Engineering, 2013, 5, 1-6
doi:10.4236/eng.2013.51b001 Published Online January 2013 (http://www.SciRP.org/journal/eng)
Distributed Intelligent Microgrid Control Using Mul-
ti-Agent Systems
Il-Yop Chung, Cheol-HeeYoo, Sang-J in Oh
School of Electrical Engineering, Kookmin University,Seoul, Republi c of Korea
Received 2013
ABSTRACT
In the f utur e, the individua l entities of microgrids such a s distributed ge nerators a nd smart load s may need to deter mine
their power generation or consumption in more econo mic ways. Intelligent agents can help the decision-making proce-
dure of the entities by intelligent algorithms and state -of-the-art communication with central controller and other local
agents. This paper presents the development of atable-top microgrid control system u sing mult i-agent syste ms and also
the demonstration of demand response programs during power shortage. In our table-top system, agents are imple-
mented using microcontrollers and Zigbee wireless communicatio n technology is applied for efficient data communica-
tion in the multi-agent system. The power system models of distributed generators and loads are implemented in the
real-time simulator using Opal-RT system. The whole test system that includes real-time system simulation and agent
hardware is imple mented in the hardware-in-the-loop simulation framework. The performance of the developed system
is tested for e mergenc y demand response cases.
Keywords: Microgrid; Multi-agent System; Energy Mana geme nt System; Hardware-in-the-Loop Simulation
1. Introduction
Microgrids have recently emerged as a new paradigm for
the future power distr ib ution system that can host mul-
tiple distributed energy resources (DERs) in the local
distribution system levels. Microgrids can also be de-
signed as autonomous independent cells in power grids
because they can control the net power flowing into and
out-of the microgrids to the pre-determined values or
contracts. Therefore, microgrids can improve flexibility
of power grid operation [1-3].
In addition, man y electricit y custo mers need to cut the
cost of their e le ctricity bill by app lyin g s mar t information
technology. Conventional centralized microgrid man-
agement systems have advant ages to effectively manipu-
late the power generation or consumption of the whole
microgrid but it is dif ficult to consider delicate re quest of
individual entities such as DERs and local loads, espe-
cially in terms of economic concerns. On the other hand,
multi-agent systems (MASs) have more interest in indi-
vidual entitie s by nature. T he agent s can obtain informa-
tion by measuring local parameters and communicating
with other agents spontaneously. The agent can make a
decision with artificial intelligence by negotiating and
coope rati ng with o ther agents [4 -8].
This paper presents the implementation of the table-
top system of MAS-based microgrid system. The micro-
grid model contai ns a b attery energy storage system (BESS),
a micro-gas turbine (MGT), and a smart load. Intelligent
agents are implemented using AVR ATmega 128 micro-
controllers. The microgrid simulation model and the
agent hardware are interfacedviahardware-in-the-loop
simulation (HILS) setup. The control signals of DERs or
loads are determined by the corresponding agents. The
control command is delivered from the agents to the
DERs and loads in the simulation via analog and digital
I/O port of the Opal-RT system. The communication be-
tween agentsuses Zigbee wireless communication proto-
col,whichis a low-cost, low-power, and wireless mesh
network standard. Therefore, agents can communicate with
other agents by using either peer-to-peer or one-to-many
communication mode.
When the grid power reserve diminishesquickly, the
grid operator needs emergent load reductionfor stable
operation. This procedure is called as demand response.
One of the popular demand response programs is emer-
gency demand response (EDR) that offers incentives to
the customers who instantly reduce their load [9]. In this
paper, the control objective of the MAS-based microgrid
control is to fi nd the optimal condition for EDR.
Detailed decision-making procedure based on MAS
configuration is presented in this paper and the perform-
ance of the control scheme is verified by the HILS ex-
peri ments .

I.-Y. CHUNG ET AL.
2
2. MAS-Based Microgrid Control
2.1. Microgr id Syst em Mo del
Figure 1 illustrates a single-line diagram of a microgrid
that contains a BESS, a MGT, and a smart load that used
in this paper.
Integratio n of a BESS into microgrids can compensate
instant power mismatch introduced by renewable energy
resources or load variation. Figure 2 illustrates the equiva-
lent mod el the BES S that cont ains a Li-io n battery mod el,
a bi-directional dc-dc converter, and a three-phase inverter.
This paper used the battery model provided by MAT-
LAB/Simulink. The no n-isolated bi-directional boost co n-
verte r has c onti nuous cur rent wa vefor m so the b atter y c an
have long life span. If there is no external control input,
the charging and discharging modes are determined by
the level of the SOC of the battery because the SOC
should be maintained between 30% and 80% during the
normal state. If the SOC is lower than 30%, the BESS
char gin g control s tart s .
The MGTis modeled by equivalent synchronous ge-
nerator and an AC/DC rectifier and DC/AC 3-level po w-
er inverter as illustrated in Figure 3. The rated power of
the MGT is 30kVA and the inverter operates in constant
power control mode. The smart load model employs a
current-controlled converter so that the load operates as a
constant-current load. In this paper assumes that the
smart load consists of controllable load and fixed load.
Figure 1.Configuration of microgrid with MAS-based con-
trol system.
Figure 2.Co nfiguration of BESS.
2.2. Mul ti -Agent System
Figure 1 also shows the concept of MAS-based micro-
grid control system. Each intelligent agent takes charge
of decision-making for a DER or a smart load.
Intelligent agents have reactive, proactive, and social
abilities so that they can react to the environmental
changes, follow the final goal, and interact between other
agents in a cooperative or competitive manner [4]. Ref-
erence [5] explains that agent should have fundamental
modules such as data collection, co mmunicatio n, decisio n-
making, action implementation, and knowledge man-
agement.
Coordination of multiple intelligent agents is an im-
portant issue. In the developed control system, the Mi-
crogrid Central Coordinator (MGCC) coordinates mul-
tiple intelligent agents. When a special control request
arrives from the main grid such as emergency demand
responses, the MGCC informs agents of the control ob-
jectives for the whole microgrid. After receiving the in-
dividualproposal of the agents, the MGCC dispatches the
control command t o the age nts.
The overall decision-making procedure follows the
Contract Net Protocol (CNP) [10].The CNP provides a
formal procedure in the coordination procedure in MAS-
based management systems. The contract between the
MGCC and the agents can be reached by the process of
decision-making and interaction based on two-way
communication. Figure 5 illustrates the concept of the
CNP based decision-making procedure. The overall pro-
cedure starts when the main grid requests for certain ac-
tions such as demand resp onse.
Figure 3. Co nfiguration of MGT .
Figure 4. Functi onal diagram of age nts

I.-Y. CHUNG ET AL.
3
In the CNP procedure, decision-making processes can
be found both in the MGCC and the agent side. Agents
make a decision such as how much it will participate in
the present task requested by the MGCC. T o approach an
optimal solution, the agents evaluate the detailed condi-
tions of the task and check local information such as
generation cost, state-of-charge of a battery, energy
market price, and so on. Agents can use artificial intelli-
gent algorithms such as knowledge-based expert system,
fuzzy systems, or neural networks to attain maximum
benefits from the task. The MGCC decides the overall
operation scheme for a microgrid after receiving the bids
from the agents. If the bids from the agents are not
enough to meet the request from the grid, the MGCC can
modify the task conditions to lead additional participa-
tion from the agents.
3. Hardware-in-the-Loop S imulation
3.1. Real-Time Simulation
The microgrid dynamic simulation model is implemen-
tedusing RT-LABTMsoftware in Opal-RT simulation en-
vironments. RT-LAB is designed to realize the real-time
simulation of Simulink models on clusters of standard
multi-core CPU co mputers. RT -LAB build s parallel tasks
from the original Simulink model and then assigns each
task on one CPU of the multi-core computer so that the
overall simulation can be accelerated. The simulation
time and accuracy of the microgrid can also be improved
by using power system solver and toolboxes of RT-LAB.
Figure 6 shows the real-time simulation model and
Opal-RT s yst em.
3.2. Agent Hardware
There are three agents in our HIL simulation setup: two
for the DERs, MGT and BESS, and one for the smart
load. The DER agents obtain the information such as
current output power, rated power, battery SOC and so
forth and determine the output power reference based on
its own opera tion strategies considering cost, p ower mar-
gins and so on.
Figure 7 shows the hardware of the agent for the
BESS. The agent hardware consists of a main control
board, a Zigbee module, a digital-to-analog converter,
and a LCD display module. The main control board of
the agent uses an AVR ATmega128 microcontroller that
is commonly used in industrial embedded system appli-
cations. The microcontroller collects various data through
built-in devices such as analog-to-digital converters and
serial communication ports of the ATmega128.
The agent can cooperate with other agents and the
central coordinator if necessary. The information between
agents and the coordinator can be transferred through
wireless communication based on ZigbeeTM protocol.
Zigbee is one of the popular solutions for short-distance
wireless personal area networks. Compared to other
communication solutions, Zigbee is advantageous for
short-distance sensor or controller networks such as res-
idential or commercial applications because of security,
low power consumption, and so on. Since there is no
physical line connection between agents, the configura-
tion of Zigbee-based MAS network and the cooperative
operation procedure are flexib le.
3.3. Microgrid Central Coordinator
The MGCC is defined as a central coordinator of a mi-
crogrid. The MGCChas three main functions: 1) to mine
relevant data for microgrid operation from the power
system, 2)to manage multiple agents and monitoring
their status and 3) to coordinate agents to achieve the
goal of the whole microgrid operation.
Figure 5 . Concept of C ontract Net P rotocol between M GCC
and agents.
Figure 6. Real-time simulation model development using
RT-LAB.
Zigbee Module
Input/Output
Conditioning Module
ATmega128
LCD
Module
Input/Output Signal
Figure7. Agent hardware using AVR microcontroller with
Zigbe e mod ule.

I.-Y. CHUNG ET AL.
4
In this paper, the MGCC colle ctsthe real-time data such
aselectricity price and request for emergency demand
response and incentives. When the MGCC receives a
request for partic ip a tio n in demand response or power
quali ty impr ove ment, t he MG CC dete rmi nes whe ther the
microgrid pa rticipates in the r equest or no t. I f the MG C C
decides to join the request, the MGCC communicates
with t he a gen ts in t he micr ogr id to r espo nd to the r eque st
optimally. The detailed protocol in the communication
between the MGCC and the agents follows the Contract
Net Protocol (CNP). Figure 8 shows the screen of the
MGCC GUI program.
3.4. Hardware-in-t he -loop Simulations
Real-time simulation can be a powerful tool for power
system studies. Especially, HILS provides means for the
operation of physical hardware such as power compo-
nents or control hardware while interfaced to a computer
simulation of the system in which the physical hardware
is intended to function. That is, HILS experiments allow
for hardware device to be tested in a true-to-life test con-
dition before the actual system is bu ilt and co mmiss ioned .
It can also mini mize the risk and c ost to e xamine e xtreme
conditions to identify hidden flaws before their impact
manifests in actual operation.
Figure8. Screen capture of MGCC GUI p rogra m.
Figure 9 shows the configuration of the HILS setup
for MAS-based intelligent microgrid control. The power
system model of the microgrid is programmed with
RT-LAB software installed in the host P C. The compiled
simulation model in the host PC is downloaded in the
Opal-RT system through Ethernet connection. Then, the
microgrid model can run in the Opal-RT system in real
time. Figure 10 shows the actual line connections be-
tween the Opal-RT system and the agents. The agents
communicate with other agents and the MGCC via Zig-
beewireless communication. The MGCC is programmed
in a laptop computer. The host PC is used for debugging
the real-time model and monitoring the overall simula-
tion system.
4. Demonstration
In this paper , the M AS -based microgrid control system is
applied to the emergency demand response (EDR) pro-
gram. In the EDR program, the grid operator pays sig-
nificant incentives to the participants who can reduce
their consump tion. For exa mple, i n the US, t he ince ntive
money is abo ut ten times hig her than the elec tricity price
Figure 9. Configuration of HILS setup of MAS-based mi-
crogrid control.
Figure 10. Interface between the Opal-RT and the agents

I.-Y. CHUNG ET AL.
5
during the off-peak period [11]. The individual agents
determine how much the corresponding DER or load will
participate in the EDR program. To this end, the agents
shoulddetermine their optimal values for power genera-
tion or consumption considering given conditions. The
MGCC collects the bids from the agents and decides the
power assignments of the agents by adjusting and coor-
dinating conflicting bids.
As shown in Figure 1, the rated energy of BESS is
80kWh and the rated power of MGT, and smart load are
30kW and 100kW, respectively. Let us assume that the
grid operator requests 40 kWh load reduction with the
incentive as1500 KRW/kWh. The initial operating con-
ditions of the BESS a nd the MGT are as follo ws: theSOC
of the BESS is 80%; the initial values of power genera-
tion of BESS and MGTare 20 kW and 15 kW, respec-
tively. The initial state of smart load is 20kW of critical
load which cannot be reduced.
Now, the MGCC informs the agents of the 40kWh
EDR request with the incentive of 1200 KRW/kWh.
Note that the incentive from the MGCC is 1200
KRW/kWh, which is less than the grid incentive 1500
KRW/ kWh in the begin ning.
The maximum particip ation power of the BESS can be
defined as
(1)
Where Qrated is the rated capacity of the battery (kWh)
that is 80 kWh,S OC is the current SOC; SOCmin is the
minimum constraint of the SOC (%) that is set to 30%,
and PBESS is the initial power output of the BESS (kW),
respectively. Then, the BESS can bid 20 kW for an
hourdue to
The generation cost of gas-turbines can be represented
as a quadratic function. The MGT agent must have the
information about this cost function. The EDR participa-
tion power is determined when the marginal cost is the
same as the incentive. Assume that the participation
power of MGT is 10kW.
The smart loads consist of controllable loads that can
be transferred to different time slots and critical loads
that cannot be shed. If we assume that the initial critical
and controllable loads are 20 kW and 40 kW respectively,
the smart load can bid as much as 40 kW.
Then, in the first round, the MGCC receives the bids
fro m the a gents as much as 6 0 kW in tot al, which mea ns
20 kW from the BESS, 20 kW from the MGT and 40 kW
from the smart load. Then, the total EDR power partici-
pation is larger than 40kW of the request power. The
MGCC dispatch the EDR participation proportionally
according the amount of bids as follows:
Figure 11 illustrates the decis ion-maki ng pr oc edur e of
the MGCC followed by the CNP framework. The MGCC
can match the requested DR po wer with lower price than
the original incentive offered by the grid operator. The
balance money can be shared by the microgrid entities or
can be used for microgrid maintenance cost.
If the sum of the bids from the agents is less than the
EDR request from the grid, the MGCC increase the in-
ternal incentive a bit larger than 1200 KRW/kWh, say
1200 KRW/kWh. Then, agents may consider additional
EDR particip ation. T he MGCC checks if the total sum of
the bi ds fr om age nts matche s wit h the gr id r eq uest. I f yes,
the MGCC dispatch final participation power to the
agents. Otherwise, the MGCC increase the internal in-
centive a bit more until the total sum of the bids exceed
the gri d reques t.
5. Conclusion
This paper presents the development of table-top system
of the MAS-based microgrid control system. The devel-
oped system consists of MGCC and multiple agents for
distributed control of microgrids. This paper elaborates
the details of the hardware development of the agents and
also the software of the MGCC. Emergency demand re-
sponse example has been tested on the overall HILS sys-
tem. The agents are programmed to flexibly talk to the
other agent s and the M GCC via t he CNP and then final l y
find a solution of each unit corresponding to a certain
EDR request for peak shaving. More efficient intelligent
algorithms for optimization and coordination will be de-
veloped for the multi-agents in the f ut ure work.
Figure 11. CNP procedure for dispatching EDR participa-
tion power.

I.-Y. CHUNG ET AL.
6
REFERENCES
[1] R.H. Lasseter, “Control and Design of Microgrid Com-
ponents,” PSERC Final Report, Jan. 2007.
[2] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay,
“Microgrids,” IEEE power & energy magazine, pp.78-94,
Jul./Aug. 2007.
[3] I. Chung, W. Liu, D.Cartes, E. Collins, and S. Moon,
“Control Methods for Multiple Distributed Generators in
a Micro grid System,” IEEE Transactions on Industry Ap-
plications,vol.46, no.3, pp.1078-1088, May/June 2010.
[4] M. Wooldridge, An Introduction to Multiagent Systems,
John Wiley and Sons, 2009.
[5] T Logenthiran, D.Srinivasan, and A.M.Khambadkone,
“Multi-agent system for energy resource scheduling of
integrated microgrids in a distributed system,” Electric
Power Systems Research, vol. 81, no.1, pp.138-148,
2011.
[6] J. Oyarzabal, J. Jimeno, J. Ruela, A. Engler, and C. Hardt,
“Agent based Micro Grid Management System,” 2005
International Conference on Future Power Systems, 18
Nov. 2005.
[7] J.Lago rse, D.Paire, and A.Miraoui, “A multi-agent sys-
tem for energy management of distributed power
sour ces, ”Renewable Energy, vol.35, no.1, pp.174-182,
2010.
[8] H. Kim and T. Kinoshita, “A Multiagent System for Mi-
crogrid Operation in the Grid-connected Mode,” Journal
of Electrical Engineering and Technology, vol. 5, no. 2,
pp. 246-254, 2010.
[9] R. Tyagi and W. Black, “Emergency Demand Response
for Distribution System Contingencies,” IEEE Transmis-
sion and Distribution Conference and Exposition, Apr.
2010.
[10] J. Wu, “Contract Net Protocol for Coordination in Mul-
ti-Agent System,” Proc. of 2nd International Symposium
on Intelligent Information Technology Application,
pp.1052-1058, 2008.
[11] M.H. Albadi and E.F. El-Saadany, “Demand Response in
Electricity Markets: An Overview,” 2007 IEEE Power
Engineering S ociety General Meetin g, 24-28 Jun. 2007.