lu, J. Wang, and C.-C. Liu, “Cybersecurity for
microgrids,” IEEE Transactions on Sustainable Energy, vol. 7, no. 3, distributed energy resources and smart inverters,” IET Cyber-Physical
pp. 1232–1244, 2016. Systems: Theory & Applications, vol. 1, no. 1, pp. 28–39, 2016.
[9] F. Chen, M. Chen, Q. Li, K. Meng, Y. Zheng, J. M. Guerrero, and [30] T. Cruz, L. Rosa, J. Proença, L. Maglaras, M. Aubigny, L. Lev, J. Jiang,
D. Abbott, “Cost-based droop schemes for economic dispatch in islanded and P. Simoes, “A cybersecurity detection framework for supervisory
microgrids,” IEEE Transactions on Smart Grid, vol. 8, no. 1, pp. 63–74, control and data acquisition systems,” IEEE Transactions on Industrial
2017. Informatics, vol. 12, no. 6, pp. 2236–2246, 2016.
[10] X. Lu, K. Sun, J. M. Guerrero, J. C. Vasquez, and L. Huang, “State- [31] A. Majdandzic, B. Podobnik, S. V. Buldyrev, D. Y. Kenett, S. Havlin,
of-charge balance using adaptive droop control for distributed energy and H. E. Stanley, “Spontaneous recovery in dynamical networks,”
storage systems in dc microgrid applications,” IEEE Transactions on Nature Physics, vol. 10, no. 1, p. 34, 2014.
Industrial electronics, vol. 61, no. 6, pp. 2804–2815, 2014. [32] M. A. Di Muro, C. E. La Rocca, H. Stanley, S. Havlin, and L. A.
[11] K. Tan, X. Peng, P. L. So, Y. C. Chu, and M. Chen, “Centralized control Braunstein, “Recovery of interdependent networks,” Scientific reports,
for parallel operation of distributed generation inverters in microgrids,” vol. 6, p. 22834, 2016.
IEEE Transactions on Smart Grid, vol. 3, no. 4, pp. 1977–1987, 2012. [33] Y. Shang, “Localized recovery of complex networks against failure,”
[12] A. Vaccaro, G. Velotto, and A. F. Zobaa, “A decentralized and cooper- Scientific reports, vol. 6, p. 30521, 2016.
ative architecture for optimal voltage regulation in smart grids,” IEEE [34] P. Yi, Y. Hong, and F. Liu, “Initialization-free distributed algorithms for
Transactions on Industrial Electronics, vol. 58, no. 10, pp. 4593–4602, optimal resource allocation with feasibility constraints and application
2011. to economic dispatch of power systems,” Automatica, vol. 74, pp. 259–
[13] L. Ding, Q.-L. Han, and X.-M. Zhang, “Distributed secondary control for 269, 2016.
active power sharing and frequency regulation in islanded microgrids us-
ing an event-triggered communication mechanism,” IEEE Transactions
on Industrial Informatics, vol. 15, no. 7, pp. 3910–3922, 2019.
[14] T. Morstyn, B. Hredzak, and V. G. Agelidis, “Distributed cooperative
control of microgrid storage,” IEEE transactions on power systems,
vol. 30, no. 5, pp. 2780–2789, 2015.
[15] J. Khazaei and Z. Miao, “Consensus control for energy storage systems,”
IEEE Transactions on Smart Grid, vol. 9, no. 4, pp. 3009–3017, 2018.
[16] D. H. Nguyen and J. Khazaei, “Multiagent time-delayed fast consensus
design for distributed battery energy storage systems,” IEEE Transac-
tions on Sustainable Energy, vol. 9, no. 3, pp. 1397–1406, 2018.
[17] J. Khazaei and D. H. Nguyen, “Multi-agent consensus design for
heterogeneous energy storage devices with droop control in smart grids,”
IEEE Transactions on Smart Grid, DOI: 10.1109/TSG.2017.2765241,
2017.
[18] T. Morstyn, B. Hredzak, and V. G. Agelidis, “Cooperative multi-agent
control of heterogeneous storage devices distributed in a DC microgrid,”
IEEE Transactions on Power Systems, vol. 31, no. 4, pp. 2974–2986,
2016.
[19] J. Hu and A. Lanzon, “Distributed finite-time consensus con-
trol for heterogeneous battery energy storage systems in droop-
controlled microgrids,” IEEE Transactions on Smart Grid, DOI:
10.1109/TSG.2018.2868112, 2018.
[20] Z. Zuo, Q.-L. Han, B. Ning, X. Ge, and X.-M. Zhang, “An overview
of recent advances in fixed-time cooperative control of multiagent
systems,” IEEE Transactions on Industrial Informatics, vol. 14, no. 6,
pp. 2322–2334, 2018.
[21] D. Ding, Q.-L. Han, Y. Xiang, X. Ge, and X.-M. Zhang, “A survey
on security control and attack detection for industrial cyber-physical
systems,” Neurocomputing, vol. 275, pp. 1674–1683, 2018. Lei Ding (M’18 ) received the B.Sc. degree in
automation, and the M.Sc. and Ph.D. degrees in
[22] W. Zeng, Y. Zhang, and M.-Y. Chow, “Resilient distributed energy
control theory and control engineering from Dalian
management subject to unexpected misbehaving generation units,” IEEE
Maritime University, Dalian, China, in 2007, 2009
Transactions on Industrial Informatics, vol. 13, no. 1, pp. 208–216,
and 2014, respectively.
2017.
From November 2010 to November 2012, he
[23] C. Peng, J. Li, and M. Fei, “Resilient event-triggering H∞ load
was a visiting Ph.D. student sponsored by China
frequency control for multi-area power systems with energy-limited dos
Scholarship Council at the Centre for Intelligent and
attacks,” IEEE Transactions on Power Systems, vol. 32, no. 5, pp. 4110–
Networked Systems, Central Queensland University,
4118, 2017.
Australia. From January 2015 to January 2016, he
[24] Q. Yu, R. Xiong, C. Lin, W. Shen, and J. Deng, “Lithium-ion battery was a visiting Research Fellow at Western Sydney
parameters and state-of-charge joint estimation based on h-infinity and University, Australia. From May 2016 to January 2017, he was a Post-
unscented kalman filters,” IEEE Transactions on Vehicular Technology, doctoral Associate at Wayne State University, USA. From February 2017
vol. 66, no. 10, pp. 8693–8701, 2017. to March 2019, He worked as a Research Fellow at Swinburne University
[25] G. He, Q. Chen, C. Kang, P. Pinson, and Q. Xia, “Optimal bidding of Technology, Australia. In March 2019, he joined Nanjing University
strategy of battery storage in power markets considering performance- of Posts and Telecommunications, where he is currently a full professor
based regulation and battery cycle life,” IEEE Transactions on Smart with Institute of Advanced Technology. His research interests include event-
Grid, vol. 7, no. 5, pp. 2359–2367, 2016. triggered control and distributed cooperative control of multi-agent systems
and their applications to smart grids.
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2955739, IEEE
Transactions on Industrial Informatics
Qing-Long Han (M’09-SM’13-F’19) received the Boda Ning (S’14-M’17) received the B.Eng. de-
B.Sc. degree in Mathematics from Shandong Normal gree in automatic control from the East China
University, Jinan, China, in 1983, and the M.Sc. and University of Science and Technology, Shanghai,
Ph.D. degrees in Control Engineering and Electrical China, in 2011, the M.Sc. degree with distinc-
Engineering from East China University of Science tion in control systems from the University of
and Technology, Shanghai, China, in 1992 and 1997, Manchester, Manchester, U.K., in 2012, and the
respectively. Ph.D. degree in Electrical and Electronic Engineer-
From September 1997 to December 1998, he ing under Vice-Chancellor’s Research Scholarship
was a Post-doctoral Researcher Fellow with the (2013?2017), from the Swinburne University of
Laboratoire d’Automatique et d’Informatique In- Technology, Melbourne, Australia, in 2017.
dustielle (currently, Laboratoire d’Informatique et Since September 2017, he has been a Research
d’Automatique pour les Systémes), École Supérieure d’Ingénieurs de Poitiers Fellow with the Swinburne University of Technology. From 2010 to 2011, he
(currently, École Nationale Supérieure d’Ingénieurs de Poitiers), Université was an Exchange Student with the University of Dundee, Dundee, U.K. His
de Poitiers, France. From January 1999 to August 2001, he was a Research current research interests include cooperative control of multi-agent systems
Assistant Professor with the Department of Mechanical and Industrial Engi- and its applications to smart grids.
neering at Southern Illinois University at Edwardsville, USA. From September
2001 to December 2014, he was Laureate Professor, an Associate Dean
(Research and Innovation) with the Higher Education Division, and the
Founding Director of the Centre for Intelligent and Networked Systems at
Central Queensland University, Australia. From December 2014 to May 2016,
he was Deputy Dean (Research), with the Griffith Sciences, and a Professor
with the Griffith School of Engineering, Griffith University, Australia. In Dong Yue (SM’08) received the Ph.D. degree in
May 2016, he joined Swinburne University of Technology, Australia, where control science and engineering from the South
he is currently Pro Vice-Chancellor (Research Quality) and a Distinguished China University of Technology, Guangzhou, China,
Professor. His research interests include networked control systems, neural in 1995.
networks, time-delay systems, multi-agent systems and complex dynamical He is currently a Changjiang Professor and the
systems. Dean of the Institute of Advanced Technology, Nan-
Professor Han is a Highly Cited Researcher according to Clarivate Analytics jing University of Posts and Telecommunications.
(formerly Thomson Reuters). He is a Fellow of The Institution of Engineers His research interests include analysis and synthesis
Australia. He is an Associate Editor of a number of international journals, of networked control systems, multiagent systems,
including the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, optimal control of power systems, and the Internet
the IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, the IEEE of Things.
TRANSACTIONS ON CYBERNETICS, and Information Sciences. He is currently an Associate Editor of the IEEE TRANSACTIONS ON
INDUSTRIAL INFORMATICS, the IEEE TRANSACTIONS ON NEURAL
NETWORKS AND LEARNING SYSTEMS, Journal of the Franklin Institute,
International Journal of Systems Science, and the IEEE Control Systems
Society Conference Editorial Board.
1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.