Electric Power Components and Systems

ISSN: 1532-5008 (Print) 1532-5016 (Online) Journal homepage: https://www.tandfonline.com/loi/uemp20

Distributed Control Strategy of Microgrid Based on

the Concept of Cyber Physical System

Bo Zhang, Chun-Xia Dou, Dong Yue, Zhan-Qiang Zhang & Kai Ma

To cite this article: Bo Zhang, Chun-Xia Dou, Dong Yue, Zhan-Qiang Zhang & Kai Ma (2019)
Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System,
Electric Power Components and Systems, 47:1-2, 55-76, DOI: 10.1080/15325008.2018.1564227

To link to this article: https://doi.org/10.1080/15325008.2018.1564227

Published online: 20 Feb 2019.

Submit your article to this journal

Article views: 196

View related articles

View Crossmark data

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uemp20
Electric Power Components and Systems, 47(1-2): 55–76, 2019
Copyright # Taylor & Francis Group, LLC
ISSN: 1532-5008 print / 1532-5016 online
DOI: 10.1080/15325008.2018.1564227

Distributed Control Strategy of Microgrid Based on

the Concept of Cyber Physical System
Bo Zhang,1 Chun-Xia Dou,1,2 Dong Yue,2 Zhan-Qiang Zhang,1 and Kai Ma1
1
Institute of Electrical Engineering, Yanshan University, Qinhuangdao, People’s Republic of China
2
Institute of Advanced Technology, Nanjing University of Posts and Telecommunications, Nanjing, People’s Republic
of China

CONTENTS
Abstract—For distributed energy resources (DERs) in islanded
1. Introduction microgrid (MG), to improve the control effect on the output
voltages and frequency from droop control, a novel distributed
2. Hierarchical Control Structure
control strategy is proposed in this article based on the concept of
3. Design of the Cyber Layer cyber physical system (CPS). The novel control structure is
4. Design of the Physical Layer consisted by the cyber and the physical layers. In the cyber layer:
5. Case Study and Analysis firstly, the communication structures of overall microgrid and
single DER are constructed; then, based on the communication
6. Conclusions structures, the influence of time-delay (delay) and packet loss
References (loss) on consensus control is analyzed in this article; And then,
an event-triggered mechanism combined with adapt virtual leader-
following consensus control (AVLFCC) and predictive
compensation is proposed in this layer. It is used to solve the
problem of time-varying delay and loss rate in the process of
consensus control. In the physical layer: firstly, the droop control
is used as the primary control on the output voltages and
frequency of DERs; then, a novel secondary control method is
proposed by using the event-triggered mechanism. And finally, the
simulation results confirm the effectiveness of the novel
distributed control strategy under communication problems (delay
and loss) and during plug-and-play operation.

1. INTRODUCTION
1.1. Literature Review
The CPS is the combination of communication system and
physical entities. It is the next generation engineering sys-
tem. Its function of real-time sense, dynamic control and
information service are realized by the cooperation of the
3C (control, computation, and communication) technology
Keywords: islanded microgrid, distributed control, cyber physical system, [1]. After the concept of “smart grid” has been put for-
event-triggered mechanism, time-varying delay and loss rate,
secondary control ward, the application of CPS in power grid is more and
Received 28 April 2017; accepted 2 December 2018 more extensive [2] and [3], and it is called as “power
Address correspondence to Chun-Xia Dou, Institute of Electrical Engineering, CPS” [4]. The basic architecture of power CPS was also
Yanshan University, Qinhuangdao 066004, People’s Republic of China.
E-mail: [email protected]; Dong Yue, Institute of Advanced Technology, constructed in [4]. Due to the MG is an important compo-
Nanjing University of Posts and Telecommunications, Nanjing 210023, nent of smart grid, a lot of research has been done on com-
People’s Republic of China. E-mail: [email protected]
Color versions of one or more of the figures in the article can be found
bining the concept of CPS and MG in recent years. This
online at www.tandfonline.com/uemp. concept is best applied to the control strategy of MG.

55
56 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

FIGURE 1. The novel control structure of the ith DER.

For an islanded MG system, the droop control is widely decoupled. But this method was not easy to be imple-
used to control the output voltages and frequency of DERs mented in actual power system. Another droop control
in the system. Due to this control method is derived by method by using the virtual power was proposed in [8], the
using many approximations, so the traditional droop con- output voltage and frequency could be also decoupled by
trol will make errors between the controlled variables and this method. However, there were still errors between the
the reference values. It needs the improved control method controlled variables and the reference values by using the
to keep the output voltages and frequency of DERs stably proposed control methods in [6]–[8]. And the use of the
and improve the control effect. Based on the analysis communication network in MG to improve the control
above, there are some corresponding researches in recent effect on DERs was not considered in [5]–[8].
years: A hybrid control strategy for MG with optimal And recently, with the development of communication
scheduling was proposed in [5]. There were two layers in technology, there are some control strategies proposed to
the strategy: the primary layer was for controlling, and the improve the control effect by using communication data. A
secondary layer was for optimization. But it did not take frequency control strategy based on the average consensus
into account the influence of external disturbances (e.g. algorithm was proposed in [9]. Another control method
variations from load). An adaptive droop control strategy with the consensus algorithm was proposed in [10] to
was proposed in [6]. It could adjust the voltage and fre- improve the control effect on frequency. A novel consensus
quency according to external disturbance. An improved algorithm was used in [11] to accomplish the distributed
droop control method based on virtual impedance and vir- load shedding. An adaptive droop control strategy for DC
tual power source was proposed in [7]. This method made MG based on the discrete consensus algorithm was used in
the output voltage and frequency of the droop control [12]. By using consensus control, the parameters in droop
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 57

FIGURE 2. The communication structure of overall MG.

control were adjusted in [13] to realize the proportional of plug and play can be also realized by using the novel
distribution of load. The secondary control strategy was control strategy.
used in [14] and [15], and the control effect of DERs had
been improved. But the problems of delay and loss in com-
1.2. Assumptions and Contributions
munication were not considered in [9]–[15].
Certainly, there are also some studies proposed to over- In this article, some assumptions are given for the sake of
come the communication problems in recent years. For simplicity: (a) the MG is considered to operate in the
example, Dou et al studied the control strategy of bridge islanded mode; (b) each DER can be regarded as an agent
type DC MG and ordinary AC MG under the consideration in communication network; (c) in the experiments, the
of delay in [16] and [17]. The problem of delay, loss, and renewable energy resources are replaced by direct current
constrained bandwidth had been also studied in [18]. (DC) or alternating current (AC) power sources; (d) when
However, the researches on delay in [16]–[18] are limited. the topology of the MG system is switched, there is
Once the delay was too long, the ideal control effect can- no transition.
not be maintained. Meanwhile, the communication struc- The novel and contribution of this article are as below:
tures of the overall MG and a single DER were not 1. Based on the concept of CPS, the communication
constructed in [9]–[18]. structures of the overall MG and single DER are con-
At present, the distributed control is still the most structed in this article. Meanwhile, a novel distributed con-
studied control strategy [19]. But there are still few studies trol strategy on the output voltages and frequency of DERs
on control strategies to solve the problems in communica- is proposed by using the communication structures. The
tion. And the communication structure in MG is not shown novel control strategy has the characteristic of distributed
in most studies. To improve the control effect on output coordination, and it needs less communication data than
voltage and frequency of each DER with considering the centralized control; 2. An event-triggered mechanism is
problems of delay and loss, a novel distributed control designed to solve the problems of time-varying delay and
strategy is proposed in this article. The control strategy is loss rate. By this method, the conservatism of consensus
also a hierarchical control (it is consisted by cyber and control for delay and loss problems is greatly reduced; 3.
physical layers). The control effect on each DER in the According to the Laplace integral theorem, differential the-
physical layer is improved by using the communication orem and terminal value theorem, an AVLFCC method is
data and consensus control from cyber layer. The capability designed in this article. The parameters in controller can be
58 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

This article is organized as follows: The novel distrib-

uted control structure is shown in Section 2. In Section 3,
the cyber layer is designed in detail. The physical layer is
designed in Section 4. In Section 5, the simulation cases
are provided to confirm the effectiveness of the proposed
method. Conclusions are collected in Section 6.

2. HIERARCHICAL CONTROL STRUCTURE

Based on the concept of CPS, the novel distributed control
structure is as Figure 1. In this figure, the associated parts
are marked with the same color. The meanings of the vari-
ables in Figure 1 are explained in following text. As shown
FIGURE 3. The communication structure of the ith DER. in Figure 1, the cyber layer is consisted of the communica-
tion network, the AVLFCC part, the event-triggered mech-
changed according to the change of delay and loss rate,
anism and the predictive compensation part. The physical
and the control effect can be maintained; 4. A predictive
layer is consisted of the following parts: DER, inverter, LC
compensation part combined extreme learning machine
filter, the Park transformation part, the power calculation
(ELM) and model predictive control (MPC) is proposed in part, the droop control part, the secondary control part, the
this article. And the predictive part has a good dynamic voltage synthesis part, the double closed loop and the
performance. PWM signal generator. The relationship between the cyber
layer and the physical layer is as follows: In the physical
1.3. Approach Overview layer, the output voltage and frequency of each DER will
be transmitted as the communication data to the cyber
Firstly, a novel distributed control strategy for islanded
layer. In the cyber layer, by using the event-trigger mech-
MG is proposed based on the concept of CPS. The control
anism, the communication data will be used to accomplish
structure of the novel control strategy is consisted by the
the secondary control of each DER in physical layer.
cyber and physical layers.
The control process is as follows: Firstly, the
Secondly, the cyber layer is designed in detail. In the
P
x=Q
U droop control is used as the primary control to
cyber layer, the design procedure is as follows: 1) Based
control the output voltage and frequency of each DER.
on the graph theory, the communication data flow model
Then, the output voltage and frequency will be transmitted
of DERs is constructed; 2) The communication structures to the cyber layer by sensor as the communication data.
of overall MG and single DER are also constructed in this After the judge of event-triggered mechanism, the commu-
article. And the bandwidth of communication lines is selected nication data (or predictive data) will be used to accom-
by the data flow model; 3) When the bandwidth of communi- plish the (adaptive) virtual leader–following consensus
cation lines is not reasonable, the problem of delay and loss control. Then again, the secondary control will be accom-
will occur in the system. And the influence of delay and loss plished by adding feedback (the feedback values are
on consensus control is also analyzed. 4) An event-triggered obtained by consensus control) on droop control. Finally,
mechanism combined with AVLFCC and predictive compen- the obtained voltage and frequency will be transmitted to
sation is proposed in this layer. It is used to solve the prob- the voltage synthesis part. And the synthesized voltage will
lems of time-varying delay and loss rate. be transmitted to the next part. After the system works
Thirdly, the physical layer is designed in detail. In the properly, switch 3 and 4 will be closed to keep the output
physical layer, the design procedure is as follows: 1) The voltage and frequency of each DER stable.
P
x=Q
U droop control is used as the primary control on Remark 1: The used switches are controllable logic
the output voltages and frequency of DERs; 2) Based on the switches. Also, changes in load and DER are human-oper-
proposed event-triggered mechanism, a novel secondary con- ated which can be known in advance. Therefore, when
trol method is designed in this layer. external disturbance occurs in the system, the related
Finally, the effectiveness of the novel distributed control logical commands will be sent to the switch 3 and 4. Thus,
strategy is confirmed by simulation results. the two switches will be disconnected. After the consensus
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 59

control is accomplished (i.e. the secondary control is

accomplished), the two switches will be closed to maintain
the controlled variables of DERs as their respective refer-
ence values.

3. DESIGN OF THE CYBER LAYER

3.1. Graph Theory
The topology of multiagent is usually represented by a
directed graph G ¼ ðV ; eÞ [20]. For the directed graph, it is
FIGURE 4. Communication topology.
composed of a set of nodes and a set of edges:
e  V  V ; ðV ¼ f1; 2    ; ngÞ. The each node of the
three parts: communicator, decision maker, and physical
directed graph could be seen as an agent. If the ith node
equipment. Communicator is the node of a communication
connects to the jth node in a directed graph, the ith node
network. Physical equipment is the node of physical layer,
should have a path point to the jth node. If any node to
such as DER or energy storage device. The decision maker
any other node is connected by a path, the directed graph
(including the consensus control part, event-triggered
is “connected strongly.” If there is a directed path connects
mechanism and the predictive compensation part) plays a
all nodes of the graph, this path is also called a “directed
coordinate role between the communicator and the physical
tree.” The adjacency matrix of the directed graph is defined
equipment. Furthermore, the bandwidth of each communi-
as Aa ¼ aij 2 Rnn . If the jth node connects to the ith
cation line should be chosen according to (1). It can be
node, aij is positive and aij ¼ 0 otherwise .
also seen from (1) that delay or loss will lead to f . So, the
reasonable methods are needed to solve these problems.
3.2. Power CPS Modeling The communication structure of the ith DER is as
Based on the research of [15] and [16], each DER in the sys- Figure 3. As shown in this figure, the communicator is consti-
tem can be regarded as an agent. So, the graph theory can be tuted by signal receiver and signal transmitter. It can receive
applied in MG. And the power CPS is an information system, the information from the decision maker and neighbor DER.
so its steady state model can be expressed as the communica- Meanwhile. Furthermore it can also transmit information to
tion data flow model. Based on graph theory, the data flow itself and neighbor DER; the physical equipment can realize
model can be constructed. The process is as below: The MG the energy transfer with neighbor DER; the decision maker is
is considered as a graph G ¼ ðV ; eÞ. In the graph, k
b þ 1 constituted by sensor and decision controller (e.g. event-trig-
represents a collection of nodes. The nodes may be DERs or gered mechanism). It is the key part to realize the coordin-
other communication devices (e.g. communicator, decision ation control between the cyber layer and the physical layer.
maker and sensor); e represents a collection of communica- In the communication structure of single DER, the decision
tion lines in the graph; Cv represents the capability of infor- maker can receive the discrete signal (sampled signal) from
mation exchange (the amount of information that can be sent the physical equipment and the communicator. Meanwhile,
in unit time) for a node. Its units are bit/s or Mbit/s; sinði;jÞ the decision maker can also transmit control signal to the
represents the inputted information from the jth DER to the physical equipment. It is used to ensure the effective oper-
ith DER; soutði;jÞ represents the outputted information from the ation of the overall control system. Based on the structure in
ith DER to the jth DER; Bl represents the bandwidth of the Figure 3, each DER can exchange the information (frequency,
ith communication line. The data flow model is as below: voltage and current et al) with its neighbor DER.
8 P P
>
> sinði;jÞ ¼ soutði;jÞ
>
>
<0P 3.3. Consensus Control
j2Ni sinði;jÞ  Cv
> XX (1)
Definition [21]: Assume there are agents marked as
>
> 0 sinði;jÞ  Bl
>
: agent1; agent2; :::; agentn in a network. The state of the ith
i2l j2Ni
agent is marked as xi ði ¼ 1; 2; :::; nÞ. Meanwhile, there is
Meanwhile, the communication structure of overall MG only one agent which can transfer its state to any other agent
is as shown in Figure 2. The cyber and physical layers are along a directed path in the communication network. The
linked by multiple communication units. Each unit includes directed path is also called as “directed spanning tree.” In
60 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

that case, the only one agent is set as the leader (or “root designed to analyze the influence of delay and loss on the
node”), and the state value of the leader is set as t. The other consensus control. And the controller is designed as the
agents are set as followers. For a multi-agent system, the kind of virtual leader-following consensus controller.
leader–following consensus control is accomplished if there is The delay and packet loss rate in communication are
lim jjxi ðtÞ
xL ðtÞjj ¼ 0 (2) assumed as below:
t!1
i. The delay occurs in the process that the
The first-order mathematical model is used universally in communication data is sent from the ith DER to the
multiagent system, and the equation is ith communicator. But the value is usually very small,
so it can be ignored in this article.
x_ i ¼ ui (3)
ii. The delay also occurs in the process that the data is
where xi is the state value of the ith DER; ui is the con- sent from the jth DER to the ith DER by
trol input. communicator. And it is set as sij . If time-varying
Therefore, the state of each follower will be consistent to delay is considered, the effect of delay on communication
ith by a consensus protocol. And the protocol can be regarded data can be regarded as changing xij ðtÞ to xij ðt
sij ðtÞÞ,
which is essentially the change of data. Furthermore, the
as the way to design controller. Without considering delay
change in time domain can be expressed more specifically
and loss, the consensus protocol of the ith agent is
X   in frequency domain. In the frequency domain, a time-

ui ðtÞ ¼ ki aij xj ðtÞ
xi ðtÞ þ bi ðxL ðtÞ
xi ðtÞÞ (4) delay system can be replaced by a first order inertial
j2Ni system. So if the data is xðt
delayÞ, it can be represented
xðSÞ
as 1þdelayS in the frequency domain, where xðSÞ is the
where xL ðtÞ is the state of the leader; Ni is the collection of Laplace transformation of xðtÞ. Therefore, the data can be
neighbor agents of the ith agent; bi represents the relation- set as kij ðtÞxij ðtÞ in time domain, where kij ðtÞ represents
ship between the leader and the ith agent. If the leader con- the change of data.
nects to the ith agent, bi >0 and bi ¼ 0 otherwise. iii. In the process that the data is sent from the jth DER
Each DER in the MG system can be regarded as an to the ith DER by communicator, the corresponding
agent (follower). Each one can transmit/receive its state packet loss rate is set as 1
lij ( i; j 2 ½1; 2 and i 6¼ j ).
data (voltage and frequency) to/from neighbor DERs. All If the loss rate is time-varying, the effect of loss on
DERs constitute the communication network in the the communication data can be regarded as changing
islanded MG. But in actually, there is not convenient to xij ðtÞ to lij ðtÞxij ðtÞ.
select a leader in the power system. So, the “virtual leader” The consensus protocol with considering time-varying
delay and loss rate is as below:
is proposed in this article to solve the problem. “Virtual
8  
leader” does not actually exist in a MG, its implementation >
> x_ 1 ¼ kx1 ½a12 l12 ðtÞk12 ðtÞx2 ðtÞ
x1 ðtÞ
>
>  
is only to design the xL of consensus controller to a fixed >
< þb1 l1L ðtÞk1L ðtÞxL ðtÞ
x1 ðtÞ 
value (e.g. voltage or frequency reference value). This can   (5)
>
> x_ 2 ¼ kx2 ½a21 l21 ðtÞk21 ðtÞx1 ðtÞ
x2 ðtÞ
be regarded as a separate agent has been constructed, and >
>  
>
:
its state value is xL . As long as the setting of aij guarantees þb2 l ðtÞk2L ðtÞxL ðtÞ
x2 ðtÞ 
2L
the existence of the directed spanning tree and the con-
where k1L ðtÞ represents the change of the data sent from
structed agent acts as the root node, the control effect of
the virtual leader to DER1; k2L ðtÞ represents the change of
consensus control can be guaranteed.
the data sent from the virtual leader to DER2. The Eq. (5)
can be also expressed as
3.4. Consider the Consensus Protocol with the x_ ¼ Ax þ Bu (6)
Problems of Delay and Loss
The coefficient matrices are as follows:
The existence of delay and loss in cyber layer will affect  
the quality of communication [22]. And the control effect
½a12 þ b1 kx1 kx1 a12 l12 ðtÞk12 ðtÞ
A¼ :
in cyber layer will be also affected. The specific analysis is kx2 a21 l21 ðtÞk21 ðtÞ
½a21 þ b2 kx2
as the follows.  
b l ðtÞk1L ðtÞ
B ¼ 1 1L
b2 l2L ðtÞk2L ðtÞ
3.4.1. Analyze the influence of delay and loss on consensus
control. In this article, the case of two DERs (followers) To verify the stability of (6), there is jSI
Aj ¼ 0. The
and one virtual leader (the topology is as Figure 4) is solutions ( S1 ; S2 ) are as (7).
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 61

8
>
>
f½a12 þ b1 kx1 þ ½a21 þ b2 kx2 g
>
> S1 ¼ þ
>
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
>
>
> f½a þ b k þ ½a þ b k g2
4k k ½ða þ b Þða þ b Þ
a a l l k k Þ
>
>
> 12 1 x1 21 2 x2 x1 x2 12 1 21 2 12 21 12 21 12 21
<
2 (7)
>
>
f½a12 þ b1 kx1 þ ½a21 þ b2 kx2 g
> S1 ¼
>

>
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
>
>
>
> f½a12 þ b1 kx1 þ ½a21 þ b2 kx2 g2
4kx1 kx2 ½ða12 þ b1 Þða21 þ b2 Þ
a12 a21 l12 l21 k12 k21 Þ
>
>
:
2

Therefore, S1 requires the negative real part. In theory, D_x 01 ¼ x_ 10
_x 1   

the system can be stabilized by selecting the suitable gains ¼ kx1 a12 1
l12 ðtÞk12 ðtÞ x2 ðtÞ þ b1 1
l1L ðtÞk1L ðtÞ xL ðtÞ
in consensus protocol. However, delay and loss rate are (9)
time-varying in practice, which leads to instability of
where D_x 01
is the equation of error. According to the
the system.
Laplace transform, there is
 
3.4.2. Design of the event-triggered mechanism. In this art- sDX10 ðsÞ ¼ k1 a12 1
l12 ðsÞk12 ðsÞ=4p2 X2 ðsÞ
icle, an AVLFCC method is proposed to solve the prob-  
lems of time-varying delay and loss rate. However, there þb1 1
l1L ðsÞk1L ðsÞ=4p2 XL ðsÞ þ Dx01 ð0Þ (10)
may still be a large overshoot by using this method. According to the Laplace terminal value theorem, the
Therefore, based on the obtained data from the predict- steady-state error is as below:
ive compensation part, another consensus control method 
lim Dx01 ðtÞ ¼ lim sDX10 ðsÞ
is also proposed to solve this problem. But there must be t!1 s!0  
errors between the predictive data and the actual data. ¼ lim kx1 a12 1
l12 ðsÞk12 ðsÞ=4p2 X2 ðsÞ
s!0
So as to make use of the two methods reasonably, an  
event-triggered mechanism is designed in this article. þb1 1
l1L ðsÞk1L ðsÞ=4p2 XL ðsÞ þ Dx01 ð0Þ
The predictive data is used only when trigger conditions  
are met. ¼ kx1 a12 1
l12 ð0Þk12 ð0Þ=4p2 X2 ð0Þ
 
þb1 1
l1L ð0Þk1L ð0Þ=4p2 XL ðsÞ þ Dx01 ð0Þ
Design of the trigger events. The events are chosen as
below: 1. Event1: The steady-state error caused by delay (11)
and loss on consensus control. 2. Event2: The overshoot in
consensus control process. And the steady-state error of Remark 2: In Eq. (11), the initial moment of Laplace
consensus control can be calculated by the following transformation is the moment when the consensus control
method. Still take the system in Figure 4 as the example, begins to be applied in the system. It is not the real 0s. In
and the consensus protocols are as below: other words, “zero” in Eqs. (10) and (11) does not mean
8 the initial moment of the system to run, but the initial
>
> x_ 1 ¼ kx1 ½a12 ðx2 ðtÞ
x1 ðtÞÞ þ b1 ðxL ðtÞ
x1 ðtÞÞ moment of the Laplace transformation.
>
>  
>
> _ 0
¼ ½a ð Þk ð Þx ð Þ
x ð Þ Since the state variation at the initial moment of consensus
>
> x 1 kx1 12 l t 12 t 2 t 1 t
>
> 
12
 control is max, the overshoot at that moment is selected as the
< þb1 l1L ðtÞk1L ðtÞxL ðtÞ
x1 ðtÞ 
(8) maximum overshoot. According to the leader-following consen-
>
> x_ 2 ¼ kx2 ½a21 ðx1 ðtÞ
x2 ðtÞÞ þ b2 ðxL ðtÞ
x2 ðtÞÞ
>
>   sus theory, the corresponding equation for the state x can be
>
>
> x_ 2 0 ¼ kx2 ½a21 l21 ðtÞk21 ðtÞx1 ðtÞ
x2 ðtÞ
> described as x_ ¼ Adaptive virtual leader
following consensus
>
>  
: þb l ðtÞk ðtÞx ðtÞ
x ðtÞ  protocol. And finally, the controlled state value of each DER
2 2L 2L L 2
will be stabilized at xL . Therefore, the overshoot can be
Ð
where x01 and x02 are the obtained state after the effect described as x_ dtjt¼0
xL . More specifically, it is Overshoot ¼
Ð
of delay and loss. Take DER1 as the example, and ðAdaptive virtual leader
following consensus protocolÞ
there is dtjt¼0
xL .
62 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

Design of the trigger conditions. For the ith DER, the

trigger values of steady-state error (event1) and overshoot
(event2) are set as Txi1 and Txi2 respectively. The event-trig-
gered mechanism can be set up by the following conditions:
1. When the steady-state error  Txi1 , the controller of
AVLFCC will be designed. And if the overshoot  Txi2 , the
controller will be used to accomplish the secondary control;
2. Otherwise, the obtained data by predictive compensation
part will be used to replace the actual communication data.
FIGURE 5. The procedure of predictive compensation.
3.4.3. Design of the AVLFCC. In this article, the adaptive
virtual leader-following consensus protocol for the ith DER 8
>
> kx11 a12 l12 ðsÞk12 ðsÞX2 ðsÞ=4p2 þ kx12 b1 l1L ðsÞXL ðsÞ=2p þ x1 ð0Þ
>
< X1 ðsÞ
(follower) is designed as s þ kx11 a12 þ kx12 b1
X h   i >
> kx21 a21 l21 ðsÞk21 ðsÞX1 ðsÞ=4p2 þ kx22 b2 l2L ðsÞXL ðsÞ=2p þ x2 ð0Þ
uxi ðtÞ ¼ kxi1 aij lij ðtÞxj t
sij ðtÞ
xi ðtÞ >
: 2
X ðs Þ
s þ kx21 a21 þ kx22 b2
j2Ni
(15)

þkxi2 bi liL ðtÞxL
xi ðtÞ (12) where x1 ð0Þ and x2 ð0Þ are the initial state values of the
where kxi1 and kxi2 are the gains; xL is a constant value. DER1 and DER2, respectively. Meanwhile, there are:
Furthermore, due to the xL has been set as a constant value, so e1 ðtÞ ¼ x1 ðtÞ
xL ðtÞ
the delay is not considered in the output data of the virtual leader. ð
Theorem: If kxi1 and kxi2 are selected properly, the states ¼ x_ 1 dt
xL ðtÞ
Ð
of DERs can follow the corresponding states of the vir- X1 ðsÞ x1 ðtÞdtjt¼0
L½e1 ðtÞ ¼ þ
XL ðsÞ
tual leader. s  sð 
In the process of this proof, the Laplace integral theorem, lim e1 ðtÞ ¼ lim X1 ðsÞ þ x1 ðtÞdtjt¼0
sXL ðsÞ
t!1 s!0
differential theorem and terminal value theorem are mainly
kx11 a12 l12 ðsÞk12 ðsÞX2 ðsÞ=4p2 þ kx12 b1 l1L ðsÞXL ðsÞ=2p þ x1 ð0Þ
used to prove our theorem. In this article, we prove the situ- ¼ þ c1
s þ kx11 a12 þ kx12 b1
ation that there is only one virtual leader and two DERs (fol-
(16)
lowers) in the network (as shown in Figure 4). However, in
other situations, the only differences are the number of fol- e2 ðtÞ ¼ x2 ðtÞ
xL ðtÞ
ð
lowers, the values of aij and the values of bi . Therefore, the ¼ x_ 2 dt
xL ðtÞ
above Laplace transformation theorems can still be used to Ð
X2 ðsÞ x2 ðtÞdtjt¼0
prove the proposed theorem in other situations. L½e2 ðtÞ ¼ þ
XL ðsÞ
s  sð 
lim e1 ðtÞ ¼ lim X2 ðsÞ þ x2 ðtÞdtjt¼0
sXL ðsÞ
Proof: Assume the state equation is a first order equation, t!1 s!0

and there are: kx21 a21 l21 ðsÞk21 ðsÞX1 ðsÞ=4p2 þ kx22 b2 l2L ðsÞXL ðsÞ=2p þ x2 ð0Þ
¼ þ c2
8 s þ kx21 a21 þ kx22 b2
>
> x_ 1 ¼ kx11 a12 ½l12 ðtÞx2 ðt
s21 ðtÞÞ
x1 ðtÞ
>
< (17)
þkx12 b1 ½l1L ðtÞxL ðtÞ
x1 ðtÞ
(13) where X1 ðsÞ; X2 ðsÞ, and XL ðsÞ are the state values obtained
>
> x_ ¼ kx21 a21 ½l21 ðtÞx1 ðt
s21 ðÞ
x2 ðtÞ Ð Ð
>
:
2
by Laplace transform; c1 ¼ x1 ðtÞdtjt¼0 ; c2 ¼ x2 ðtÞdtjt¼0 .
þkx22 b2 ½l2L ðtÞxL ðtÞ
x2 ðtÞ
If the gains meet the following condition, the states of
where x1 ðtÞ is the state of the DER1; x2 ðtÞ is the state of DERs can be consistency to the corresponding states of the
the DER2. The errors are as below: virtual leader:
 8
e1 ðtÞ ¼ x1 ðtÞ
xL ðtÞ > kx11 a12 l12 ð0Þk12 ð0ÞX2 ð0Þ=4p2 þ kx12 b1 l1L ð0ÞXL ð0Þ=2p þ x1 ð0Þ
(14) >
> þ c1 ¼ 0
e2 ðtÞ ¼ x2 ðtÞ
xL ðtÞ <
kx11 a12 þ kx12 b1
> kx21 a21 l21 ð0Þk21 ð0ÞX1 ð0Þ=4p þ kx22 b2 l2L ð0ÞXL ð0Þ=2p þ x2 ð0Þ
>
2
where xL is the state of the virtual leader. According to the >
: þ c2 ¼ 0
kx21 a21 þ kx22 b2
Laplace transformation and the approximation principle
(18)
(the delay part can be approximated as a first-order inertial
part). The following equations can be obtained. The proof is accomplished.
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 63

FIGURE 6. The operation process in the physical layer.

FIGURE 7. The experiment model.

3.4.4. Design of the predictive compensation part. In the – 1)T is used to obtain the subsequent data. The specific
case of delay or loss occurs in the system, the communica- design process is as the follows.
tion data will be changed. This will further affect the con-
trol effect of consensus control. Therefore, if the existence Design the MPC of the ith DER. Based on the research
of delay and loss results in steady-state error or overshoot of [23] and the structure of Figure 1, the MPC can be used
exceeding the trigger values, the actual communication in MG. For the ith DER, the procedure of constructing the
data will be discarded. Meanwhile, the predictive data MG mathematical model is as below:
obtained by ELM and MPC will be used to replace the dis-
carded data, and thus ensure the consensus control can be 1. The control system in Figure 1 is needed to be discrete,
realized. The predictive process is as shown in Figure 5. and the sampling time is set as T. The output voltage
The operation process is as follows: when the event-trig- and frequency of droop control at KT are set as Ui ðKÞ
gered mechanism is triggered at t ¼ KT, the data of t ¼ (K and xi ðKÞ respectively.
64 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

(  
2. Ui ðKÞ and xi ðKÞ will be transferred to the voltage Pi ðK þ 1Þ ¼ f10 Uodi ðK þ 1Þ; Uoqi ðK þ 1Þ; Iodi ðK þ 1Þ; Ioqi ðK þ 1Þ 
synthesis part. This process can be represented as Qi ðK þ 1Þ ¼ f11 Uodi ðK þ 1Þ; Uoqi ðK þ 1Þ; Iodi ðK þ 1Þ; Ioqi ðK þ 1Þ
 
Ui /hi ðK Þ ¼ f1 Ui ðK Þ; xi ðK Þ (19) (22)

where Ui /hi ðKÞ is the obtained voltage by the voltage 6. Pi ðK þ 1Þ and Qi ðK þ 1Þ will be transferred to the
synthesis part. droop control part. Ui ðK þ 1Þ and xi ðK þ 1Þ can be
3. Ui /hi ðKÞ will be transferred to the Park transformation also obtained by this part. The process can be
part. Udrefi ðKÞ and Uqrefi ðKÞ can be also obtained by expressed as

this part. According to the structure of voltage and Ui ðK þ 1Þ ¼ f12ðPi ðK þ 1ÞÞ
current double closed loop in [7], Udrefi ðKÞ and (23)
xi ðK þ 1Þ ¼ f13 Qi ðK þ 1Þ
Uqrefi ðKÞ are also the inputs of voltage loop. After
inputting Udrefi ðKÞ and Uqrefi ðKÞ in the voltage loop, In summary, the output of droop control part at t ¼
Idrefi ðKÞ and Iqrefi ðKÞ can be obtained. Meanwhile, the (K þ 1)T can be obtained from the output data of droop
two variables are the inputs of current loop. The control at t ¼ KT. The initial parameters for MPC includ-
process can be also expressed as ing ZInverteri , ZLinei , Li , Ci and UDCi , et al. In actually, the
8
> Udrefi ðK Þ ¼ f2 ðUi /hi ðK ÞÞ running time in the MG is continuous. Therefore, it needs
>
>
>
< U ðK Þ ¼ f ðU /h ðK ÞÞ to add the D-A converter on the output of MPC.
qrefi 3 i i
  In summary, when the data at t ¼ t0 is known, the data
>
> I ð K Þ ¼ f U ð K Þ; Uodi ðK Þ; Uoqi ðK Þ; Iodi ðK Þ
> drefi 4 drefi
: I ðK Þ ¼ f  U ðK Þ; U ðK Þ; U ðK Þ; I ðK Þ
> at any time after t0 can be obtained by recursion from (19)
qrefi 5 qrefi oqi odi oqi to (23).
(20)
Design the ELM. ELM is a fast training method for sin-
4. Idrefi and Iqrefi will be transferred to the current loop.
gle hidden layer feed-forward neural network. The initial
Meanwhile, the PWM trigger signal will be obtained
parameters for ELM including historical data, operation
by this part at first. Then, the trigger signal will be
used to create the PWM signal. And then, the PWM time, number of neurons and initial weight, et al. This
signal will be used to control the inverter. The method is operated as the follows.
U0i ðk þ 1Þ (including three-phase, thus A set of training samples is set as
U0ai ðk þ 1Þ; U0bi ðk þ 1ÞandU0ci ðk þ 1Þ ) will be also TELM ¼ fðxi ; yi Þ; i ¼ 1; L; Lg, and xi 2 Rn ; yi 2 Rn . The
obtained. And finally, through the LC filter, Uoi ðK þ 1Þ ELM model with N hidden layer nodes can be expressed
(including Uoai ðk þ 1Þ; Uobi ðk þ 1ÞandUoci ðk þ 1Þ ) and as below [24]:
Ioi ðK þ 1Þ (including U0ai ðk þ 1Þ, U0bi ðk þ 1Þ, and
X
N
U0ci ðk þ 1Þ ) will be obtained. The process can be also bi gðai xi þ ci Þ ¼ Hb (24)
expressed as i¼1
8  
>
> PWMi ðK Þ ¼ f6 Idrefi ðK Þ; Iqrefi ðK Þ; Ildi ðK Þ; Ilqi ðK Þ where bi is the weight of the ith hidden layer node to the
>
>
< U ðK þ 1Þ ¼ f PWM ðK Þ; U ; Z
>  output layer; ai is the weight for the input layer to the ith
0i 7 i DCi Inverteri
hidden layer node; ci is the degree of bias of the ith hidden
>
> Uoi ðK þ 1Þ ¼ f8 ðU0i ðK þ 1Þ; Li ; Ci ; ZLinei Þ
>
> layer node; H is the output matrix of the hidden layer; b is
>
:
Ioi ðK þ 1Þ ¼ f9 ðU0i ðK þ 1Þ; Li ; Ci ; ZLinei Þ the matrix with the weights of the hidden layer to the out-
(21) put layer. According to Eq. (24), the weights of the hidden
layer to the output layer can be obtained by the least
where Ildi and Ilqi are obtained by Park transformation
square solution of the following linear equation:
on the inductive current Ili (including Ilai ðk þ 1Þ,
Ilbi ðk þ 1Þ, and Ilci ðk þ 1Þ ); UDCi is the voltage of the min kHb
Y k (25)
b
ith DER; Zinverteri is the impedance of the ith inverter;
And the solution is b ^ ¼H þ Y , H þ is the Moore-Penrose
ZLinei is the line impedance.
5. The Uodi ðK þ 1Þ, Uoqi ðK þ 1Þ, Iodi ðK þ 1Þ and generalized inverse matrix of H. According to the method
Ioqi ðK þ 1Þ will be obtained by the Park transformation of ELM, the mathematical model of the MG can be built
on Uoi ðK þ 1Þ and Ioi ðK þ 1Þ. And then, the four quickly by historical data. The input data of ELM is time,
things will be transferred to the PQ calculation part. while the target data is the communication data (i.e. the
Meanwhile, Pi ðK þ 1Þ and Qi ðK þ 1Þ can be obtained. output of each DER by droop control). When the weight of
The process can be expressed as the input layer to the hidden layer is determined, the
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 65

Parameter Value Parameter Value

Power V1 ¼ V2 ¼ V3 ¼ V4 ¼ V5 ¼ V6 ¼ V7 ¼ 700V; Line ZLine:DER1
AC
BUS ¼ 0:612 þ j3:18e
3

f1 ¼ f2 ¼ f3 ¼ f4 ¼ f5 ¼ f6 ¼ f7 ¼ 50HZ impedance ZLine:DER2
AC
BUS ¼ 0:613 þ j2:99e
3
Sine filter L ¼ 3  10
3 H ZLine:DER3
AC
BUS ¼ 0:614 þ j3:18e
3
C ¼ 0:9  10
6 F ZLine:DER4
AC
BUS ¼ 0:615 þ j3:18e
3
Sampling time T ¼ 0:5us ZLine:DER5
DER6 ¼ 0:612 þ j3:18e
3
ZLine:DER6
DER7 ¼ 0:613 þ j2:99e
3
ZLine:DER7
DER5 ¼ 0:614 þ j3:18e
3
Load1/Load2/ P ¼ 600KWQ ¼ 60KVar= reference Uref ¼ 311V; fref ¼ 50HZ;
Local load P ¼ 200KWQ ¼ 20KVar= value xref ¼ 314rad=s
P ¼ 30KWQ ¼ 3KVar Droop m ¼ 2:5e
5
coefficient n ¼ 1:04e
3
TABLE l. The experiment parameters in Section 5.

weight of the hidden layer to the output layer can be 3. Obtain the predictive data
obtained according to (24) and (25), thus the training of The final predictive data at t ¼ KT is
ELM is completed. ^y ðKÞ ¼ cELM ðKÞyELM ðKÞ þ cMPC ðKÞyMPC ðKÞ; ðy ¼ U=y ¼ xÞ

Weight processing. In this article, the weight processing 4. Retrain ELMs

on ELMs and MPC is divided into four steps. The specific If there is large error between the predictive data and
procedure is as below: the actual historical data, the ELMs will be retrained
1. Weight processing on ELMs and then return to 1) until the requirement on error
There are m ELMs used to predict, and the weight is met.
processing is carried out according to the predictive
Remark 3: Weight processing always exists during oper-
accuracy of each ELM. To make the predictive model
ation, and the historical data includes the data in various
with the small predictive error have a larger weight, the
following method is used to design the weight of each operating conditions of the system.
ELM.
ei ðK Þ ¼ yðK Þ
yELMi ðK Þ; cELMi ðK Þ 4. DESIGN OF THE PHYSICAL LAYER
" #
Xm 4.1. Construct the Structure of Physical Layer
¼ 1
jei ðK Þj= jei ðK Þj =ðm
1Þ (26) The basic components of physical layer have been pro-
i¼1
posed in Figure 1. In this subsection, the Figure 6 is used
where ci ðKÞ is the weight of the ith ELM at t¼KT; to illustratethe implementation process of distributed con-
yðKÞ is the historical data at t¼KT; yELMi ðKÞ is the trol. In other words, it intends to explain how the compo-
output data of the ith ELM at t¼KT. The output data nents of the cyber layer interact with the related
of ELMs after weight processing is as below: components of the physical layer, and then accomplish the
X
m distributed control together.
yELM ðK Þ ¼ ci ðK ÞyELMi ðK Þ (27) For the ith DER, the P
x=Q
U droop control equation
i¼1
is
2. Weighted ELMs are then weighted processing with 
xi ¼ xref
mi Pi
MPC (29)
8 Ui ¼ Uref
ni Qi
>
> yðK Þ
yELM ðK Þ
>
< cELM ðK Þ ¼ where mi and ni are the droop coefficient; xref and Uref
2yðK Þ
yELM ðK Þ
yMPC ðK Þ are the frequency and the voltage reference values, respect-
(28)
>
>c yðK Þ
yMPC ðK Þ
> ð Þ
: MPC K ¼ ð Þ ively. If the secondary control is not added in the MG, the
2y K
yELM ðK Þ
yMPC ðK Þ
equation of the voltage synthesis part is

where cELM ðKÞ is the weight of ELMs at t¼KT;  Ð

cMPC ðKÞ is the weight of MPC at t¼KT; yMPC ðKÞ is hi ¼ xi dt
(30)
the output data of MPC at t¼KT. Ui /hi ¼ Ui &hi
66 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

8
>
> 1: If lim DUi0 <TUi1 and the overshoot of AVLFCC<TUi2
>
> t!1
X h   i 
where Ui /hi is the obtained voltage by this part. The pro- >
>
>
> u ¼ kUi1 aij lij ðtÞUj t
sij ðtÞ
Ui ðtÞ þ kUi2 bi liL ðtÞUL
Ui ðtÞ
>
>
Ui
cess of designing the doubled closed loop and virtual con- >
< j2Ni
2: lim DUi0
TUi1 or the overshoot of AVLFCC
TUi2
troller is as same as [7]. And the virtual controller is used >
>
t!1
> ð1Þ Use the predictive data
>
to decouple the voltage and frequency of each DER. >
> "  #
> ð2 Þ u ¼ k X a U
>
>

^ i ðtÞ þ bi UL
U
^ j ðtÞ
U ^ i ðt Þ
>
> Ui Ui ij
: j2Ni

4.2. Design the Novel Secondary Control (34)

The control process of secondary control is as below: where U ^ j are the predictive voltage values of the
^ i and U
Firstly, the P
x=Q
U droop control is used as the pri- ith DER and the jth DER respectively; KUi , kUi1 and kUi2
mary control on the output voltage and frequency of each are gains. When selecting protocol, it should judge at first:
DER; then, based on the event-triggered mechanism, the When the influence of delay and loss on consensus control
consensus protocol of voltage and frequency will be is not up to the trigger conditions, AVLFCC is selected to
designed; afterward, the feedback values on voltage and accomplish the secondary control; otherwise, the VLFCC
frequency will be obtained by the consensus control; will be accomplished by using the predictive data, and thus
finally, the feedback values will be added to the output of ensure the secondary control can be realized.
primary control. After the system is stable, Switch 4 will be closed.
Meanwhile, there will be
4.2.1. Secondary control on voltage. Take the ith DER as (
the example, and the P
x=Q
U droop control equation is ^ i1 ¼ Ui1 þ dUi
U
(35)
as Eq. (29). The output voltage can be expressed in the dq dUi ¼ Uref
Ui1
coordinate system as below [16]:
( qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where Ui1 is the final voltage value obtained by consensus
Ui ¼ Udi 2 þ Uqi2 ^ i1 is the obtained voltage after adjusting Ui1 .
control; U
(31) ^ i1
Ui .
Uqi ¼ 0 The feedback value on voltage is U

where Udi and Uqi are voltages obtained by Park transform- 4.2.2. Secondary control on frequency. The goal of second-
ation. So, the primary control on voltage can be also writ- ary control on frequency is adjusting xi to xref . To
ten as achieve this goal, based on the differential of Eq. (29), an
 auxiliary variable uxi is set up in this article. Meanwhile,
Udi ¼ Uref
ni Qi
(32)
Uqi ¼ 0 the following equation can be established.

Therefore, the goal of the secondary control on voltage x_ i ¼ x_ ref
ni Q_ i ¼ uxi ; i ¼ 1; 2; :::; n (36)
is adjusting Udi to Uref . To achieve the goal, based on the It is similar to Eq. (34), the consensus protocol for fre-
differential of (32), an auxiliary variable uUi is set in this quency can be designed as below:
article. Meanwhile, the following equation can be 8
>
> 1: lim Dx0i <Txi1 and the overshoot of AVLFCC<Txi2
established. >
> X
>
>
t!1
>
> uxi ¼ kxi1 ½aij ðlij ðtÞxj ðt
sij ðtÞÞ
xi ðtÞ þ kxi2 bi ½liL ðtÞxL
xi ðtÞ
>
>
_ i uUi
U_ di ¼ U_ ref
ni Q (33) < j2Ni
2: lim Dx0i
Txi1 or the overshoot of AVLFCC
Txi2
>
> t!1
Based on the analysis in Section 3, the delay and loss >
> ð1ÞUse the predictive
>
> X data
>
> ^ j ðtÞ
x
^ i ðtÞÞ þ bi ðx
^ L
x
^ i ðtÞÞ
problems are considered in this article. To obtain a better > ð2Þuxi ¼ kxi ½
> aij ðx
: j2Ni
control effect, the consensus protocol for voltage is
designed as below: (37)

Parameter of MG1 Value Parameter of MG1 Value

sj1 =sL1 0.1s Other lij 0

lj1 =lL1 0.9 kxi (used in normal situation) kxi ¼
4
Othersij 0 kxi1 ,kxi2 (used in system when delay occurs) kxi1 ¼
0:43; kxi2 ¼
43
kxi1 ,kxi2 (used in system when loss occurs) kxi1 ¼ 4:3; kxi2 ¼ 430
TABLE 2. Parameters selected in Case 2.
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 67

FIGURE 8. The effect on predictive compensation part. FIGURE 9. Variations of frequency and voltage in MG1.
(a) The effect on predictive compensation part without dis- (a) Variations of frequency (delay occurs in the system);
turbance; (b) The effect on predictive compensation by (b) Variations of frequency (loss occurs in the system); (c)
MPC (the Load2 is added at t ¼ 2.5 ms); (c) The effect on Variations of frequency by using AVLFCC (delay occurs
predictive compensation by ELM (the Load2 is added at
in the system); (d) Variations of frequency by using
t ¼ 2.5 ms); (d) The effect on predictive compensation by
the proposed part (the Load2 is added at t ¼ 2.5 ms). AVLFCC (loss occurs in the system).
68 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

Case
Time 1) 2) 3)

t ¼ 1s sj1 =sj5 ¼ 0:5s; lj1 =lj5 ¼ 0:99; sj1 =sj5 ¼ 0:01s; lj1 =lj5 ¼ 0:99; sj1 =sj5 ¼ 0:005s; lj1 =lj5 ¼ 0:99;
other sij ¼ 0s; other lij ¼ 1 other sij ¼ 0s; other lij ¼ 1 other sij ¼ 0s; other lij ¼ 1
t ¼ 2s Turn sj1 =sj5 to 0.01s Turn sj1 =sj5 to 0.5s Turn sj1 =sj5 to 0.01s
TABLE 3. The running process in Case 3.

where x^ i and x
^ j are the predictive frequency values of the used to test the accuracy of the predictive compensation part.
ith DER and the jth DER, respectively; Kxi , Kxi1 and Kxi2 The simulation results are as shown in Figure 8.
are the gains. When selecting consensus protocol, it should The analysis is as below: As shown in Figure 8(a),
be the same as (34), judging first. there is no significant difference among the three predict-
After the system is stable, Switch 3 will be closed, and ive methods when the system is stable. After comparing
xi1 will be adjusted to x ^ i1 . Meanwhile, there will be Figures 8(b)–(d), it has found that: When the system has
 a large fluctuation, the adaptability of MPC method is
^ i1 ¼ xi1 þ dxi
x
(38) poor. There is the maximum error between the predictive
dxi ¼ xref
xi1
data by MPC and the true data; ELM is a method of ran-
where xi1 is the final frequency value obtained by the domly selecting input weights, different ELM models
consensus control. The feedback value on frequency is Tx1 . have different adaptive abilities to the input data.
Therefore, the predictive error of the ELM model may be
5. CASE STUDY AND ANALYSIS also very large; the predictive data is the most similar to
the true data by using the proposed part combined with
To verify the control effect of the control strategy by appli-
MPC and ELM.
cation, an islanded MG model similar to the actual system
is designed as the experimental model. The specific struc-
ture is as shown in Figure 7. The model is composed of
MG1 and MG2. In order to be reasonable and simplify the
5.2. Case 2: Verify the Effectiveness of the Distributed
experiment process, the DERs are replaced by DC (or AC)
Control Strategy in the Actual System
power sources. The simulation parameters are shown in the
Table 1. The virtual leader-following consensus protocol is used in
this article to accomplish the distributed control (secondary
5.1. Case 1: Verify the Effectiveness of the Predictive control). The voltage and frequency of the virtual leader
Compensation Part are set as Uref and xref , respectively. According to the MG
In this case, the obtained historical frequency data of DER1 system shown in Figure 7, the corresponding matrices are
by droop control is selected as the experiment object. It is set as below:

8 2 3 2 3
>
> 0 1 1 1 0 0
>
> 6 7 6 7
>
> 6
>
> T off : A ¼ 41 0 175; other aij ¼ 0 B ¼ diag½b1 ; b2 ; b3  ¼ 6 40 1 07
5
>
>
2 a
Aa : a57 ¼ 1; a65 ¼ 1; a76 ¼ 1and other aij ¼ 0
>
> 2 3
>
< 1 1 0 0 0 1
2 3 2 3 1 0 0
MG1 : 0 1 1 1 1 0 0 0 MG2 : 6 7
>
> B ¼ diag ½ b ; b ; b  ¼ 60 1 07
>
> 6 7 6 7 5 6 7 4 5
>
> 61 0 1 17 60 1 0 07
>
> T on : A ¼ 6 7; other aij ¼ 0 B ¼ diag½b1 ; b2 ; b3 ; b4  ¼ 6 7 0 0 1
>
>
2 a 61 1 0 17 60 07
>
> 4 5 4 0 1 5
>
:
1 1 1 0 0 0 0 1
(39)
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 69

Based on the analysis in Section 3 part D, the gains in the adjusted to their respective reference values after a transi-
consensus protocol should satisfy the following conditions: ent process.

8
>
> kx51 a57 l57 ð0Þk57 ð0ÞX7 ð0Þ=4p2 þ kx52 b5 l5L ð0ÞXL ð0Þ=2p þ x5 ð0Þ
>
> þ c5 ¼ 0
>
> kx51 a57 þ kx52 b5
>
>
<
k a l ð0Þk65 ð0ÞX5 ð0Þ=4p2 þ kx62 b6 l6L ð0ÞXL ð0Þ=2p þ x6 ð0Þ
MG2 : x61 65 65 þ c6 ¼ 0
>
> kx61 a65 þ kx62 b6
>
>
>
> kx71 a76 l76 ð0Þk76 ð0ÞX6 ð0Þ=4p2 þ kx72 b7 l7L ð0ÞXL ð0Þ=2p þ x7 ð0Þ
>
>
: þ c7 ¼ 0
kx31 a76 þ kx72 b7

8 
>
> kx11 a12 l12 ð0Þk12 ð0ÞX2 ð0Þ þ a13 l13 ð0Þk13 ð0ÞX3 ð0Þ =4p2 þ kx12 b1 l1L ð0ÞXL ð0Þ=2p þ x1 ð0Þ
>
> þ c1 ¼ 0
>
> kx11 ða12 þ a13 Þ þ kx12 b1
>
>
>
< 
kx21 a21 l21 ð0Þk21 ð0ÞX1 ð0Þ þ a23 l23 ð0Þk23 ð0ÞX3 ð0Þ =4p2 þ kx22 b2 l2L ð0ÞXL ð0Þ=2p þ x2 ð0Þ
MG1 : T2 off : þ c2 ¼ 0
>
> kx21 ða21 þ a23 Þ þ kx22 b2
>
> 
>
>
>
> kx31 a32 l32 ð0Þk32 ð0ÞX2 ð0Þ þ a31 l31 ð0Þk31 ð0ÞX1 ð0Þ =4p2 þ kx32 b3 l3L ð0ÞXL ð0Þ=2p þ x3 ð0Þ
>
: þ c3 ¼ 0
kx31 ða32 þ a31 Þ þ kx32 b3
8 
>
> kx11 a12 l12 ð0Þk12 ð0ÞX2 ð0Þ þ a13 l13 ð0Þk13 ð0ÞX3 ð0Þ þ a14 l14 ð0Þk14 ð0ÞX4 ð0Þ =4p2 þ kx12 b1 l1L ð0ÞXL ð0Þ=2p þ x1 ð0Þ
>
> þ c1 ¼0
>
> kx11 ða12 þ a13 þ a14 Þ þ kx12 b1
>
>
>
> 
>
> kx21 a21 l21 ð0Þk21 ð0ÞX1 ð0Þ þ a23 l23 ð0Þk23 ð0ÞX3 ð0Þ þ a24 l24 ð0Þk24 ð0ÞX4 ð0Þ =4p þ kx22 b2 l2L ð0ÞXL ð0Þ=2p þ x2 ð0Þ þ c2
> 2
>
< ¼0
kx21 ða21 þ a23 þ a24 Þ þ kx22 b2
T2 on : 
>
> kx31 a32 l32 0 k32 0 X2 0 þ a31 l31 0 k13 0ÞX1 ð0Þ þ a34 l34 ð0Þk34 ð0ÞX4 ð0Þ =4p2 þ kx32 b3 l3L ð0ÞXL ð0Þ=2p þ x3 ð0Þ
ð Þ ð Þ ð Þ ð Þ ð
>
> þ c3 ¼0
>
> kx31 ða23 þ a13 þ a43 Þ þ kx32 b3
>
>
>
> 
>
> kx41 a42 l42 ð0Þk42 ð0ÞX2 ð0Þ þ a43 l43 ð0Þk43 ð0ÞX3 ð0Þ þ a41 l41 ð0Þk41 ð0ÞX1 ð0Þ =4p2 þ kx42 b4 l4L ð0ÞXL ð0Þ=2p þ x4 ð0Þ
>
> þ c4 ¼0
: kx41 ½a24 þ a34 þ a14  þ kx42 b4
(40)

Ð Ð Ð
where c3 ¼ x3 dtjt¼0 ;c4 ¼ x4 dtjt¼0 ; c5 ¼ x5 dtjt¼0 ; c6 ¼ 5.3. Case 3: Verify the Effectiveness of the Event-
Ð Ð
x6 dtjt¼0 ; c7 ¼ x7 dtjt¼0 ; x represents the voltage or frequency Triggered Mechanism
of DER; X represents the Laplace equation of x. Take the To verify the effectiveness of the trigger mechanism on solv-
variations of frequency in MG1 (T2 on) as the example, and the ing the problems of time-varying delay and loss rate, the
selected parameters are as shown in the Table 2. MG1 and MG2 have all been experimented. For the parame-
The simulation results are as Figure 9. ters used in the experiment, the selecting rules are as follows.
The analysis is as below: As shown in Figure 9(a) and Delay is selected as 0.5 s and 0.01 s. The parameters are
(b), the obtained voltages and frequency by secondary con- obtained by simulation experiment. After a lot of simulation
trol will oscillate over a period of time when delay or loss experiments, it has found that: when the delay is 0.5 s, there
occurs in the cyber layer. And the overshoot in the control will be a large steady-state error and a large overshoot; when
process is very large. The control effect of AVLFCC is delay is 0.01 s, the control effect of consensus control will be
shown in Figure 9(c) and (d). As shown in the two figures, less; Tf 12 and TU12 : For a MG system, the generally permis-
the output frequency and voltage of each DER can be sible frequency and voltage deviations are
0:05fref 0:05fref
70 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

and
0:1Uref 0:1Uref , respectively. Therefore, Tf 12 is

selected as þ
2.5Hz, and TU12 is selected as þ
31V;
Tf 11 and TU11 : For Event1 (i.e. steady-state error), it can be
artificially selected. In this article, the steady-state error of fre-
quency and voltage are selected as þ
0.5 Hz and þ
1
V, respectively.
To get closer to the actual situation, the experiments are
divided into three categories: case 1) The delay will be
changed from 0.5 s to 0.01 s at t ¼ 2 s (from trigger to no
trigger); case 2) The delay will be changed from 0.01 s to
0.5 s at t ¼ 2 s (from no trigger to trigger); case 3) The
delay will be changed from 0.005 s to 0.01 s at t ¼ 2 s
(from no trigger to no trigger). In this experiment, the sec-
ondary control is used at t ¼ 1 s, and the running process is
shown in the Table 3.
Remark 4: In this case, the effect of trigger mechanism is
verified only by changing the delay in the output data of
DER1/DER5.
The variations in MG1 are shown as Figure 10.
The variations in MG2 are shown as Figure 11.
The analysis is as below: As shown in Figures 10(a)
and (b) and 11(a) and (b), after the trigger mechanism
is triggered at t ¼ 1s, the predictive data will be used to
accomplish the consensus control. The completion of
secondary control can be also ensured. When the delay
of DER1 is adjusted to 0.01 s at t ¼ 2s, the AVLFCC
will be used to accomplish the secondary control. After
a period of transient process (adjust the parameters in
the controller), the output frequency and voltage of
each DER can be also adjusted to their respective refer-
ence values. In Figures 10(c) and (d) and 11(c) and
(d), the AVLFCC is used to accomplish the consensus
control before t ¼ 2s. When the delay of DER1 is
adjusted to 0.5 s, the event-triggered mechanism will be
triggered, and the predictive data will be used to
accomplish the secondary control. It also ensures the
output frequency and voltages of DERs can be adjusted
to their respective reference values. In Figures 10(e)
and (f) and 11(e) and (f), when the delay of DER1 is
changed from 0.005 s to 0.05 s at t ¼ 2s, the event-trig-
gered mechanism will not be triggered. And the
AVLFCC will be still used to accomplish the secondary
FIGURE 10. Variations of frequency and voltage in MG1 control. The secondary control ultimately keeps the con-
by using event-triggered mechanism. The variations in
MG2 are shown as Figure 11. (a)Variations of frequency in trolled variables of each DER at the corresponding ref-
MG1 under case 1); (b) Variations of voltage in MG1 erence values respectively. In summary, the simulation
under case 1); (c) Variations of frequency in MG1 under results have shown that the delay and packet loss prob-
case 2); (d)Variations of voltage in MG1 under case 2); (e)
Variations of frequency in MG1 under case 3); (f) lems can be effectively solved by the event-trig-
Variations of voltage in MG1 under case 3). gered mechanism.
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 71

5.4. Case 4: Verify the Effectiveness of the Distributed

Control on Plug and Play of DER (Load)
In this case, it is used to verify the effectiveness of the dis-
tributed control on plug and play of load (or DER). The
simulation results are as shown in Figures 12–15.

5.4.1. Load disturbance exists in the system. In this case,

the control effect of distributed control is compared with
the drop control. And the running process is as follows:
the load2 will be added at t ¼ 2s, and cut down at t ¼ 3s;
the secondary control will be added at t ¼ 1s.
The analysis is as follows:
For MG1: As shown in Figure 12(a), the droop control
makes the frequency of DERs tend to be 49.3Hz at
t ¼ 0.5s. After adding load2 at t ¼ 2s, the frequency of each
DER will drop to 49.2 Hz. And the frequency of DERs
will be raised to almost 49.3 Hz again at t ¼ 3s. During the
process, there are errors between the output frequency of
DERs and the corresponding reference value. The varia-
tions of voltage by using droop control are shown in
Figure 12(b). As shown in this figure, the droop control
makes the output frequency of DERs tend to be 309.5 V
before t ¼ 0.5s. By adding load2 at t ¼ 2s, it is found that
the output voltage of each DER will drop at t ¼ 2s. After
cutting down load2 at t ¼ 3s, the output voltage of each
DER will raise to almost 309.5 V again. There are also
errors between the output voltages of DERs and the corre-
sponding reference value. In Figure 12(c) and (d), the
novel distributed control is added at t ¼ 1s. Through this
control method, the output frequency of DERs will be
adjusted to 50 Hz at t ¼ 1.3s. Meanwhile, the output vol-
tages of DERs will be adjusted to 311 V. For MG2: The
variations of frequency obtained by droop control are
shown in Figure 13(a). In this figure, the output frequency
of DERs will drop to almost 48.95 Hz at t ¼ 2s. And then,
it will be raised to almost 49.1 Hz at t ¼ 3s. There are
errors between the obtained frequency and the correspond-
ing reference value. In Figure 13(b), the output voltages of
DER5, DER6 and DER7 are different by using droop con-
trol. And the voltages are between 308.5 V and 308.8 V.
After adding load2 at t ¼ 2s, the voltage of each DER has
dropped at first. And then, it will be raised at t ¼ 3 s. there
FIGURE 11. Variations in MG2 by using event-triggered are also errors between the obtained voltages and the corre-
mechanism. (a)Variations of frequency in MG2 under case
1); (b) Variations of voltage in MG2 under case 1); (c) sponding reference value. By analyzing the variations in
Variations of frequency in MG2 under case 2); Figure 13(c) and (d), it is found that the output frequency
(d)Variations of voltage in MG2 under case 2); (e) and voltage of each DER can still be adjusted to their
Variations of frequency in MG2 under case 3); (f) respective reference values by using the novel distrib-
Variations of voltage in MG2 under case 3).
uted control.
72 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

5.4.2. DER disturbance exists in the system. Due to the

increase or decrease of DER4 has more influence on MG1,
so only the experiment results in MG1 are discussed in this
case. To compare the control effect, the experiment results
on droop control are also carried out in this article. The
running process is as follows: the DER4 will be added/cut
down at t ¼ 2s; the secondary control will be added
at t ¼ 1 s.
The analysis is as follows:
Adding DER4: As shown in Figure 14(a) and (b), the
droop control makes the output frequency and voltages of
DER1, DER2, and DER3 tend to be almost 49.3Hz and
309 V respectively. After adding DER4 at t ¼ 2s, there are
drastic changes of frequency and voltage between t ¼ 2s
and t ¼ 2.65s. This is because when the DER4 is added in
the system, the system tackles it as a load. So, there is a
transition before DER4 works normally in the system.
Finally, the output frequency of the four DERs is adjusted
to almost 49.4 Hz. And the output voltages of the four
DERs are still not stable before t ¼ 3s. The control effect
of distributed control is shown in Figure 14(c) and (d). As
shown in the two figures, the output frequency and voltage
of each DER will be adjusted to their respective reference
values at t ¼ 1.3s. After adding DER4 at t ¼ 2s, there is
also a transition between t ¼ 2s and t ¼ 4.5s. But finally,
the frequency and voltage will be adjusted to their respect-
ive reference values again. Cutting down DER4: As shown
in Figures 15(a) and (b), the droop control makes the fre-
quency tend to be 49.45 Hz. After cutting down DER4 at
t ¼ 2s, the frequency of DER1, DER2 and DER3 has
drooped to almost 49.3 Hz. And the output voltages of the
four DERs are between 309.6 V and 309.8 V. After cutting
down DER4 at t ¼ 2s, the voltages of DER1, DER2 and
DER3 have drooped. The control effect of distributed con-
trol is shown in Figures 15(c) and (d). And it’ found that
the frequency and voltage of each DER can be still
adjusted to their respective reference values by the novel
distributed control method.
Furthermore, the distributed control strategy proposed in
this article is compared with the prior technique: Take the
control strategy in [18] as an example, the two control
strategies have many similar advantages as below: 1. The
secondary control is effect: Both of them have a good sec-
FIGURE 12. variations of frequency and voltage in MG1 ondary control effect on voltage and frequency; 2. The per-
(load disturbance exists in the system). (a)Variations of fre-
formance of plug and play is effect: The plug and play of
quency in MG1 by using the droop control; (b) Variations
of voltage in MG1 by using the droop control; (c) DERs can be implemented in both articles. However, no
Variations of frequency in MG1 by using the novel distrib- appropriate method is designed in [18] to solve the prob-
uted control; (d) Variations of frequency in MG1by using lems of time-varying delay and loss rate. This is the big-
the novel distributed control. gest difference between the two articles. After comparing
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 73

FIGURE 13. Variations of frequency and voltage in MG2 FIGURE 14. Variations of frequency and voltage in MG1
(load disturbance exists in the system). (a) Variations of when the DER4 is added at t ¼ 2 s. (a)Variation of fre-
voltage in MG2 by using droop control; (b) Variations of quency in MG1 by using the droop control; (b) Variations
frequency in MG2 by using droop control; (c) Variations of voltage in MG1 by using the droop control; (c)
of frequency in MG2 by using the novel distributed con- Variations of frequency in MG1 by using the distributed
trol; (d) Variations of voltage in MG2 by using the novel control; (d) Variations of voltage in MG1 by using the dis-
distributed control. tributed control.
74 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

the experimental results in this article and those in [18],

the analysis is as below: Although the experiments of delay
and loss for secondary control are carried out in [18], it is
concluded that the method can overcome the problems in a
finite range. The method in [18] has a good robustness in
facing with small delay and small loss rate. Different from
this, an event-triggered mechanism is designed in this art-
icle to solve the communication problems. When the delay
or loss rate is large, the actual communication data will be
replaced by the predictive data. Through this mechanism,
the conservatism of consensus control for delay and loss
problems is greatly reduced.

6. CONCLUSIONS
In this article, a novel distributed control strategy is proposed
to improve the control effect on output voltage and frequency
of each DER in MG. Based on the concept of CPS, the novel
control structure is consisted by cyber and physical layers. In
the cyber layer, the communication structures of overall MG
and single DER are constructed. Meanwhile, to solve the prob-
lems of time-varying delay and loss rate, an event-triggered
mechanism combined with AVLFCC and predictive compensa-
tion is also proposed in this layer. In the physical layer, based
on the proposed mechanism, a novel secondary control method
is proposed to obtain a better control effect on DERs. Finally,
the case 1 has confirmed that the expected predictive data can
be obtained through the predictive compensation part; the case
2 has confirmed that the problems of delay and loss have been
solved effectively in this article; the case 3 has confirmed that
the output voltage and frequency of each DER can be adjusted
to their respective reference values by using the secondary con-
trol method and event-triggered mechanism; the stability of the
novel control strategy has been confirmed in the case 4 under
the condition of load or DER variations.

FUNDING
This work is supported by the Key projects of the National
Natural Science Foundation of China under Grant
61833008, National Natural Science Foundation of China
under Grants 61573300 and 61533010, Jiangsu Provincial
FIGURE 15. Variations of frequency and voltage in MG1 Natural Science Foundation of China under Grant
when the DER4 is cut down at t ¼ 2 s. (a) Variations of fre- BK20171445, Key research and development plan of
quency in MG1 by using the droop control; (b) Variations Jiangsu Province-Industrial foresight and common key
of voltage in MG1 by using the droop control; (c) technical projects under Grant BE2016184, and Hubei
Variations of frequency in MG1 by using the distributed
control; (d) Variations of voltage in MG1 by using the dis-
Provincial Natural Science Foundation of China under
tributed control. Grant E2016203374.
 Zhang et al.: Distributed Control Strategy of Microgrid Based on the Concept of Cyber Physical System 75

REFERENCES cooperative control of multi-agent systems,” IET Generation

Transmission Distribution, vol. 7, no. 8, pp. 822–831, 2013.
[1] K. D. Kim, and P. R. Kumar, “Cyber-Physical systems: a [16] C. X. Dou, D. Yue, J. M. Guerrero, X. P. Xie, and S. L. Hu,
perspective at the centennial,” Proc. IEEE, vol. 100, no. 1, “Multiagent system-based distributed coordinated control for
pp. 1287–1308, 2012. radial DC microgrid considering transmission time delays,”
[2] Y. Mo et al., “Cyber-physical security of a smart grid infra- IEEE Trans. Smart Grid, vol. 8, no. 5, pp. 1–12, 2017.
structure,” Proc. IEEE, vol. 100, no. 1, pp. 195–209, 2012. [17] C. X. Dou, D. Yue, Z. Q. Zhang, and J. M. Guerrero,
[3] S. Siddharth, H. Adam, and G. Manimaran, “Cyber-physical “Hierarchical delay-dependent distributed coordinated con-
system security for the electric power grid,” Proc. IEEE, trol for DC ring-bus microgrids,” IEEE Access, vol. 5, pp.
vol. 100, no. 1, pp. 210–224, 2012. 10130–10140, 2017.
[4] T. Facchinetti, and M. L. Delia Vedova, “Real-time model- [18] Q. Shafiee, V. Nasirian, J. C. Vasquez, J. M. Guerrero, and
ing for direct load control in cyber-physical power systems,” A. Davoudi, “A multi-functional fully distributed control
IEEE Trans. Ind. Inform., vol. 7, no. 4, pp. 689–698, 2011. framework for AC microgrids,” IEEE Trans. Smart Grid.
[5] C. X. Dou, D. L. Liu, X. B. Jia, and F. Zhao, “Management vol. 9, pp. 3247–3258, 2018. .
and control for smart microgrid based on hybrid control the- [19] X. Feng, A. Shekhar, F. Yang, R. E. Hebner, and P. Bauer,
ory,” Electr. Power Components Syst., vol. 39, no. 8, pp. “Comparison of hierarchical control and distributed control
813–832, 2011. for microgrid,” Electric Power Components Syst., vol. 45,
[6] N. Vahidreza, D. Ali, L. Frank, and J. M. Guerrero, no. 10, pp. 1043–1056, 2017.
“Distributed adaptive droop control for DC distribution sys- [20] C. E. Ren, L. Chen, and C. L. Philip Chen, “Adaptive fuzzy
tems,” IEEE Trans. Energy Convers., vol. 29, no. 4, pp. Leader-Following consensus control for stochastic multi-
944–956, 2014. agent systems with heterogeneous nonlinear dynamics,”
[7] C. X. Dou, Z. Q. Zhang, D. Yue, and M. M. Song, IEEE Trans.Fuzzy Syst., vol. 25, no. 1, pp. 181–190, 2017.
“Improved droop control based on virtual impedance and [21] X. You, C. C. Hua, D. Peng, and X. P. Guan,
virtual power source in low-voltage microgrid,” IET “Leader–following consensus for multi-agent systems sub-
Generation Transmission Distribution, vol. 11, no. 4, pp. ject to actuator saturation with switching topologies and
1046–1054, 2017. time-varying delays,” IET Control Theory Appl., vol. 10, no.
[8] T. Wu, Z. Liu, J. Liu, S. Wang, and Z. You, “A unified vir- 2, pp. 144–150, 2016.
tual power decoupling method for droop-controlled parallel [22] M. Yu, C. Yan, D. M. Xie, and G. M. Xie, “Event-triggered
inverters in microgrids,” IEEE Trans. Power Electron., vol. tracking consensus with packet losses and time-varying
31, no. 8, pp. 5587–5603, 2016. delays,” IEEE/CAA J. Automatica Sin., vol. 3, no. 2, pp.
[9] W. Liu, W. Gu, W. X. Sheng, X. L. Meng, Z. J. Wu, and 165–173, 2016.
W. Chen, “Decentralized multi-agent system based coopera- [23] Y. Zhang, R. Wang, T. Zhang, Y. J. Liu, and B. Guo, “Model
tive frequency control for autonomous microgrids with com- predictive control-based operation management for a residential
munication constraints,” IEEE Trans. Sustain. Energy, vol. microgrid with considering forecast uncertainties and demand
5, no. 2, pp. 446–456, 2014. response strategies,” IET Generation Transmission
[10] Z. G. Wang, W. C. Wu, and B. M. Zhang, “A fully distrib- Distribution, vol. 10, no. 10, pp. 2367–2378, 2016.
uted power dispatch method for fast frequency recovery and [24] G. B. Huang, Q. Y. Zhu, and C. K Siew, “Extreme learning
minimal generation cost in autonomous microgrids,” IEEE machine: a new learning scheme of feedforward neural
Trans. Smart Grid, vol. 7, no. 1, pp. 19–31, 2016. networks,” in IEEE International Joint Conference on
[11] W. Liu et al., “Improved average consensus algorithm based Neural Networks, IEEE, Budapest, Hungary, pp. 985–990.
distributed cost optimization for loading shedding of autono-
mous microgrids,” Int. J. Electr. Power Energy Syst., vol. 7,
BIOGRAPHIES
no. 3, pp. 89–96, 2015.
[12] Lv, Z. Y. Wu, Z. J. Dou, X. B. Hu, M. Q. Zhao, and B. Bo Zhang received the B. S. degrees in electrical engineer-
“An adaptive Droop control for the islanded DC microgrid ing and automation from the Liren college of Yanshan uni-
based on discrete Consensus algorithm,” Proc. CSEE, vol. versity, Qinhuangdao, China, in 2016. He is currently
35, no. 17, pp. 4397–4407, 2015.
pursuing the M. S. degree in automation with Yanshan
[13] K. O. Oureilidis, and C. S. Demoulias, “A decentralized
impedance-based adaptive droop method for power loss University, Qinhuangdao, China. He is also currently a stu-
reduction in a converter-dominated islanded microgrid,” dent member of the Chinese Association of Automation.
Sustain. Energy Grids Networks, vol. 5, pp. 39–49, 2016. His current research interests include microgrids and dis-
[14] P. B. Wang, X. N. Lu, X. Yang, W. Wang, and D. G. Xu, tributed generation technology.
“An improved distributed secondary control method for DC
microgrids with enhanced dynamic current sharing perform-
Chunxia Dou received the B.S. and M.S. degrees in auto-
ance,” IEEE Trans. Power Electron., vol. 31, no. 9, pp.
6658–6673, 2016. mation from the Northeast Heavy Machinery Institute,
[15] A. Bidram, A. Davoudi, F. L. Lewis, and Z. H. Qu, Qiqihaer, China, in 1989 and 1994, respectively, and the
“Secondary control of microgrids based on distributed Ph.D. degree in Institute of Electrical Engineering from
76 Electric Power Components and Systems, Vol. 470 (2019), No. 1-2

Yanshan University, Qinhuangdao, China, in 2005. In Zhanqiang Zhang received the dual B.S. degree in elec-
2010, she joined the Department of Engineering, Peking trical engineering and automation/mathematics and applied
University, Beijing, China, where she was a Postdoctoral mathematics from Hebei University of Science and
Fellow for two years. Since 2005, she has been a Technology, Shijiazhuang, China, in 2015. He is currently
Professor in Institute of Electrical Engineering, Yanshan pursuing the M.S. degree in automation with Yanshan
University. Her current research interests include multi- University, Qinhuangdao, China. He is currently a Chinese
agent based control, event-triggered hybrid control, dis- Association of Automation Member. His current research
tributed coordinated control, and multi-mode switching interests include distributed generation control and micro-
control and their applications in power systems, grid networked control research.
Microgrids and smart grids.
Kai Ma received the B.Sc. degree in automation and Ph.D.
Dong Yue received the Ph.D. degree from the South China degree in control science and engineering from Yanshan
University of Technology, Guangzhou, China, in 1995. He University, Qinhuangdao, China, in 2005 and 2011,
is currently a professor and dean of Institute of Advanced respectively. In 2011, he joined Yanshan University. From
Technology, Nanjing University of Posts and 2013 to 2014, he was a Postdoctoral Research Fellow with
Telecommunications and also a Changjiang Professor with Nanyang Technological University, Singapore. He is cur-
the Department of Control Science and Engineering, rently an Associate Professor with the Department of
Huazhong University of Science and Technology. His Automation, School of Electrical Engineering, Yanshan
research interests include analysis and synthesis of net- University. His research interests include demand response
worked control systems, multi-agent systems, optimal con- in smart grids and resource allocation in communica-
trol of power systems, and internet of things. tion networks.