IEEE POWER & ENERGY SOCIETY SECTION

Received 23 June 2022, accepted 5 August 2022, date of publication 16 August 2022, date of current version 29 August 2022.
Digital Object Identifier 10.1109/ACCESS.2022.3198949

Consensus-Based Hybrid Multiagent Cooperative

Control Strategy of Microgrids Considering
Load Uncertainty
WOON-GYU LEE 1 , (Member, IEEE), THAI-THANH NGUYEN 2, (Member, IEEE),
HYEONG-JUN YOO 3 , (Member, IEEE), AND HAK-MAN KIM 4, (Senior Member, IEEE)
1 Department of Electrical Engineering, Incheon National University, Incheon 406-772, South Korea
2 Advanced Grid Innovation Laboratory for Energy (AGILe), New York Power Authority, White Plains, NY 10601, USA
3 Division of Smart Distribution Research Center, Korea Electrotechnology Research Institute, Gwangju 61751, South Korea
4 Research Institute for Northeast Asian Super Grid, Incheon National University, Incheon 406-772, South Korea

Corresponding author: Hak-Man Kim ([email protected])

This work was supported by the Korea Institute of Energy Technology Evaluation and Planning (KETEP) and the Ministry of Trade,
Industry & Energy (MOTIE) of the Republic of Korea (No. 20206910100020).

ABSTRACT The energy management systems used in microgrid systems are either centralized or decen-
tralized. The centralized method is easy to implement, but has low reliability and it is computationally
intensive due to its reliance on a single control unit. The decentralized control is a more reliable strategy
as multiple units are coordinated to manage the microgrid system, however, its algorithm and implemen-
tation are complicated for long-term operation. This paper proposes a consensus-based hybrid multi-agent
management system including both centralized and decentralized agents taking advantages of both methods
into account. The hierarchical cooperative optimization system is adapted for the distributed economic
dispatch. The primary control layer is responsible for the voltage and frequency regulations whereas the
secondary control layer is implemented in a consensus-based distributed multi-agent control scheme for
optimal microgrid management. The centralized agent is used to optimally schedule the energy storage
system. The proposed hybrid control strategy consistently maintains the economic operation and demand
load balance of the microgrid system under load uncertainty. A comparison study with the centralized
control strategy is presented to demonstrate the effectiveness of the proposed control strategy. The feasibility
of the proposed controller is validated by the control-hardware-in-the-loop simulation using OPAL-RT
Technologies. Real-time simulation results show that the proposed hybrid strategy has a better optimization
performance compared to the centralized strategy under load uncertainty conditions.

INDEX TERMS Economic dispatch, hybrid multi-agent cooperative system, load uncertainty, microgrid.

NOMENCLATURE PESS
e Charging or discharging power
of ESS.
Cd (PDG
d ) Quadratic generation cost function
PESS
e,max Upper bound of ESS.
of DGs.
αeESS , βeESS , γeESS Quadratic cost coefficients of ESS.
PDG
d Output power of d-th DGs.
ω Angular frequency.
αdDG , βdDG , γdDG Quadratic cost coefficients.
1V ∗ Amplitude deviation of voltage.
d,min , Pd,max
PDG DG Lower and upper bounds of DGs.
kdp , kdq Droop coefficients.
Ce (PESS
e ) Quadratic generation cost function 1P∗ Deviation of active power.
of ESS. Q∗ Reactive power reference.
The associate editor coordinating the review of this manuscript and ε Communication link between all
approving it for publication was Guangya Yang . agents.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
88798 VOLUME 10, 2022
W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

ni , nj Number of agents adjacent to the MG system to optimize the economic dispatch problem,
agents i and j. resulting in optimal operation of the MG system [4], [5], [6],
δPi (k) Mismatch power between EMS [7]. Traditional MG control system schemes for economic
set-point and the measured power dispatch can be classified into two approaches: centralized
at iteration k. and decentralized optimization [8].
PEMS (k) Power set-point from the The centralized energy management systems (EMSs)
i
centralized EMS at iteration k. gather all MG system information, such as generations and
Pi (k) Measured power of i-th agent load profiles to find the optimal operation points of the
at iteration k. MG system. Various centralized optimization techniques
λDG have been suggested to solve economic dispatch problem,
d (k) Incremental cost of DG at
iteration k. such as the gradient search method [9], linear program-
λESS ming method [10], lamda iteration strategy [11], genetic
e (k) Incremental cost of ESS at
iteration k. algorithm [12], population dynamics approach [13], particle
λi (k + 1) Incremental cost at swarm optimization [14], evolutionary programming [15].
iteration k + 1. However, these strategies require the powerful centralized
Pdis Distributed power reference of DG. EMS to gather global data and process the huge amounts of
e,DG
Pdis Distributed power reference of ESS received information. In the centralized methods, information
e,ESS
on the load profiles is predicted for long-term scheduling.
agents
Inaccuracy in the load prediction can negatively affect the
Pref (k) Generation power reference of each
optimal operation of the MG system [16]. Therefore, the
agent at iteration k.
power system uses various load forecasting methods to effi-
CdDG Per-unit generation cost of
ciently integrate and allocate energy resources, and adjust
d-th DG.
supply and demand power [17]. Some papers proposed prob-
PDG
d (k) Output power of d-th DG
abilistic load forecasting (PLF) strategies, such as density
at iteration k.
estimation and quantile regression to generate probabilis-
yd (k) Start-up status identifier at
tic prediction [18], [19], [20]. The centralized optimization
iteration k.
problem typically assumes a perfect forecast for DER output
zd (k) Shutdown status identifier at
and demand load. However, the demand load consisting of
iteration k.
various characteristic patterns cannot be fully predicted [21].
CdSU , CdSD Start-up and shutdown cost.
Thus, these strategies are limited in responding immediately
ud (k) Commitment status identifier at
to changes in operating conditions, such as the intermit-
iteration k.
tent characteristic of renewable energy resources and load
d,min , Pd,max
PDG DG Minimum and maximum uncertainty [22]. Recently, several intelligent load allocation
power generation. solutions have been proposed to avoid load prediction [23],
PESS+
e (k), PESS+
e (k) Charging and discharging amount [24], [25]. In [23], the author proposed the two-stage feeder
of ESS at iteration k. load decomposition algorithm based on a pivot point using
PESS
e,cap Maximum capacity of ESS. smart meters. The improved load profile for residential net-
e,loss , Pe,loss
PESS+ ESS−
Charging and discharging loss work customers was proposed through the Gaussian mixture
of ESS. model and factor analyzers mixture model in [24]. In [25],
ESS , SOC ESS
SOCe,min Lower and upper limits for SOC the author has developed a novel smart energy management
e,max
of ESS framework based on the quality of electrical service. How-
PLoad
l (k) Demand load at iteration k. ever, these solutions require a powerful centralized controller
PPV (k) Generation power of PV at performance to gather and process huge amounts of global
v
iteration k. information. In addition, the load prediction would increase
the complexity and computation burden of the centralized
control system.
I. INTRODUCTION In contrast to the centralized solution, distributed solutions
Microgrids consist of various types of distributed energy can reduce the computing burden as all DERs participate
resources (DERs) including fuel cells, wind power, and solar in solving the optimization problem and improve scalabil-
energies [1]. With an increased variety of DERs, dealing with ity through the plug-and-play function [26], [27], [28]. The
uncertainties of demand load and renewable energy sources distributed strategies based on a consensus algorithm have
could be complex in operation and control of the microgrid been presented [29], [30], [31], [32]. However, these strate-
(MG) systems [2]. The main objective of the MG operation gies were developed for the secondary layers to improve the
and control is to optimally operate while maintaining sys- power qualities of MG system such as frequency, voltage,
tem constraints such as power supply balance and output and reactive power regulations. Recently, some decentral-
stabilization [3]. Various optimization methods were used in ized solutions were suggested to solve the operation cost

VOLUME 10, 2022 88799

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

problem [27], [33], [34], [35], [36], [37], [38], [39], [40], operation, have been studied extensively for the microgrid
[41]. In [33], the decentralized economical operation strat- control and operation strategy. Existing centralized strategies
egy was proposed to schedule different types of participants have limitations on the huge computational burden, load
without considering the tripping of participants. A distributed uncertainties, and low reliability whereas the decentralized
method using self-organizing dynamic agents was proposed strategies lack the energy management for long-term opera-
in [34]. The incremental cost consensus algorithm methods tion. In order to overcome the limitations of these centralized
were proposed by selecting the incremental cost of each DER and decentralized strategies, this paper proposes a hybrid
in [35]. The incremental welfare consensus algorithm for multi-agent management system including both centralized
solving an energy dispatch problem was proposed in [27]. EMS and distributed agents. This paper proposed a hybrid
In [36], the distributed economic dispatch algorithm was multi-agent system where both secondary and tertiary layers
suggested, which can learn a mismatch between demand and take part in optimizing operational cost, which distinguishes
total generation power. In [37], [38], and [39], the authors our approach from existing methodologies. The central agent
proposed an optimization method to implement the multi- calculates a power reference considering future intervals
agent system by modeling the DER and load agents. The based on the mismatched power. The distributed secondary
distributed methods based on a multi-agent system including control layer computes the incremental cost through the
the energy storage system (ESS) unit were proposed in [40] mismatched power received from the central agent, derives
and [41]. However, these solutions do not consider the energy the converged incremental cost using a consensus algorithm,
management of ESS agents. The incremental cost of the ESS and transmits the result to the primary controller. Through
agent is determined by the remaining state of charge (SoC). this hierarchical cooperative control structure, it can main-
Thus, the existing distributed system charges and discharges tain the cost optimization for economic dispatch under load
according to the incremental cost of the ESS agent at the uncertainty without load forecasting step. Compared to the
corresponding interval. Without SOC management for future existing distributed methods, the proposed strategy enables
intervals, the distributed control system is required to curtail SOC management of ESS considering the future load infor-
load to maintain the supply and demand power balance. mation. The conventional control strategies cannot achieve
The communication system is utilized for information cost optimization as communication failure occurs. On the
sharing among DERs in the distributed network. In a multi- other hand, the proposed control strategy of this paper can
agent-based distributed control system, the stability of the perform the economic dispatch by bypassing other control
secondary control layer affects the optimization performance. layers when update process of local information for cost
Recently, various control strategies of communication fail- optimization is not possible due to communication failure.
ure have been proposed in distributed control systems [42], Therefore, the hybrid multi-agent cooperative control system
[43], [44], [45]. A resilient distributed control strategy in enables continuous economical optimization through a hier-
which each participant detects and blocks a compromised archical control structure even if communication line failure
communication link and controller was proposed in [42]. occurs. The main contributions of this study are listed as
An event-based frequency control strategy using the dis- follows:
tributed consensus was proposed in [43]. The communication 1) A hybrid multi-agent management system is proposed
topology update mechanism integrates event-based data to to optimally manage the MG system. The proposed
improve the stability of control strategy in untrusted com- strategy has advantage of combining centralized and
munication network status. The MG modeling architecture decentralized approaches to achieve economic dispatch
to improve communication system stability was proposed of MG system.
in [44]. This paper analyzed the effect of communication 2) The proposed strategy continues the optimal operation
failure on the distributed control system. It also performed of the MG system without significant effort in the load
performance verification on the proposed robust communi- forecasting process. It can achieve cost optimization,
cation system. These control strategies are implemented to even in harsher environments than in [36], [37], [38],
achieve the operational objective by blocking the communi- [39], [40], and [41], such as load uncertainty and com-
cation line where an error is detected. Literature [45] pro- munication failures.
posed a distributed control strategy based on resilient double 3) The controller hardware-in-the-loop test platform is
gradient-descent method. This paper responds to DoS attacks proposed to verify the capability of the proposed hybrid
by switching multiple control variables according to the com- multi-agent management system.
munication state. In case of the communication failure, the The rest of this paper is organized as follows: Section II
local information exchanged between agents is corrupted describes the hierarchical control structure of the proposed
and consequently the economic dispatch fails. Thus, it is hybrid multi-agent system, including the incremental cost
important to always ensure a reliable bidirectional flow of consensus algorithm and a centralized EMS unit. Section III
exchanged data even in communication failures. provides the simulation results considering load uncertainty
A hierarchical control approaches in which the secondary and compares them with the centralized EMS method from
control layer is in charge of improving power qualities the real-time test platform. Finally, the main conclusion of
while the tertiary control layer is responsible for economic this study is summarized in Section IV.

88800 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

FIGURE 1. Schematic diagram of the microgrid system. FIGURE 2. Configuration of the hybrid multi-agent system.

II. HYBRID MULTI-AGENT SYSTEM METHODOLOGY

In this study, Fig. 1 shows the MG configuration consisting
of 10 nodes and centralized EMS. The MG system includes
dispatchable agents consisting of four distributed generators
(DG) and one ESS, and non-dispatchable agents consisting
of three photovoltaics (PV) and two loads. Each dispatchable
agent shares its own power information using the distributed
communication line and a centralized EMS sends power set-
point to all dispatchable agents.

A. DESCRIPTION OF AGENTS
1) DISTRIBUTED GENERATOR FIGURE 3. Control strategy of three layer with different sampling times.

It is assumed that there are n-DGs in the hybrid multi-agent

system. The operational cost of such DG is assumed to be the
same as that of the conventional gas or diesel generators [36], according to the remaining SOC in each interval. Besides, the
[37], [38], [39], [40], [41]. The cost function for DG can be battery state of health (SOH) is related to the deterioration
expressed in quadratic function, as given by (1) and (2). The cost, which is affected by the number of operation cycles
power generation output of DG is limited as given in (3). and the depth of discharge (DOD) in each cycle. Thus, the
DOD-based simplified battery cost function including SOH
d ) = αd · (Pd ) + βd · Pd + γd
DG 2
Cd (PDG DG DG DG DG
(1) can be expressed in quadratic function, as given in (4)
∂Cd (PDG
d )
and (5).
d + βd
= 2αdDG · PDG DG
(2)
∂PdDG
e ) = αe (Pe
2
Ce (PESS ESS ESS
+ 3PESS
e,max · (DOD))
PDG ≤ PDG ≤ PDG (3) + βeESS (PESS e,max · (DOD)) + γe
+ 3PESS
d,min d d,max ESS
e

where Cd (PDG (4)

d ) is the quadratic generation cost function of
each distributed generator; PDG ∂Ce (PESS
e )
d is the output power of d-th = 2αeESS (PESS e,max · (DOD)) + βe
+ 3PESS ESS
DG; αdDG , βdDG and γdDG are the quadratic cost coefficients; ∂Pe ESS e

PDG DG
d,min and Pd,max are the lower and upper bounds of the (5)
output power, respectively. The Lagrange multiplier method − PESS ≤ PESS ≤ PESS (6)
e,max e e,max
is used to solve the objective function Cd (PDG d ), thus, the
quadratic cost function given by (1) is rewritten as in (2). where Ce (PESS
e ) is the quadratic generation cost function of
the energy storage system agent; PESS e is the charging or
2) ENERGY STORAGE SYSTEM AGENTS discharging power; PESS e,max is the upper bound; αeESS , βeESS
The ESS agent utilizes a bidirectional capability to absorb or and γe ESS are the coefficients of quadratic generation cost;
compensate power to maintain the demand load balance of ∂Ce (PESS
e )/∂P ESS is the incremental cost, which is defined
e
the system. Battery charging or discharging process operates as the derivative of the quadratic cost function.

VOLUME 10, 2022 88801

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

FIGURE 4. Overall of proposed hybrid multi-agent system.

3) RENEWABLE ENERGY AGENTS control layer (SC) calculates the incremental cost based on
The renewable agents include wind turbine generator (WTG) the mismatched power, that is the deviation power between
and PV. It is difficult to accurately predict the amount of the generation set-point from the EMS and the output of the
power generation due to intermittent output characteristics. agent. Fig. 4 describes the overall control configuration of the
The renewable power source is mainly operated according to proposed hybrid multi-agent system. The controller of each
the maximum power point tracking (MPPT) control method. DER includes the primary control layer based on the voltage
Thus, the renewable agent is considered the negative load in and current control loops and the second control layer based
terms of the MG system. on consensus algorithm. The distributed secondary controller
calculates the mismatched power using the set-point from the
B. CONSENSUS-BASED HYBRID MULTI-AGENT SYSTEM centralized EMS for optimal operation in the future intervals.
The centralized control does not guarantee economical oper- The incremental cost of each DER, which is calculated based
ation due to long control cycles in load uncertainty. The on the mismatched power is converged through a distributed
distributed control including the battery cannot perform SOC communication network. The generation power reference is
management because future loads are unknown. In this paper, sent to the primary layer. In the primary control layer, the
the hybrid multi-agent control strategy, which combines both droop control method is applied to follow the reference power
EMS and distributed control based on consensus algorithm to from the upper layer.
continuously maintain optimal operation in load uncertainty,
is shown in Fig. 2. The proposed multi-agent cooperative 1) PRIMARY CONTROL BASED ON DROOP CONTROL
control strategy enables a tractable design by setting different The primary control layer is aims to maintain power balance
control sampling times for each layer to ensure sufficient and power quality in islanded operation. The inner current
intervals for optimal operation. The control sequence for the and voltage control loops of this layer are shown in Fig. 5(a).
hierarchical control layers is shown in Fig. 3. The primary The proportional integral (PI) controllers are used to regulate
control layer has the shortest control interval (Ts ), as it is the current and voltage. The inner current controller is based
required an immediate response speed to maintain the load on the decoupling of the dq-reference frame, as given by (7)
balance. The secondary control layer operates in control inter- and (8).
val (Tcom ) according to the network communication speed, Z
and the tertiary control layer has the longest control interval Ud∗ = kp1
c
i∗d − id + ki1

c ∗
id − id dt − ωLf iq


(7)
(Tems ) for economical operation. The EMS of the tertiary con-
  Z  
trol layer (TC) determines the generation power considering Uq∗ = kp2
c
i∗q − iq + ki2
c
i∗q − iq dt − ωLf id (8)
the next interval load and send to each agent. The secondary

88802 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

N = {1, 2, . . . , n} and edges ε ∈ N 2 , in which n represents

the number of agents and the edges set ε represents the agent-
to-agent communication link. The adjacency matrix that is
associated with the undirected graph G is A = {ai,j }i,j=1,.···n ∈
Rn×n , as given by (11). If the communication line is normally
connected, it has a value of 1, whereas it has a value of 0 in
case of not connected or is blocked due to a line failure.
(
1 if (i, j) ∈ ε
aij = (11)
/ ε or i = j
0 if (i, j) ∈

Algorithm 1: Consensus Algorithm for Secondary

Controller
1: Initial value
2: Updated weighting matrix W
3: Step 1: Determine average incremental cost value (Cavg )
4: While errorc < acceptable value do
FIGURE 5. Control diagram of primary control layer: (a) inner control for all i < N do
loop; (b) Outer control loop. n
ωij Cj (k) − Ci (k)
P

Ci (k + 1) = Ci (k) +
i∈Ni
where Ud∗ and Uq∗ are the output voltage references; kp1 c and error ci = |Ci (k + 1) − Ci (k)|
c
ki1 are PI control coefficients of the inner current controller end
c and k c are PI control coefficients of the inner
of d axes; kp2 errorc = max{errorci }
i2 end while
current controller of q axes; i∗d and i∗q are current references n
Cavg = 1n
P
of the measured currents id and iq of d and q axes. Ci [0]
The outer control loop is based on the droop controller i=n

for power sharing among multi-agents in the MG system,

as shown in Fig. 5(b). The droop controller computes refer- The Laplacian matrix L = {li,j }i,j=1,.···n ∈ Rn×n , which is
ence for angular frequency and voltage of the inner control a matrix of graph G, as given by (12).
loop as given by (9) and (10).  n
X
ω = ω∗ − kdp P − 1P∗


(9) aij for i = j


lij = i6=j,j=1 (12)
1V = V − kdq Q − Q
∗ ∗ ∗


(10) 

−aij for i 6= j
where ω and 1V ∗ are the angular frequency and amplitude
deviation of voltage, respectively; kdp and kdq are the droop The coefficient of information exchange between adjacent
coefficients; 1P∗ and Q∗ are the deviation of active power agents is based on Metropolis method. In order to adapt to
and reactive power reference, respectively; P and Q are the changes in the system configuration, the Metropolis method
measurements. was presented in [28]. The corresponding weights could be
The deviation of active power reference for power control represented by the updating matrix W = {ωi,j }i,j=1,.···n ∈
of each agent is received from the secondary control layer. Rn×n , which is given by (13).

2) CONSENSUS-BASED SECONDARY CONTROL 
 1/[max(ni + nj ) + 1], j ∈ Ni
 X
The dynamic consensus algorithm is used to obtain power 
ωij = 1 − 1/[max(ni + nj ) + 1], j = i (13)
reference assigned to each agent for optimal operation in 
 j∈Ni
hybrid control system. The purpose of consensus algorithm 
 0, otherwise.
is to design an updated law that can converge the information
status of all agents to a same value. The graph theory is used where ni and nj are the number of agents adjacent to agents i
to represent the communication network of a hybrid multi- and j, respectively.
agent system. The agent can be considered as one of the nodes The agent exchanges local information with neighbors
in a communication system, each node performs an update according to distributed linear iteration at each interval. The
process based on its information and communication with average value of the incremental cost can be found as a result
adjacent nodes at each step. of several iterations. Thus, the secondary control layer should
The communication topology can be presented using maintain equally incremental cost among all agents. The
the undirected graph G = (N , ε) with nodes set dynamic consensus algorithm can be represented at discrete

VOLUME 10, 2022 88803

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

where λDG
d (k) and λe (k) are the incremental cost of DG and
ESS

ESS agents, respectively;

Each agent exchanges the local incremental cost with the
adjacent agent. These following adjacent agents are regulated
their incremental cost according to the distributed average
rule, the next agents sequentially update their information.
The consensus-based distributed convergence algorithm can
be represented by (18).
n
X
λi (k + 1) = λi (k) + ωij (λj (k) − λi (k)) (18)
j∈Ni

where λi (k + 1) is the marginal cost at iteration k + 1;

The distributed power references of DG and ESS agents
can be rewritten as in (19) and (20).
λi (k + 1) − βd,DG
Pdis
d,DG (k + 1) = (19)
2αd,DG
λi (k + 1) − βe,ESS
Pdis
e,ESS (k + 1) = − 3Pmax
e,ESS (DOD)
2αe,ESS
(20)
where Pdis dis
e,DG and Pe,ESS are the distributed power reference
of DG and ESS agents, respectively;
Then, the generation power reference that is determined by
secondary controller can be expressed by (21).
P∗i (k + 1) = Pdis EMS
i (k + 1) − Pi (k) (21)
FIGURE 6. Flowchart of the distributed secondary controller. where Pref (k) is the generation power reference of each agent
at iteration k; ksp and ksi are the proportional and integral
time instant k + 1, as given by (14). terms of secondary PI controller, respectively;
n
X The generation power reference of the consensus-based
xi (k + 1) = xi (k) + ωij (xj (k) − xi (k)) (14) secondary controller is sent to the primary controller of each
j∈Ni DER agent.
where n is the number of agent units in the system; ωij is the
3) CENTRALIZED EMS OF TERTIARY CONTROL
weights updating matrix.
The consensus-based secondary layer performs economic The centralized EMS receives all information from each
operation and the power balance control of a hybrid control agent and determines the set-point to optimize operation cost.
system in uncertainty loads. Fig. 6 presents the secondary It is assumed that the hybrid multi-agent system is operated
control algorithm for hybrid multi-agent cooperative control in island mode. The generation amount and unit cost of DG
proposed in this study. The incremental cost is computed are considered in the objective function since ESS does not
based on mismatch power that is the difference between the occur generation cost. The objective function of centralized
power set-point from an EMS and the measured power. The EMS is to minimize the total operation cost, as given by (22).
mismatch power can be expressed in (15). XX 
min CdDG · PDG
d (k) + y d (k) · Cd
SU
+ zd (k) · Cd
SD

δPi (k) = PEMS

i (k) − Pi (k) (15) d∈D k∈K
(22)
where, δPi (k) is the mismatch power at iteration k; PEMS
i (k)
is the power set-point from the centralized EMS; Pi (k) is the where CdDG and PDG d (k) are the per-unit generation cost and
measured power of i-th agent. power of d-th DG at k, respectively; yd (k) and zd (k) are the
The incremental costs of (2) and (5) can be expressed start-up and shutdown status identifier; CdSU and CdSD are the
as (16) and (17) according to the mismatch power with start-up and shutdown cost. The constraints for each DG units
∂Cd (PDGd )/∂Pd
DG = λDG (k) and ∂C (PESS )/∂PESS =
d e e e
are given as follows
λESS
d (k). ud (k) · PDG DG DG
d,min ≤ Pd (k) ≤ ud (k) · Pd,max (23)
λDG
d (k) = 2αd δPi (k) + βd
DG DG
(16) s.t. yd (k) = max {(ud (k) − ud (k − 1)) , 0} (24)
λESS ESS ESS ESS
e (k) = 2αe (δPi (k)+3Pe,max (DOD))+βe (17) zd (k) = max {(ud (k − 1) − ud (k)) , 0} (25)

88804 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

where ud (k) is the commitment status identifier; PDG

d,min and
TABLE 1. Tested system parameters.

PDG
d,max are the minimum and maximum power generation
amount;
The charging and discharging bounds of the ESS are given
by (26) and (27).
  1
0 ≤ PESS+
e (k) ≤ PESS ESS
e,cap · 1 − SOCe (k − 1) ·
1 − PESS+
e,loss
(26)
 
ESS− ESS ESS ESS−
0 ≤ Pe (k) ≤ Pe,cap · SOCe, (k − 1) · 1 − Pe,loss
TABLE 2. Control sampling time of each layer.
(27)
where PESS+
e (k) and PESS+
e (k) are the charging and discharg-
ing amount of the ESS, respectively; PESS e,cap is the capacity;
ESS+ ESS−
Pe,loss and Pe,loss are the charging and discharging loss of
the ESS. The SOC of the ESS at interval k is constrained
by (28) and (29).
1 PESS−
e (k)
SOCeESS (k) = SOCeESS (k − 1) − · ESS−
The fast convergence speed improves the performance of
PESS
e,cap 1 − Pe,loss hierarchical control system. However, the increase in commu-
1   
nication lines causes economic problems such as installation
− · 1 − PESS+
e,loss · PESS+
e (k)
PESS
e,cap and maintenance costs. Thus, it is required to determine
(28) topology according to the purpose of the system. In order
ESS to verify the feasibility of the proposed control strategy, the
SOCe,min ≤ SOCeESS (k) ≤ ESS
SOCe,max (29)
test platform for hybrid multi-agent system is implemented
ESS and SOC ESS are the lower and upper limits
where SOCe,min using the real-time simulator of OPAL-RT, as shown in Fig. 8.
e,max
for SOC of the ESS, respectively; In addition, the power The host computer and real-time simulator OP5600 of the
balance constraint to maintain the system frequency and the test platform use RT-LAB to model the primary controller of
voltage, as given by (30). each agent, PV agents, and loads, and generate the conversion
X X X code for the real-time simulation. The five computers are
PLoad
l (k) = P DG
d (k) + PESS−
e (k) utilized as the distributed controller of the secondary layer,
l∈L d∈D e∈E and the control PC is used to implement the centralized EMS
X X
ESS+
− Pe (k) + PPV
v (k) (30) of tertiary layer. The agents of distributed control system
e∈E v∈V are connected to the OP5600 and control PC using Modbus
based Ethernet HUB, as shown in Fig. 9. The communication
where PLoad
l (k) is the load at k, and PPV
v (k) is the generation topology consists of ring connection. The centralized EMS
power of PV at k.
sends the set-point to the distributed secondary control layer.
III. REAL-TIME SIMULATION
The tested system consists of three PV units, two loads,
and four inverter-based DER agents, which are three DG
agents and one ESS agent. The tested system parameters are
shown in Table 1. The control sampling time of each layer
is shown in Table 2. The sampling time of the primary layer
is 50 µs to ensure the stability of the converter system. The
secondary layer is run with the sampling time of 0.1 s. The
tertiary layer is designed with the sampling time of 30 s to
ensure that the secondary layer has enough time to converge
toward a common value, ensuring the optimal solution of the
tertiary layer. In this paper, all the tested cases were run for
24 intervals. The time window of each interval is 30s, which
results in a total simulation time of 720s.
The convergence speed according to the communication
topology is shown in Fig. 7. The line topology with few
communication lines has the slowest convergence speed, FIGURE 7. Convergence time comparison according to communication
while the full topology has a fast convergence speed. topology: (a) line; (b) star; (c) ring; (d) full.

VOLUME 10, 2022 88805

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

FIGURE 10. Net load (load uncertainty:0%) and PV output.

FIGURE 8. Real-time simulation test platform setup.

FIGURE 9. Communication network diagram.

Then, the consensus-based distributed controller receives the

measured power and DOD value of ESS from the OP5600
simulator. The optimal control operations that minimize the
generation cost are computed and the power reference is sent
to the OP5600 simulator. The generation costs for the agents FIGURE 11. Output power of each DER: (a) centralized EMS strategy;
(b) proposed control strategy.
corresponding to (1) and (4) are shown in Table 3.

TABLE 3. Generation cost of each agent parameters.

FIGURE 12. Energy storage system state of charge comparison.

A. CASE 1: WITHOUT LOAD UNCERTAINTY
The microgrid net power is being diffused through the multi-
agent system, as shown in Fig. 10. The load profile is similar The output power of the DERs is shown in Fig. 11. The cen-
to the residential load profile, and PV agents are generated tralized method determines the generation power reference
around noon to compensate for a large portion of the load. at every interval, and satisfies the load balance by sending
In this paper, the PV agent operates under maximum power it to each DER agent, as shown in Fig. 11(a). In contrast,
point tracking mode. Thus, it can be considered as a negative the proposed control strategy optimizes power generation
load. The net load is defined as the total load demand minus cost through hybrid control system, as shown in Fig. 11(b).
PV generation. The marginal cost of each agent through the It can be seen that the ESS agent with the lowest cost of
secondary layer converges in about 5.1 s. The sampling time generation operates to supply more power. Fig. 12 shows a
of the tertiary layer should be set longer than the incremental comparison result of SOC according to charging/discharging
cost convergence time through the secondary layer. Thus, the of an energy storage system agent. The SOC of the battery
control sampling time of the tertiary layer is set to 30 s to is charged to the initial value for continuous operation. The
ensure sufficient interval to obtain an optimal solution. marginal cost of all DERs is shown in Fig. 13. It can be

88806 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

FIGURE 13. Marginal cost update based on consensus algorithm. FIGURE 15. Net load (load uncertainty:5%) and PV output.

FIGURE 14. Cost optimization comparison (load uncertainty:0%).

TABLE 4. Generation cost comparison.

FIGURE 16. Output power of each DER: (a) centralized EMS strategy;
(b) proposed control strategy.

observed that marginal costs always converge to minimize

operating costs. The optimization results of the cost func-
tion are compared to the centralized solution, as shown in
Fig. 14. It can be seen that the generation cost of the proposed
control strategy is always cheaper compared to a centralized
EMS strategy. Table 4 presents aggregated generation costs
over 24-intervals period for hybrid multi-agent system. The
proposed control strategy is validated in that the total cost is
5.87% cheaper than centralized strategy. FIGURE 17. Energy storage system state of charge comparison.

B. CASE 2: WITH LOAD UNCERTAINTY

In this case, the performance of the proposed control strategy operating costs for every iteration. Thus, the outputs of all
is verified by considering load uncertainty. Fig. 15 shows the DERs are different depending on the marginal cost.
load profile and the sum of PV outputs. It is assumed that Fig. 17 shows the SOC of an energy storage system agent,
the load uncertainty is 5%. The output power of the DERs is which is performed frequent charging and discharging for
shown in Fig. 16. In case of the centralized EMS strategy, the optimal operation. In the central EMS strategy, there is an
output of DG1 , DG2 , and DG3 is higher due to low generation error between the SOC of the last interval and the initial
cost, as shown in Fig. 16(a). In interval 5, it can be seen that value in the load uncertainty due to the control sampling
the output of DG3 decreases as PV generates. The proposed interval of the tertiary layer, whereas the proposed strategy
control strategy determines the output of all DERs using the recovers the SOC to the initial value to enable continuous
set-point received from the EMS considering the future load operation. Fig. 18 presents the marginal cost of all DERs.
and the power reference through secondary optimization, It can be seen that the marginal cost always converges even
as shown in Fig. 16(b). The hybrid control system optimizes in the load uncertainty. The optimization result of the cost

VOLUME 10, 2022 88807

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

TABLE 5. Total generation cost comparison.

FIGURE 18. Marginal cost update based on consensus algorithm.

TABLE 6. Generation cost comparison.

C. CASE 3: CENTRALIZED EMS FAILURE

In this case, the performance of the proposed control strategy
is verified by considering the case where the central EMS
has failed at load uncertainty of 15%. It is assumed that the
EMS failure occurs at interval 1 and recovers at interval 24,
and all DERs cannot receive the set-point information for
economic operation. The output power of DERs is shown in
Fig. 20. Since the centralized strategy cannot receive set-point
from the EMS, only the droop controller of the primary layer
operates to control the power quality, as shown in Fig. 20(a).
Thus, the outputs of DERs with the same capacity converge
and fail to optimize the operation. On the other hand, the
proposed control strategy blocks the set-point information
received from the tertiary layer in the event of an EMS failure
FIGURE 19. Cost optimization comparison: (a) load uncertainty 5%; and can maintain the economic operation using the distributed
(b) load uncertainty 10%; (c) load uncertainty 15%. communication system of the secondary layer, as shown in
Fig. 20(b). The marginal cost of all DERs is shown in Fig. 21.
It can be observed that marginal cost always converges to
function is compared versus a centralized solution, as shown minimize operating cost even when information is blocked
in Fig. 19. As can be seen, the cost optimization of the from the EMS. The cost optimization comparison result is
proposed strategy is improved. Table 5 presents the total cost compared to the centralized control strategy, as shown in
over 24-intervals period for the load uncertainty. In order to Fig. 22. The generation cost of the proposed strategy is
compare the total cost of the centralized EMS strategy and always lower than that of the centralized control strategy. The
the proposed strategy according to the load uncertainty, load centralized strategy maintains a constant value in the total
uncertainties are assumed to be 0%, 5%, 10%, and 15%. It can generation cost as frequency control is performed during the
be seen that the cost difference is 5.87% compared to the failure of the EMS. Whereas, the proposed control strategy
centralized EMS strategy when the load uncertainty is 0%, always reduces the generation cost compared to the cen-
however, the cost difference increases to 6.04%, 6.52%, and tralized control strategy as optimal operation is performed.
7.23% as the load uncertainty increases. Thus, the proposed Table 6 presents a comparison of total generation cost. The
control strategy has been validated in that the total generation total cost of the proposed control strategy is 7.99% cheaper,
cost is always cheaper than centralized EMS strategy in load and the difference is increased by 0.76% compared to the
uncertainty. result of Case 2 in which EMS operates normally.

88808 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

FIGURE 23. Distributed communication topology state: (a) normal state;

(b) single communication line failure; (c) two communication line failures.

FIGURE 20. Output power of each DER: (a) centralized EMS strategy;
(b) proposed control strategy.

FIGURE 24. Marginal cost update: (a) distributed control strategy;

(b) proposed control strategy.
FIGURE 21. Marginal cost update based on consensus algorithm.

FIGURE 25. Cost optimization comparison.

FIGURE 22. Cost optimization comparison.

D. CASE 4: DISTIBUTED COMMUNICATION LINE FAILURE connection to line connection. Since DG3 agent is isolated
In this case, the impact on cost optimization performance of due to two communication line failures, each information
distributed communication line failure in the secondary layer cannot be exchanged between agents. The conventional dis-
was analyzed at load uncertainty of 15%. The distributed tributed strategy disables the secondary control layer of the
communication topology state according to the line failure isolated agent and only the droop controller is operated,
is shown in Fig. 23. The communication lines between ESS- resulting in a difference of the marginal cost, as shown in
DG4 and DG3 -DG4 are disconnected at intervals 8 and 16 due Fig. 24(a). In contrast, the proposed strategy reduces the
to communication failure. The DG4 operates in an isolated incremental cost by receiving set-point from the EMS for
state according to the two communication failures. cost optimization even when DG4 is isolated, as shown in
The marginal cost of each agent is shown in Fig. 24. Fig. 24(b). Fig. 25 shows the cost optimization comparison
In the event of a single communication line failure between with the distributed control strategy. The proposed strategy
ESS-DG4 , the marginal cost converges in both control can maintain the economic operation even after DG4 is iso-
strategies as the communication topology changes from ring lated, and it can be confirmed that the power generation cost

VOLUME 10, 2022 88809

 W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

TABLE 7. Generation cost comparison. [10] J. Hu, M. Z. Q. Chen, J. Cao, and J. M. Guerrero, ‘‘Coordinated active
power dispatch for a microgrid via distributed lambda iteration,’’ IEEE
J. Emerg. Sel. Topics Circuits Syst., vol. 7, no. 2, pp. 250–261, Jun. 2017.
[11] M. F. Zaman, S. M. Elsayed, T. Ray, and R. A. Sarker, ‘‘Evolutionary
algorithms for dynamic economic dispatch problems,’’ IEEE Trans. Power
Syst., vol. 31, no. 2, pp. 1486–1495, Mar. 2016.
[12] J. Barreiro-Gomez, G. Obando, and N. Quijano, ‘‘Distributed population
dynamics: Optimization and control applications,’’ IEEE Trans. Syst., Man,
Cybern., Syst., vol. 47, no. 2, pp. 304–314, Feb. 2017.
[13] J. Sun, V. Palade, X.-J. Wu, W. Fang, and Z. Wang, ‘‘Solving the power
economic dispatch problem with generator constraints by random drift
particle swarm optimization,’’ IEEE Trans. Ind. Informat., vol. 10, no. 1,
pp. 222–232, Jun. 2014.
[14] A. Pereira-Neto, C. Unsihuay, and O. R. Saavedra, ‘‘Efficient evolutionary
is always reduced compared to the distributed strategy. The strategy optimisation procedure to solve the nonconvex economic dispatch
total generation cost of the proposed strategy is 4.23% lower problem with generator constraints,’’ IEE Proc.-Gener., Transmiss. Dis-
than the results of the distributed strategy, as shown in Table 7. trib., vol. 152, no. 5, pp. 653–660, Sep. 2005.
[15] N. Lu and Y. Zhang, ‘‘Design considerations of a centralized load controller
using thermostatically controlled appliances for continuous regulation
IV. CONCLUSION reserves,’’ IEEE Trans. Smart Grid, vol. 4, no. 2, pp. 914–921, Jun. 2013.
This paper has proposed a hybrid multi-agent management [16] X. Wang, W. J. Lee, H. Huang, R. L. Szabados, D. Y. Wang, and
P. V. Olinda, ‘‘Factors that impact the accuracy of clustering-based load
system including both centralized and decentralized agents. forecasting,’’ IEEE Trans. Ind. Appl., vol. 52, no. 5, pp. 3625–3630,
Since the proposed strategy can integrate the optimal schedul- Sep. 2016.
ing of ESS and the economic dispatch of DERs, it can min- [17] T. Hong, J. Wilson, and J. Xie, ‘‘Long term probabilistic load forecasting
and normalization with hourly information,’’ IEEE Trans. Smart Grid,
imize the generation cost while maintaining load balance vol. 5, no. 1, pp. 456–462, Jan. 2014.
under load uncertainty. The controller hardware-in-the-loop [18] B. Liu, J. Nowotarski, T. Hong, and R. Weron, ‘‘Probabilistic load fore-
simulation has been performed to validate the feasibility casting via quantile regression averaging on sister forecasts,’’ IEEE Trans.
Smart Grid, vol. 8, no. 2, pp. 730–737, Mar. 2017.
of the proposed strategy for the practical application. The [19] J. Xie, T. Hong, T. Laing, and C. Kang, ‘‘On normality assumption in
real-time simulation results showed the superior performance residual simulation for probabilistic load forecasting,’’ IEEE Trans. Smart
of the proposed controller in terms of cost optimization under Grid, vol. 8, no. 3, pp. 1046–1053, Jul. 2017.
load uncertainty and communication failure compared to [20] R. Bo and F. Li, ‘‘Probabilistic LMP forecasting considering load uncer-
tainty,’’ IEEE Trans. Power Syst., vol. 24, no. 3, pp. 1279–1289, Aug. 2009.
the conventional control method. As the proposed controller [21] H. Jiang, K. Wang, Y. Wang, M. Gao, and Y. Zhang, ‘‘Energy big data:
relies on the secondary communication network to exchange A survey,’’ IEEE Access, vol. 4, pp. 3844–3861, 2016.
information, the cyber security of the communication net- [22] J. Wang, X. Zhu, M. Liang, Y. Meng, A. Kling, D. L. Lubkeman, and
N. Lu, ‘‘A data-driven pivot-point-based time-series feeder load disaggre-
work is an important issue to be addressed, which will be gation method,’’ IEEE Trans. Smart Grid, vol. 11, no. 6, pp. 5396–5406,
considered for our future works. Nov. 2020.
[23] B. Stephen, A. J. Mutanen, S. Galloway, G. Burt, and P. Järventausta,
‘‘Enhanced load profiling for residential network customers,’’ IEEE Trans.
REFERENCES Power Deliv., vol. 29, no. 1, pp. 88–96, Feb. 2014.
[1] A. Silani and M. J. Yazdanpanah, ‘‘Distributed optimal microgrid energy [24] Y. Huang, S. Mao, and R. M. Nelms, ‘‘Adaptive electricity scheduling in
management with considering stochastic load,’’ IEEE Trans. Sustain. microgrids,’’ IEEE Trans. Smart Grid, vol. 5, no. 1, pp. 270–281, Jan. 2014.
Energy, vol. 10, no. 2, pp. 729–737, Apr. 2019. [25] W. Liu, W. Gu, W. Sheng, X. Meng, Z. Wu, and W. Chen, ‘‘Decen-
[2] A. Khodaei, S. Bahramirad, and M. Shahidehpour, ‘‘Microgrid planning tralized multi-agent system-based cooperative frequency control for
under uncertainty,’’ IEEE Trans. Power Syst., vol. 30, no. 5, pp. 2417–2425, autonomous microgrids with communication constraints,’’ IEEE Trans.
Sep. 2015. Sustain. Energy, vol. 5, no. 2, pp. 446–456, Apr. 2014.
[26] C. M. Colson and M. H. Nehrir, ‘‘Comprehensive real-time microgrid
[3] Y. Du, H. Tu, and S. Lukic, ‘‘Distributed control strategy to achieve
power management and control with distributed agents,’’ IEEE Trans.
synchronized operation of an islanded MG,’’ IEEE Trans. Smart Grid,
Smart Grid, vol. 4, no. 1, pp. 617–627, Mar. 2013.
vol. 10, no. 4, pp. 4487–4496, Jul. 2019.
[27] S. Yang, S. Tan, and J.-X. Xu, ‘‘Consensus based approach for economic
[4] P. Shamsi, H. Xie, A. Longe, and J.-Y. Joo, ‘‘Economic dispatch for an dispatch problem in a smart grid,’’ IEEE Trans. Power Syst., vol. 28, no. 4,
agent-based community microgrid,’’ IEEE Trans. Smart Grid, vol. 7, no. 5, pp. 4416–4426, Nov. 2013.
pp. 2317–2324, Sep. 2016.
[28] W. Zhang, W. X. Liu, X. Wang, L. M. Liu, and F. Ferrese, ‘‘Online optimal
[5] A. Cherukuri and J. Cortés, ‘‘Distributed generator coordination for ini- generation control based on constrained distributed gradient algorithm,’’
tialization and anytime optimization in economic dispatch,’’ IEEE Trans. IEEE Trans. Power Syst., vol. 30, no. 1, pp. 35–45, Jan. 2015.
Control Netw. Syst., vol. 2, no. 3, pp. 226–237, Sep. 2015. [29] M. Aragüés-Peñalba, T. L. Nguyen, R. Caire, A. Sumper,
[6] X. He, J. Yu, T. Huang, and C. Li, ‘‘Distributed power management for S. Galceran-Arellano, and Q.-T. Tran, ‘‘General form of consensus
dynamic economic dispatch in the multimicrogrids environment,’’ IEEE optimization for distributed OPF in HVAC-VSC-HVDC systems,’’ Int.
Trans. Control Syst. Technol., vol. 27, no. 4, pp. 1651–1658, Jul. 2019. J. Electr. Power Energy Syst., vol. 121, Oct. 2020, Art. no. 106049.
[7] C.-S.-G. Karavas, K. A. Plakas, K. F. Krommydas, A. S. Kurashvili, [30] I. Khan, Z. Li, Y. Xu, and W. Gu, ‘‘Distributed control algorithm for
C. N. Dikaiakos, and G. P. Papaioannou, ‘‘A review of wide-area moni- optimal reactive power control in power grids,’’ Int. J. Electr. Power Energy
toring and damping control systems in Europe,’’ in Proc. IEEE Madrid Syst., vol. 83, pp. 505–513, Dec. 2016.
PowerTech, Jun. 2021, pp. 1–6. [31] I. Khan, Y. Xu, H. Sun, and V. Bhattacharjee, ‘‘Distributed optimal reactive
[8] Z. Tang, D. J. Hill, and T. Liu, ‘‘A novel consensus-based economic power control of power systems,’’ IEEE Access, vol. 6, pp. 7100–7111,
dispatch for microgrids,’’ IEEE Trans. Smart Grid, vol. 9, no. 4, 2018.
pp. 3920–3922, Jul. 2018. [32] I. Khan, Y. Xu, S. Kar, M.-Y. Chow, and V. Bhattacharjee, ‘‘Compres-
[9] A.-H. Mohsenian-Rad and A. Leon-Garcia, ‘‘Optimal residential load sive sensing and morphology singular entropy-based real-time secondary
control with price prediction in real-time electricity pricing environments,’’ voltage control of multiarea power systems,’’ IEEE Trans. Ind. Informat.,
IEEE Trans. Smart Grid, vol. 1, no. 2, pp. 120–133, Sep. 2010. vol. 15, no. 7, pp. 3796–3807, Jul. 2019.

88810 VOLUME 10, 2022

W.-G. Lee et al.: Consensus-Based Hybrid Multiagent Cooperative Control Strategy of Microgrids

[33] V. Loia and A. Vaccaro, ‘‘Decentralized economic dispatch in smart grids THAI-THANH NGUYEN (Member, IEEE)
by self-organizing dynamic agents,’’ IEEE Trans. Syst., Man, Cybern., received the B.S. degree in electrical engineering
Syst., vol. 44, no. 4, pp. 397–408, Apr. 2014. from the Hanoi University of Science and Tech-
[34] Z. Zhang and M.-Y. Chow, ‘‘Convergence analysis of the incremental cost nology, Vietnam, in 2013, and the Ph.D. degree in
consensus algorithm under different communication network topologies electrical engineering from Incheon National Uni-
in a smart grid,’’ IEEE Trans. Power Syst., vol. 27, no. 4, pp. 1761–1768, versity, South Korea, in 2019. From 2019 to 2022,
Nov. 2012. he was a Postdoctoral Researcher and a Research
[35] N. Rahbari-Asr, U. Ojha, Z. Zhang, and M. Y. Chow, ‘‘Incremental wel-
Professor with Incheon National University,
fare consensus algorithm for cooperative distributed generation/demand
South Korea, and a Research Associate at Clarkson
response in smart grid,’’ IEEE Trans. Smart Grid, vol. 5, no. 6,
pp. 2836–2845, Nov. 2014. University, USA. Since April 2022, he has been
[36] Y. S. F. Eddy, H. B. Gooi, and S. X. Chen, ‘‘Multi-agent system for an Engineer Scientist with the Advanced Grid Innovation Laboratory for
distributed management of microgrids,’’ IEEE Trans. Power Syst., vol. 30, Energy (AGILe), New York Power Authority, USA. His research interests
no. 1, pp. 24–34, Jan. 2015. include power system modeling and control, power converter control, and
[37] C.-S. Karavas, G. Kyriakarakos, K. G. Arvanitis, and G. Papadakis, the application of power electronics to power systems.
‘‘A multi-agent decentralized energy management system based on dis-
tributed intelligence for the design and control of autonomous polygen-
eration microgrids,’’ Energy Convers. Manage., vol. 103, pp. 166–179,
Oct. 2015.
[38] C.-S. Karavas, K. Arvanitis, and G. Papadakis, ‘‘A game theory approach
to multi-agent decentralized energy management of autonomous polygen-
eration microgrids,’’ Energies, vol. 10, no. 11, p. 1756, 2017.
[39] W. Zhang, Y. Xu, W. Liu, C. Zang, and H. Yu, ‘‘Distributed online optimal
energy management for smart grids,’’ IEEE Trans. Ind. Informat., vol. 11,
no. 3, pp. 717–727, Jun. 2015. HYEONG-JUN YOO (Member, IEEE) received
[40] T. Zhao and Z. Ding, ‘‘Distributed finite-time optimal resource manage- the Ph.D. degree in electrical engineering from
ment for microgrids based on multi-agent framework,’’ IEEE Trans. Ind. Incheon National University, South Korea, in
Electron., vol. 65, no. 8, pp. 6571–6580, Aug. 2018. 2018. From 2018 to 2019, he was a Postdoc-
[41] R. de Azevedo, M. H. Cintuglu, T. Ma, and O. A. Mohammed, ‘‘Multi toral Research Associate with Incheon National
agent-based optimal microgrid control using fully distributed diffusion University. He is currently working as a Senior
strategy,’’ IEEE Trans. Smart Grid, vol. 8, no. 4, pp. 1997–2008, Jul. 2017. Researcher with the Korea Electrotechnology
[42] B. Huang, Y. Li, F. Zhan, Q. Sun, and H. Zhang, ‘‘A distributed robust Research Institute (KERI). His research interests
economic dispatch strategy for integrated energy system considering include the control of power conversion systems,
cyber-attacks,’’ IEEE Trans. Ind. Informat., vol. 18, no. 2, pp. 880–890,
microgrids, and dc (LV&MVDC) systems.
Feb. 2022.
[43] S. Deng, L. Chen, X. Lu, T. Zheng, and S. Mei, ‘‘Event-based distributed
frequency control in harsh communication conditions,’’ IEEE Trans. Ind.
Informat., vol. 18, no. 6, pp. 3777–3786, Jun. 2022.
[44] S. Rath, D. Pal, P. S. Sharma, and B. K. Panigrahi, ‘‘A cyber-secure
distributed control architecture for autonomous AC microgrid,’’ IEEE Syst.
J., vol. 15, no. 3, pp. 3324–3335, Sep. 2021.
[45] Y. Li, T. Li, H. Zhang, X. Xie, and Q. Sun, ‘‘Distributed resilient double-
gradient-descent based energy management strategy for multi-energy sys-
tem under DoS attacks,’’ IEEE Trans. Netw. Sci. Eng., vol. 9, no. 4, HAK-MAN KIM (Senior Member, IEEE) received
pp. 2301–2316, Jul. 2022, doi: 10.1109/TNSE.2022.3162669.
the Ph.D. degree in electrical engineering from
Sungkyunkwan University, South Korea, in 1998,
WOON-GYU LEE (Member, IEEE) received the and the Ph.D. degree in information sciences
B.S. and M.S. degrees in electrical engineer- from Tohoku University, Japan, in 2011. From
ing from Incheon National University, Incheon, October 1996 to February 2008, he worked with
South Korea, in 2017 and 2019, respectively, the Korea Electrotechnology Research Institute
where he is currently pursuing the Ph.D. degree (KERI), South Korea. He is currently a Profes-
with the Department of Electrical Engineering. His sor with the Department of Electrical Engineer-
research interests include the control of power con- ing, Incheon National University, South Korea. His
version systems, microgrid, multiagent systems, research interests include microgrid operation and control, and dc power
and distributed control systems. systems.

VOLUME 10, 2022 88811