Electric Power Components and Systems

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

An Intelligent Multi Agent based Approach for

Autonomous Energy Management in a Microgrid

Sujil Areekkara, Rajesh Kumar & Ramesh C Bansal

To cite this article: Sujil Areekkara, Rajesh Kumar & Ramesh C Bansal (2021) An Intelligent
Multi Agent based Approach for Autonomous Energy Management in a Microgrid, Electric Power
Components and Systems, 49:1-2, 18-31, DOI: 10.1080/15325008.2021.1937390

To link to this article: https://doi.org/10.1080/15325008.2021.1937390

Published online: 30 Jun 2021.

Submit your article to this journal

Article views: 52

View related articles

View Crossmark data

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

https://www.tandfonline.com/action/journalInformation?journalCode=uemp20
Electric Power Components and Systems, 49(1-2): 18–31, 2021
# 2021 Taylor & Francis Group, LLC
ISSN: 1532-5008 print / 1532-5016 online
DOI: 10.1080/15325008.2021.1937390

An Intelligent Multi Agent based Approach for

Autonomous Energy Management in a Microgrid
Sujil Areekkara,1 Rajesh Kumar ,2 and Ramesh C Bansal3,4
1
Department of Electrical Engineering, Rajarambapu Institute of Technology, Rajaramnagar, Maharashtra, India
2
Department of Electrical Engineering, Malaviya National Institute of Technology, Jaipur, Rajasthan, India
3
Department of Electrical Engineering, University of Sharjah, Sharjah, United Arab Emirates
4
Department of Electrical, Electronic and Computer Engineering, University of Pretoria, Pretoria, South Africa

CONTENTS
Abstract—The integration of highly intermittent and unpredictable
1. Introduction distributed energy resources increases the complexities in the energy
management system (EMS). Therefore, the conventional control used
2. Multi Agent System (MAS)
in microgrid EMS is no more effective and requires alternative
3. Microgrid Modeling control strategies that can cope up with the new dynamics in the
4. Simulation Results and Analysis system. Adopted from computer science, a distributed artificial
5. Conclusion intelligence technique called multi agent system (MAS) can serve
this purpose well due to its inherent autonomous nature, scalability,
References flexibility, and adaptability. Consequently, in this paper, a new multi
agent based architecture for microgrid energy management system
has been proposed. The proposed model considers different
forecasting agents and a real time correction agent for microgrid
energy management. The performance of proposed MAS based
microgrid EMS modeled in a new platform called Stateflow has
been evaluated by developing different scenarios considering
resource fluctuations and load variations. The simulation results
demonstrate the effectiveness of the proposed architecture in
outperforming the conventional microgrid energy management
system and the effective usage of each component of the microgrid.

Keywords: microgrid, energy management system, multi agent system,
distributed energy resource, unit commitment
Received 2 December 2019; accepted 30 April 2021
Address correspondence to Sujil Areekkara, Department of Electrical
Engineering, Rajarambapu Institute of Technology, Rajaramnagar,
Maharashtra, India. E-mail: sujilavijayan@gmail.com
1. INTRODUCTION
A microgrid (MG) consists of a cluster of loads and dis-
tributed generators serving as a single controllable entity
that can be operated autonomously or in association with
the main grid for providing power to its local area [1]. The
addition of new renewable energy resources increases the
complexity and changes the dynamics of the system due to
highly intermittent and unpredictable nature of distributed
energy resources (DERs). Consequently, the control strat-
egies for energy management that have been used in earlier
decades need to be replenished with modern control para-
digms and control techniques for the smooth operation of
the system [2–4]. The main functions of the energy man-
agement system (EMS) in a microgrid are to ensure the
power balance in the system, provide reliability, flexibility,
and power quality to the system [5].

18
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid
NOMENCLATURE
DSM Demand Side Management
DG Diesel Generator
DER Distributed Energy Resources
EMS Energy Management System
LF Load Forecasting
MG Microgrid
MAS Multi Agent System
Initially, most of the researchers in the field of micro-
grid EMS are concentrated on conventional EMS i.e. cen-
tralized control schemes. In the centralized scheme, a
microgrid central controller is used to optimize the eco-
nomic power dispatch for DERs and storage system [6].
As the number of components in the microgrid increase,
the central controller faces certain concerns like the
requirements of large measurements and it also increases
the complexities in solving the optimization problem.
Moreover, the chance of single point failure is another
major concern in the centralized scheme. This single point
failure in the system can cause complete system failure and
which questions the reliability of the system [6]. Although
the design of a practical centralized MG controller can be
achieved with expert systems and heuristic algorithms, it
does not guarantee the flexibility and scalability of the sys-
tem [7].
When compared to centralized management, the distrib-
uted control for the microgrid energy management system
is less complicated and more vigorous [8]. The components
of distributed management are treated as a module having
enough intelligence to provide the control calculations and
actions. Also, the control modules in the distributed man-
agement help to fulfill the requirements of other modules
as well as reduce the communication and computation bur-
den [9]. Multi Agent System (MAS) has the capabilities
mentioned above to solve the extremely difficult problem
by assigning subtasks within the system in a distributed
manner. So, it is anticipated that a MAS based framework
can help a microgrid to perform robustly in the presence of
extremely intermittent renewable distributed resources and
to meet dynamic load variations. Authors in [10] developed
distributed coordinated control strategies based on a multi
agent system for large power systems.
As evident from the review of recent literature, the
power system researchers are concentrating toward the
implementation of multi agent systems for microgrid
energy management. A detailed review of MAS applica-
tions in the electrical power systems can be found in [11,
12]. In [13], the authors have presented an energy
19
PV Photovoltaic
PRER Portable Renewable Energy Resource
RES Renewable Energy Source
SOC State of Charge
UC Unit Commitment
management scheme for multiple MG by a distributed opti-
mization approach. A daily risk-based optimal scheduling
of reconfigurable smart microgrids in the presence of wind
turbines for microgrid operator profit maximization is pre-
sented in [14]. The authors in [15] developed a distributed
control architecture for AC microgrid with renewable gen-
eration and storage. Energy management and scheduling in
MG by incorporating portable renewable energy resources
(PRER) as an option for demand response have been pro-
posed in [16]. By utilizing the mixed-integer linear pro-
gramming, the optimal scheduling of renewable microgrid
in an isolated load area has been presented in [17]. By con-
sidering communication delays and slow switching topolo-
gies, the authors in [18], presented a robust neighbor-based
distributed control strategy for a DC microgrid. The pro-
posed control can synchronize the voltage of the DC MG
to a reference value and found to be capable of optimal
load sharing that minimizes DER generation cost. In [19],
proposed a distributed control scheme for the effective util-
ization of RES in a microgrid considering noisy disturb-
ance and limited bandwidth constraints. The authors also in
[20], demonstrate the frequency restoration and accurate
load power-sharing problem of ac MG with huge penetra-
tion of distributed generators under time-delay communica-
tion in noisy environments. The authors in [21] introduced
a MAS-based monitoring and control system for integrated
homes/buildings and microgrid in the presence of different
RES and loads.
All the aforementioned researchers conclude that the
MAS based EMS can provide better efficiency and can
also manage complex distributed generators effectively due
to the capability of MAS that can give several key advan-
tages to the system as reported in [22]. Also, the authors in
[23] proposed a MAS based decentralized control for the
energy management of a microgrid. In [24], the authors
suggested that distributed coordination can be implemented
by utilizing a MAS with the distributed energy system and
should be coupled with hardware to find the entire agent
based solution for distributed energy resource management.
Also, the authors have suggested that more intelligent MG
20 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2
FIGURE 1. Interaction of agents in stateflow with micro-
grid in simulink.
operations can be achieved by incorporating load forecast-
ing, Renewable Energy Source (RES) power production
forecasting, resource scheduling and demand side manage-
ment (DSM) into the MG modeling in the future.
The main contributions of this article can be summar-
ized as follows.

This article enables a control scheme that includes the
output power characteristics of different distributed
energy resources and also presents a way for effective
usage of each component in the MG.

Explored a new platform called Stateflow for modeling
and simulating multi agent system for microgrid
energy management system.

The proposed model incorporates the forecasting and
estimation of renewable energy sources and loads as
directed in [24] for the microgrid energy manage-
ment system.
The rest of the paper is organized as follows: Section 2
details the basics of the multi agent system, proposes multi
agent system and different agents deployed for microgrid
management. Section 3 explains microgrid modeling.
Section 4 describes different scenarios simulated with
results and discussions before and after the implementation
of the proposed scheme along with performance evaluation
of the proposed scheme. Section 5 concludes the article.
2. MULTI AGENT SYSTEM (MAS)
The essential ingredients of any distributed system are
autonomous components and coordination. Distributed sys-
tems involving some heterogeneous entities have some lim-
itations like lack of adaptive behavior in runtime and the
costs involved in maintaining continuous communication
for some applications, which have to operate in different
environments. This makes a distributed system with a num-
ber of ongoing interactions nearly infeasible [25]. These
limitations have inspired the development of approaches
toward agent based distributed systems having the
competencies to provide continuous communication and
methods for adapting to dynamic environments.
An agent is a software program or computational entity,
holding the ability to accumulate the information from its
surroundings and perform accordingly without the help of
any human intervention [26]. A group of such several
agents working together to achieve a particular task beyond
the capabilities of a single agent is called Multi Agent
System (MAS). The inherent properties of MAS are scal-
ability, flexibility, autonomy and it can also reduce the
complexity of a problem by breaking a complicated prob-
lem into small, simple sub-problems and solving them in a
distributed manner [27, 28]. The concept of a multi-agent
system, different platforms used to implement MAS and
different multi-agent architecture including centralized, dis-
tributed, and hierarchical can be found in [11].
2.1. Agent Platform
An agent platform is an environment where agents live,
identical to the universe for human beings. The key pur-
pose of an agent platform is to guarantee communication
that is reliable and deliver multiple communication proto-
cols, languages, and communication media. Different
authors have used different platforms namely JADE,
ZEUS, SWARM for implementing a multi agent system in
the power system domain. The platforms used by different
authors in different areas of power systems have been sum-
marized in [11].
In this paper, a new multi agent platform called
Stateflow is introduced to analyze the performance of multi
agent system based microgrid energy management.
Stateflow is an environment that utilizes the concept of
state machines and flow charts for modeling and simulating
the decision logic and controls. The Stateflow enables to
model, how a system responds to particular conditions,
events, and external factors, etc. A state machine represents
a system that is driven by some events, and which can
make the transition from one mode to another mode
according to the changes in the environment. State
machines are the best known and classical method of mod-
eling the behavior of any system [29]. State machine
allows designers to imagine the state of any system at a
particular point of time and also describe the behavior of
the system based on that state. Moreover, this modeling
method is not limited to the software system [30].
Mathematically, a finite state machine can be defined as
[31]:f ðQ, q0 , F,
, dÞ, where
- Finite inputs to the machine,
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid
FIGURE 2. Proposed MAS architecture for microgrid EMS.
Q- Finite set of states, q0 2 Q - Initial states, F  Q -
Final states, and d : Q 
! Q - Transition function.
The agents and their behavior have been modeled using
Stateflow by utilizing the concepts of state machines and
flowchart, and the microgrid is modeled in MATLAB
Simulink in this paper. The integration of agents developed
in the Stateflow and microgrid model developed in
Simulink in MATLAB environment and its interaction is
shown in Figure 1.
2.2. Proposed MAS
A three layer multi agent architecture namely Forecasting and
Estimation Layer, Control and Action Layer, and Real Time
Monitoring Layer is proposed for the microgrid energy man-
agement as illustrated in Figure 2. The different agents with
particular responsibilities have been deployed in three layers
in the proposed architecture for energy management in MG.
In the proposed scheme, the unit commitment agent receives
the forecasted power availability of RES, forecasted load and
State of Charge (SOC) information from the forecasting and
estimation agents. After having received the information from
the respective agents, it takes unit commitment decisions and
sends control signals to the diesel generator agents and bat-
tery agents. Each diesel generator agents in the system also
collect the information from correction agents that are
21
deployed in real-time monitoring layer. After receiving infor-
mation, the DG agents compare both and then give an intelli-
gent decision in a way that maintains the load requirement of
the microgrid. The unit commitment agent in the system
updates its state every hour, whereas the correction agent
updates its values every 5 minutes. The detailed modeling of
the forecasting model used by PV and wind forecasting
agents are explained in [32, 33] and load forecasting agent is
illustrated in [34].

Forecasting and Estimation Layer: This layer is con-
sidered as the top layer in the proposed scheme. The
forecasting and estimation layer consists of seven
types of agents (2  wind Agents, 2  PV Agents,
Load Forecasting (LF) Agent, Unit Commitment (UC)
Agent and a State of Charge (SOC) Agent). The input
to this layer is historical data PV, historical data of
wind, historical load values, and current estimated
SOC information.

Control and Action Layer: This layer is viewed as
the middle layer in the proposed agent architecture. In
this layer, diesel generator and battery agents are
deployed, and the input to this layer is the information
from the bottom layer and the top layer.

Real Time Monitoring Layer: This layer is consid-
ered as the bottom layer in the proposed scheme where
the correction agent is deployed. The inputs to this
layer are real time information from the microgrid.
22 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2

Component Rating system, the unit commitment agent receives the forecasted
output power of PV farm from PV agents, the forecasted
Diesel Generator 1 2.5 MW power output of the wind farm from wind agents and
Diesel Generator 2 1000 kW upcoming load demand information from LF agent and the
Diesel Generator 3 500 kW
Diesel Generator 4 500 kW
current state of charge of the battery from SOC agent.
Wind Farm 1 250 kW After receiving the required information from the wind
Wind Farm 2 250 kW agent, PV agent, LF agent, and SOC agent respectively,
PV Farm 1 500 kW the unit commitment agent takes decisions to activate the
PV Farm 2 500 kW diesel generators to fulfill the load demand of the micro-
Battery 10800 Ah
grid. The decision carried out by the unit commitment
Residential Load 1.14 MW
Commercial Load 0.2 MW agent is transferred to each DG agents. The UC agent pro-
vides unit commitment decision in every one hour span for
TABLE 1. Model parameters.
24 hours. The decision-making process of the unit commit-
ment agent is represented as in Eq.(1) and Eq.(2). where
Preq is the required power to fulfill the load in the micro-
The function of each agents in the system are grid after the complete utilization of renewable distributed
as follows: energy resources. PPVA1 , PPVA2 , PWA1 and PWA2 respect-
ively represent the forecasted output power of the PV
and wind received from the forecasting and estimation
2.2.1. PV Agent. PV Agent forecasts the PV active power
layer. The PDG1max , PDG2max , PDG3max and PDG4max are the
output of the current hour from the historical data and
maximum power generation capacity of the diesel gener-
other heterogeneous factors and gives information to the
ators can be found in Table 1. The DG1on , DG2on , DG3on
unit commitment agent for further decision making.
and DG4on represent the respective diesel generator
which should be activated to fulfill the load requirement
2.2.2. Wind Agent. Wind agent gives the forecasted infor- of the microgrid.
mation of wind active power output of the present hour to
the unit commitment agent. Preq ¼ PLFA  ðPPVA1 þ PPVA2 þ PWA1 þ PWA2 Þ (1)

8
>
> Preq < 0 ; FullfillLoad þ ChargeBattery
>
>
>
> Preq > 0 ; DG1on
<
UCD ¼ Preq  PDG1max > 0 ; DG1on þ DG2on (2)
>
>
>
> Preq  PDG1max  PDG2max > 0 ; DG1on þ DG2on þ DG3on
>
>
:
Preq  PDG1max  PDG2max  PDG3max > 0 ; DG1on þ DG2on þ DG3on þ DG4on

2.2.3. LF Agent. The LF agent is modeled to give the
information of the load forecast for the current hour to the
unit commitment agent.
2.2.4. Soc Agent. The SOC agent will give the current state
of charge (SOC) of the battery to the unit commitment
agent. The battery SOC is calculated as in [35].
2.2.5. Unit Commitment Agent. The function of the unit
commitment agent is to provide the diesel generators dis-
patch strategy in the proposed MG EMS. In the proposed
2.2.6. Correction Agent. The requirement of correction
agent in the system is mainly because taking into account
that the control period is one hour for the unit commitment
decision from the unit commitment agent and that the fore-
cast of the renewable energy sources used (photovoltaic
and wind farms) and residential loads are usually very
inaccurate due to their high volatility. How the power bal-
ance can be assured in the microgrid? Let us consider, for
example, the sudden lack of energy in the microgrid within
the control period, when the diesel generators are not acti-
vated (the forecast did not indicate the necessity to activate
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid 23

FIGURE 3. Agent communication diagram.

any diesel generator) and the batteries become quickly dis- 3

Pwind1 ¼  kVwind1 k  kIwind1 k  cos ð/Vwind1  /Iwind1 Þ
charged. How MAS-EMS can react to this situation when 2
it observes the state of the microgrid only once an hour?, (7)
this situation can be avoided in simulation by introducing 3
Pwind2 ¼  kVwind2 k  kIwind2 k  cos ð/Vwind2  /Iwind2 Þ
the correction agent that updates the status of the real 2
microgrid in every 5 minutes. It is also assumed that the (8)
battery can supply power to satisfy the load during this 5- 3
PLoad ¼  kVLoad k  kILoad k  cos ð/VLoad  /ILoad Þ
minute interval when some sudden requirement of power 2
appears during this interval. (9)
The function of the correction agent in the proposed
3
scheme is to help DG agents to alter the UC decision from PCload ¼  kVCload k  kICload k  cos ð/VCload  /ICload Þ
2
the forecasting and estimation layer. The correction agent
(10)
in the system provides the correction information in every
five minutes to DG agents when and where required. The The necessary diesel generators that should be activated
correction agent estimate the current power availability of to provide the power balance in the system has been calcu-
PV farm 1, PV farm 2, wind farm 1, wind farm 2, residen- lated using Eq.(11).

8
>
> 0; DPreq  0
>
> PDG1 ; DPreq > 0
<
PDG ¼ PDG1 þ PDG2 ; DPreq  PDG1ðmaxÞ > 0 (11)
>
>
>
> P DG1 þ P DG2 þ P DG3 ; DP req  PDG1ðmaxÞ  PDG2ðmaxÞ > 0
:
PDG1 þ PDG2 þ PDG3 þ PDG4 ; DPreq  PDG1ðmaxÞ  PDG2ðmaxÞ  PDG3ðmaxÞ > 0

tial load demand and commercial load demand (Ppv1 , Ppv2 , where, PDG1ðmaxÞ , PDG2ðmaxÞ , PDG3ðmaxÞ , PDG4ðmaxÞ are the
Pwind1 , Pwind2 , PLoad , PCload ) respectively as in Eq.(5-10) and maximum power capacities of diesel generator 1, diesel
then determines the required power DPreq to be produced generator 2, diesel generator 3 and diesel generator 4
to fulfill the power balance in system as in Eq.(3). respectively as represented in Table 1.
The total renewable energy power is calculated as shown
in Eq. (4).
2.2.7. DG Agents. The DG agents in the system receive the
DPreq ¼ PLoad þ PCload  PRes (3) UC decision from the UC agent for the next hour and the
PRes ¼ Ppv1 þ Ppv2 þ Pwind1 þ Pwind2 (4) status of the generator to be activated to fulfill the power
3 requirement of the microgrid in every 5 minutes. After
Ppv1 ¼  kVpv1 k  kIpv1 k  cos ð/Vpv1  /Ipv1 Þ (5)
2 receiving both the information from the unit commitment
3 agent and correction agent, the DG agents recognize if
Ppv2 ¼  kVpv2 k  kIpv2 k  cos ð/Vpv2  /Ipv2 Þ (6)
2 there is any necessity to alter the decision from the unit
24 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2
FIGURE 4. Proposed microgrid MAS-EMS model.
commitment agent. The DG agents send the control signals
to diesel generators (either UC decision or modified UC
decisions accordingly).
2.2.8. Battery Agent. The energy storage system for the
developed microgrid is modeled in such a way that it can
store energy during excess power production from renew-
able sources and is also able to supply to meet load
requirements of the microgrid during a power shortage.
Also, the battery control in the proposed model is con-
strained in such a way that the battery will charge only
when there is excess power production from the distributed
renewable energy resources, not from the diesel generators.
The battery agent function is to take charging and dis-
charging actions accordingly after receiving the decisions
from the unit commitment agent to fulfill the load require-
ment of the microgrid. Figure 3 displays the communica-
tion of different agents during the simulation. Where,
PVA1, PVA2, WA1, WA2, LFA, SOCA, UCA, DGA1,
DGA2, DGA3, DGA2, BA and CA respectively represent
PV 1 agent, PV 2 agent, wind 1 agent, wind 2 agent, LF
agent, SOC agent, unit commitment agent, DG1 agent,
DG2 agent, DG3 agent, DG4 agent, battery agent and cor-
rection agent.
3. MICROGRID MODELING
Figure 4 shows the schematic diagram of the proposed
energy management system for a smart microgrid. The
microgrid in the proposed model has been developed in
Matlab Simulink with phasor-mode simulation. The motiv-
ation for using the phasor-mode simulation is efficiency
and attaining the timely running of the model. For perform-
ing microgrid energy management by considering a
FIGURE 5. Irradiance, wind speed and loads used
for simulation.
24 hours’ simulation scenario, the operation will become
more complicated and time consuming for both, continuous
and discrete modes of simulation. The proposed islanded
microgrid considers four diesel generators, two wind farms,
two photovoltaic arrays, energy storage, and loads. The
diesel generator used in the microgrid provides base power.
The more detailed modeling of the microgrid can be found
in [23]. Table 1 shows the parameters used in the simula-
tion of the microgrid.
4. SIMULATION RESULTS AND ANALYSIS
The simulation of the model shown in Figure 4 was carried
out for 24 hours without the MAS and with the MAS. The
simulation is carried out for only 24 hours due to the limi-
tation of data availability. While considering the case study
without the MAS, it has been considered that all the four
diesel generators are activated and provide power to the
microgrid along with the distributed energy resources. On
the other hand, the case study with MAS considered the
proposed MAS will activate only required diesel generators
to satisfy the power balance in the system by ensuring the
complete utilization of distributed energy resources. The
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid
FIGURE 6. Power output of PV and wind farm in S1.
solar irradiation profile and wind speed used for the simu-
lation study are shown in Figure 5(a) and (b) respectively
[36]. The irradiation profile follows a typical pattern with
the maximum irradiation occurring during the middle of
the day. The wind speed varies with several peaks and
troughs significantly throughout the day. Since the simula-
tion is based on the data collected from the Tioman island
resort, the load follows a distinct profile due to tourist
activities. The load demand of the island at the beginning
of the day is minimum, and it gradually increases and
meets the peak by evening and drops at night as shown in
Figure 5(c). The load profile data used for the simulation is
extracted from [23]. Also, Figure 5(d) shows the commer-
cial load profile used in the simulation case study. For ana-
lyzing the performance of the proposed multi agent system
for energy management, three different scenarios have
been created with some events that generally occur in the
system. The power output of only single PV and wind
farms are shown in each scenario because each PV farm
and wind farm are identical and produce the same power.
Also, the output power produced by all the diesel genera-
tors are shown for all the scenario to satisfy the load
requirements of the microgrid after the complete utilization
of RES.
4.1. Scenario 1
In scenario 1, the model has been simulated without any
disturbances in the system. This scenario is simulated to
analyze the performance of the system during normal oper-
ating conditions. Figure 6(a) and (b) respectively show the
output power generated from PV farm and wind farm in
scenario 1. It is clear from both waveforms that solar
power follows the irradiance pattern represented in Figure
5(a), and since the wind power output is proportional to
25
FIGURE 7. Battery SOC in S1.
the cube of the wind velocity, it trails the pattern of wind
speed represented in Figure 5(b). The load demand of the
system in scenario 1 is represented in Figure 5(c).
Figure 7 shows the state of charge of the battery during
scenario 1. In Figure 7, the battery agent is performing the
battery control action as explained in Section 3.2.8 to sup-
ply the load when there is a shortage of power in the
microgrid system. The battery is not charged throughout
the simulation because the battery controller is constrained
so the battery will charge whenever there is an excess
power produced by renewable distributed energy resources.
The output power produced by diesel generators to meet
the load demand of the microgrid in the absence and pres-
ence of the proposed MAS based energy management sys-
tem separately is shown in Figure 8(a) and (b)
respectively. The channels 1-4 in Figure 8 show the power
produced by all the diesel generators (PDG1-PDG4)
respectively. In Figure 8(b) it can be seen that the diesel
generator 1 is active for 24 hours and diesel generator 2 is
active only during the period 18:00-20:00 hours. The diesel
generator 3 and diesel generator 4 are not active through-
out the simulation. This proves the proposed system can
help to use each component of microgrid effectively. On
the other hand in Figure 8(a), in the absence of the pro-
posed scheme, all the diesel generators are active for the
complete duration of the scenario.
4.2. Scenario 2
In scenario 2, a partial shading effect occurred in the PV
system at 10:30 hours for a duration of 30 minutes and the
condition at which the wind speed exceeds the maximal
wind speed between 21:00  23:00 hours of the simulation
has been considered. This scenario is simulated to study
how the proposed system will react to the power output
variations of distributed energy resources in the microgrid.
Figure 9(a) shows the output power generated by PV
farm in scenario 2. A sudden drop in PV output power,
found at 10:30 hours, for a time span of 30 minutes in
Figure 9(a) can be attributed to the effect of the partial
shading considered in the study. The power output of wind
farm during scenario 2 can be found in Figure 9(b). In
26 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2
FIGURE 8. Power output of diesel generator in S1.
FIGURE 9. Power output of PV and wind farm in S1.
Figure 9(b), the wind power output trails the wind speed
since it follows the cube of the wind velocity. However, in
between 21:00-23:00 hours, the output wind power is zero
because the simulation considers that during this time, the
wind speed exceeds the maximum wind speed and the
wind turbine is isolated from the microgrid and stays off-
line until the wind speed returns to its nominal value. The
load demand of the system used in scenario 2 was repre-
sented in Figure 5(c). Figure 10 shows the state of charge
of the battery in this scenario for the complete 24-hours
duration. In Figure 10 the battery agent can be seen per-
forming the battery control action to supply the load when
there is a power shortage in the microgrid system. The bat-
tery is not charged throughout the simulation because of
the same reason as explained in scenario 1.
The output power produced by diesel generators to meet
the load demand of the microgrid in the presence and in
the absence of the proposed MAS based energy manage-
ment system separately is shown in Figure 11(a) and (b).
The output powers of the four diesel generators (PDG1-
PDG4) are shown by the channels 1-4 in Figure 11,
respectively. In Figure 11(b), it can be seen that the diesel
generator 1 is active for 24 hours and diesel generator 2 is
active only during the 18:00-20:00 hours as in scenario 1.
In addition from 21:00 hours for some duration due to the
effect, the wind speed exceeds the maximum wind speed
as considered in scenario 2 the diesel generator 2 is remain
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid
FIGURE 11. Power output of diesel generator in S2.
FIGURE 12. Power output of diesel generator in S3.
FIGURE 13. Battery SOC in S3.
activated by the proposed multi agent system to fulfill the
requirement of microgrid during the same time. As soon as
the diesel generator 1 and the other distributed energy
27
resources can manage the requirements of microgrid the die-
sel generator 2 is deactivated even during the absence of
wind farm. It supports the fact that the proposed system can
be able to detect the output power variations in the distributed
energy resources and manage the power balance in the sys-
tem by activating the necessary diesel generators if required.
Also, during 10:30 hours the diesel generator 2 is not acti-
vated because during that time the diesel generator 1 and all
other renewable energy resources can together provide power
to nullify the effect of partial shading effect and manage the
power balance in the system. Also, diesel generator 1 sud-
denly starts providing more power at 10:30 hours and
22:00 hours to compensate the reduction in power output of
the PV and wind farms at the same time due to partial shad-
ing effect and crossing of the maximum wind speed respect-
ively within its capacity. The diesel generator 3 and diesel
generator 4 are not active throughout the simulation, showing
that the proposed system can support the effective utilization
of diesel generator. On the other hand, in Figure 11(a), in the
absence of the proposed scheme, all the diesel generators are
active for the complete duration of the scenario.
4.3. Scenario 3
In scenario 3, to show the effect of commercial load in the
system, an asynchronous machine having a power
28 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2
FIGURE 14. Power output of diesel generator in S3.
Power System
Scenario loss in % efficiency in %
S1 3.6971 96.30
S2 4.0512 95.95
S3 3.7093 96.29
TABLE 2. Performance of proposed MG EMS in different
scenario.
consumption of 0.2 MW was introduced in the system at
20:00 hours of simulation using a circuit breaker as repre-
sented in Figure 5(d). This scenario is created to analyze
the performance of microgrid with proposed multi agent
system based energy management system during the sud-
den variation in load. Figure 12(a) and (b) displays the out-
put power generated from the PV farm and the output
power generated by the wind farm respectively in scenario
3. The load profile illustrated in Figure 5(c) (residential
power demand) and Figure 5(d) (demand of commercial
loads) together have been considered in the simulation for
scenario 3. Figure 13 shows the state of charge of the bat-
tery during 24-hours simulation in scenario 3.
The battery control in this scenario also works as
explained in scenario 1 and scenario 2. The output power
produced by diesel generators to meet the load demand of
the microgrid in presence and absence of proposed MAS
based energy management system separately is shown in
Figure 14(a) and Figure 14(b). In Figure 14(b) it can be
clearly seen that the diesel generator 1 is active during
the complete duration of the simulation. Whereas, in the
case of diesel generator 2 which is activated at
18:00 hours and deactivated after a greater duration than
in the scenario 1 because of the commercial load intro-
duced in the scenario at 20 hours, which increases the
total load requirement of the microgrid as compared to
scenario 1. So, here the proposed multi agent system iden-
tifies the variation in load and acts accordingly to provide
the power balance in the system by activating the diesel
generator 2. In this scenario, also the DG 3 and DG 4 are
completely offline during the entire simulation, and DG2
provides power only for some duration. This result indi-
cates that the proposed multi agent scheme can effectively
act to the load changes in the system and utilize the diesel
generator effectively.
4.4. Microgrid Performance Evaluation
The microgrid efficiency is evaluated as expressed in Eq.
(12) and the the percentage of power losses in the micro-
grid system is calculated using Eq.(13).
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid
PLoad þ i  Pbat
gmicrogrid ¼
PDER þ PDG þ j  Pbat
ðPDER þ PDG þ j  Pbat Þ  ðPLoad þ i  Pbat Þ
PLoss ð%Þ ¼
PDER þ PDG þ j  Pbat
 100

i ¼ 1 and j ¼ 0 ; battery is chariging ðloadÞ
i, j ¼
i ¼ 0 and j ¼ 1 ; battery is dischariging ðsourceÞ
Table 2 represents the performance of the proposed MG
multi agent energy management scheme in terms of per-
centage power loss and system efficiency in differ-
ent scenarios.
5. CONCLUSION
An intelligent multi agent system for microgrid energy
management system was developed in this paper. Different
agents namely PV agents, wind agents, LF agent, SOC
agent, unit commitment agents, diesel generator agents,
battery agent, and real time correction agent have been
developed and deployed in the simulation which repre-
sented each component of the microgrid EMS. The micro-
grid was simulated in MATLAB Simulink environment,
and agent behavior was modeled in Simulink Stateflow.
The simulation was carried out by considering the fluctua-
tions in renewable energy distributed generators and load
variations. The effectiveness of the proposed multi agent
based energy management system was tested by comparing
the power losses and microgrid system efficiency in differ-
ent scenarios. The simulation results indicate that the pro-
posed MAS based EMS has better performance as
compared to without the MAS in terms of power loss and
MG system efficiency. Also, the proposed EMS has the
capability to use the required diesel generators to satisfy
the load requirement of the microgrid after the complete
utilization of DERs. In this study, the diesel generators are
allowed to operate up to the maximum capacity to fulfill
the load requirement of the microgrid. In the future, meth-
ods to find the optimum operating point for each diesel
generator to provide maximum environmental benefits can
be incorporated. Moreover, the proposed system can be
used for any microgrid and large power system by add-
ition/deletion of agents with their responsibilities due to the
scalability of the MAS.
29
FUNDING
(12)
The authors gratefully acknowledge the support of the
Ministry of Electronics and Information Technology,
Government of India for funding this project sponsored by
MoCIT and monitored by Media Lab Asia.
(13)
ORCID
Rajesh Kumar http://orcid.org/0000-0002-6019-0702
(14)
REFERENCES
[1] Y. Zoka, H. Sasaki, N. Yorino, K. Kawahara and C. Liu,
“An interaction problem of distributed generators installed
in a microgrid,” in Proceedings of the IEEE International
Conference on Electric Utility Deregulation, Restructring
and Power Technologies, vol. 2, pp. 795–799, 2004.
[2] B. Kroposki, R. Lasseter, T. Ise, S. Morozumi, S.
Papathanassiou and N. Hatziargyriou, “Making microgrids
work,” IEEE Power Energy Mag., vol. 6, no. 3, pp. 40–53,
2008. DOI: 10.1109/MPE.2008.918718.
[3] S. A, R. Kumar and R. C. Bansal, “Multi agent based
autonomous energy management system with self-healing
capabilities for a microgrid,” IEEE Trans. Ind. Inf., vol. 15,
no. 12, pp. 6280–6290, 2019. DOI: 10.1109/TII.2018.
2889692.
[4] A. Sujil and R. Kumar, “Multi agent based energy manage-
ment system for smart microgrid,” in Recent Developments
in Control, Automation & Power Engineering (RDCAPE).
IEEE, 2017, pp. 125–130.
[5] T. Adefarati and R. C. Bansal, “Reliability, economic and
environmental analysis of a microgrid system in the presence
of renewable energy resources,” Appl. Energy, vol. 236, pp.
1089–1114, 2019. DOI: 10.1016/j.apenergy.2018.12.050.
[6] J. Zhao, C. Wang, B. Zhao, F. Lin, Q. Zhou and Y. Wang,
“A review of active management for distribution networks:
Current status and future development trends,” Electr.
Power Compon. Syst., vol. 42, no. 3-4, pp. 280–293, 2014.
DOI: 10.1080/15325008.2013.862325.
[7] K. Tan, P. So, Y. Chu and M. Chen, “Coordinated control and
energy management of distributed generation inverters in a
microgrid,” IEEE Trans. Power Delivery, vol. 28, no. 2, pp.
704–713, 2013. DOI: 10.1109/TPWRD.2013.2242495.
[8] C.-S. Karavas, G. Kyriakarakos, K. G. Arvanitis and G.
Papadakis, “A multi-agent decentralized energy management
system based on distributed intelligence for the design and
control of autonomous polygeneration microgrids,” Energy
Conversion Manag., vol. 103, pp. 166–179, 2015. DOI: 10.
1016/j.enconman.2015.06.021.
[9] Z. Jun, L. Junfeng, W. Jie and H. Ngan, “A multi-agent
solution to energy management in hybrid renewable energy
generation system,” Renew. Energy, vol. 36, no. 5, pp.
1352–1363, 2011. DOI: 10.1016/j.renene.2010.11.032.
30 Electric Power Components and Systems, Vol. 49 (2021), No. 1-2
[10] C.-X. Dou, D.-W. Hao, B. Jin, W.-Q. Wang and N. An,
“Multi-agent-system-based decentralized coordinated control
for large power systems,” Int. J. Elect. Power Energy Syst.,
vol. 58, pp. 130–139, 2014. DOI: 10.1016/j.ijepes.2014.01.012.
[11] A. Sujil, J. Verma and R. Kumar, “Multi agent system:
Concepts, platforms and applications in power systems,”
Artif. Intell. Rev., vol. 49, no. 2, pp. 153–182, 2018. DOI:
10.1007/s10462-016-9520-8.
[12] S. D. McArthur, et al., “Multi-agent systems for power engin-
eering applications part i: Concepts, approaches, and technical
challenges,” IEEE Trans. Power Syst., vol. 22, no. 4, pp.
1743–1752, 2007. DOI: 10.1109/TPWRS.2007.908471.
[13] T. Liu, X. Tan, B. Sun, Y. Wu and D. H. Tsang, “Energy
management of cooperative microgrids: A distributed opti-
mization approach,” Int. J. Elect. Power Energy Syst., vol.
96, pp. 335–346, 2018. DOI: 10.1016/j.ijepes.2017.10.021.
[14] M. Hemmati, B. Mohammadi-Ivatloo, S. Ghasemzadeh and
E. Reihani, “Risk-based optimal scheduling of reconfigur-
able smart renewable energy based microgrids,” Int. J.
Elect. Power Energy Syst., vol. 101, pp. 415–428, 2018.
DOI: 10.1016/j.ijepes.2018.04.005.
[15] Y. Gao and Q. Ai, “Distributed cooperative optimal control
architecture for ac microgrid with renewable generation and
storage,” Int. J. Elect. Power Energy Syst., vol. 96, pp.
324–334, 2018. DOI: 10.1016/j.ijepes.2017.10.007.
[16] V. S. Tabar, M. A. Jirdehi and R. Hemmati, “Energy man-
agement in microgrid based on the multi objective stochastic
programming incorporating portable renewable energy
resource as demand response option,” Energy, vol. 118, pp.
827–839, 2017. DOI: 10.1016/j.energy.2016.10.113.
[17] H. Morais, P. Kadar, P. Faria, Z. A. Vale and H. Khodr,
“Optimal scheduling of a renewable micro-grid in an iso-
lated load area using mixed-integer linear programming,”
Renew. Energy, vol. 35, no. 1, pp. 151–156, 2010. DOI: 10.
1016/j.renene.2009.02.031.
[18] J. Lai, X. Lu, X. Yu, A. Monti and H. Zhou, “Distributed
voltage regulation for cyber-physical microgrids with cou-
pling delays and slow switching topologies,” IEEE Trans.
Syst. Man Cybern., Syst., vol. 50, no. 1, pp. 100–110, 2020.
DOI: 10.1109/TSMC.2019.2924612.
[19] J. Lai and X. Lu, “Nonlinear mean-square power sharing
control for AC microgrids under distributed event
detection,” IEEE Trans. Ind. Inf., vol. 17, no. 1, pp.
219–229, Jan. 2021. DOI: 10.1109/TII.2020.2969458.
[20] J. Lai, X. Lu and X. Yu, “Stochastic distributed frequency
and load sharing control for microgrids with communication
delays,” IEEE Syst. J., vol. 13, no. 4, pp. 4269–4280, 2019.
DOI: 10.1109/JSYST.2019.2901711.
[21] A. Anvari-Moghaddam, A. Rahimi-Kian, M. S. Mirian and
J. M. Guerrero, “A multi-agent based energy management
solution for integrated buildings and microgrid system,”
Appl. Energy, vol. 203, pp. 41–56, 2017. DOI: 10.1016/j.
apenergy.2017.06.007.
[22] C.-X. Dou, W.-Q. Wang, D.-W. Hao and X.-B. Li, “MAS-
based solution to energy management strategy of distributed
generation system,” Int. J. Elect. Power Energy Syst., vol.
69, pp. 354–366, 2015. DOI: 10.1016/j.ijepes.2015.01.026.
[23] M. R. B. Khan, R. Jidin and J. Pasupuleti, “Multi-agent
based distributed control architecture for microgrid energy
management and optimization,” Energy Conversion Manag.,
vol. 112, pp. 288–307, 2016. DOI: 10.1016/j.enconman.
2016.01.011.
[24] T. Logenthiran, D. Srinivasan and D. Wong, “Multi-agent
coordination for DER in Microgrid, in: IEEE International
Conference on Sustainable Energy Technologies,” in
ICSET, IEEE, 2008, pp. 77–82.
[25] L. Raju, R. Milton and S. Mahadevan, “MAS based model-
ing and implementation of dynamic energy management of
smart microgrid using macsimjx,” Scientific World J., vol.
2016, pp. 1–14, 2016. DOI: 10.1155/2016/9858101.
[26] A. Sujil, S. K. Agarwal and R. Kumar, “Centralized multi-
agent implementation for securing critical loads in PV based
microgrid,” J. Mod. Power Syst. Clean Energy, vol. 2, no.
1, pp. 77–86, 2014. DOI: 10.1007/s40565-014-0047-1.
[27] A. Sujil, S. Choudhary, J. Verma and R. Kumar, “Agent
based intelligent fault detection, isolation and restoration in
smart power system,” in IEEE Students’ Conference on
Electrical, Electronics and Computer Science (SCEECS),
IEEE, 2016, pp. 1–6.
[28] A. Sujil, S. K. Agarwal, R. Kumar, Feasibility study of cen-
tralized multi-agent self-healing power system with super-
conducting fault current limiter,” Technol. Econ. Smart
Grids Sustain. Energy, vol. 1, no. 1, pp. 1–12, 2016. DOI:
10.1007/s40866-015-0002-x.
[29] F. J. Sequeira, A. A. Mallya, S. Hebbar, A. Shetty and D.
Divakar, Design of 4 bit 4 pulse finite state machine serial
adder, in Third IEEE International Conference on Advances
in Electrical, Electronics, Information, Communication and
Bio-Informatics (AEEICB), 2017., pp. 18, 20.
[30] D. R. Wright, Finite state machines. 2005. URL http://www4.
ncsu.edu/drwrigh3/docs/courses/csc216/fsm-notes.pdf
[31] S. Masina, K. Y. Lee and R. Garduno-Ramirez, “An archi-
tecture of multi-agent system applied to fossil-fuel power
unit,” in IEEE Power Engineering Society General Meeting,
IEEE, 2004, pp. 1982–1988.
[32] J.-S. Jang, “ANFIS: Adaptive-network-based fuzzy inference
system,” IEEE Trans. Syst., Man, Cybern., vol. 23, no. 3,
pp. 665–685, 1993. DOI: 10.1109/21.256541.
[33] A. Sujil, R. Kumar and R. C. Bansal, “FCM clustering-
ANFIS-based PV and wind generation forecasting agent for
energy management in a smart microgrid,” J. Eng., vol.
2019, no. 18, pp. 4852–4857, 2019. DOI: 10.1049/joe.2018.
9323.
[34] C. Holmes and B. Mallick, “Bayesian regression with multi-
variate linear splines,” J R. Statist. Soc. B, vol. 63, no. 1,
pp. 3–17, 2001. DOI: 10.1111/1467-9868.00272.
[35] A. Sujil and R. Kumar, “Smart micro grid test system for
agent based energy management system,” IEEE Seventh
India International Conference on Power Electronics
(IICPE-2016), 2016.
[36] Wind, PV data, https://www.renewables.ninja, Accessed:
April 26, 2017.
 Areekkara et al.: An Intelligent Multi Agent based Approach for Autonomous Energy Management in a Microgrid 31

BIOGRAPHIES Access, IEEE ITeN, Swarm and Evolutionary

Computation, Elsevier, IET Renewable and Power
Sujil Areekkara received his Masters and PhD Systems
Generation, IET Power Electronics, International Journal of
from the Malaviya National Institute of Technology
Bio Inspired Computing, and Deputy Editor-in-Chief,
(MNIT), Jaipur, India in 2013 and 2019 respectively.
CAAI Transactions on Intelligence Technology, IET. Dr.
Currently, he has been working as an assistant
Kumar is Fellow of IET (UK), Fellow of IE (INDIA),
Professor at Department of Electrical Engineering,
Fellow of IETE, Life Member of CSI, Senior Member of
Rajaramnagar, Islampur, Sangli, Maharashtra, India. In
IEANG and Life Member of ISTE. His research interests
2014, he received POSOCO POWR SYSTEM AWARD
focus on intelligent systems, machine intelligence, power
(PPSA-2014) jointly organized by power Grid
management and robotics.
Corporation LTD and IIT-Delhi for Best M-tech thesis
all over India. He is a member of International
Ramesh C Bansal has more than 25 years of teach-
Association of Computer Science and Information
ing, research, academic leadership, and industrial
Technology, Singapore and has been associated with
experience. Currently, he is a Professor in EE
various scientific societies. As of 2021, Sujil has 5
Department, University of Sharjah (UAE). In previous
years of experience in teaching in the Electrical domain.
postings, he was Professor and Group head (Power) in
Dr Sujil has published more than 40 research papers in
EEC Department, University of Pretoria (South Africa)
Journals and Conferences of International and National
and worked with the University of Queensland
of repute and has supervised 4 M. Tech. thesis. His
(Australia); University of the South Pacific (Fiji);
research interests include power systems, smart grid
BITS Pilani (India); and Civil Construction Wing, All
applications, multi-agent systems, Artificial Intelligence
India Radio. Prof. Bansal has published over 325 jour-
and machine learning applications to a power system.
nal articles, conference papers, books, books chapters.
He has Google citations of 11000 and h-index 50. He
Rajesh Kumar has more than 20 years of teaching,
has supervised 25 PhD and 4 Post Docs. He is an
research and academic leadership experience. Currently, he
Editor/AE of IEEE Systems Journal, IET-RPG, Tech.
has been working as Professor with the Department of
Eco. Smart Grids, and Sust. Energy, and Electric
Electrical Engineering, MNIT, Jaipur. Dr. Kumar has pub-
Power Comp. and Systems. He is a Fellow, and CP
lished over 500 research articles, has supervised 20 Ph.D.
Engg IET-UK, Fellow Institution of Engineers (India).
and more than 30 M. Tech. thesis. He has 12 patents to his
His research interests include renewable energy, power
name and received 53 awards. He is on 12 Journal
systems, and smart grid.
Editorial Boards. He has been Associate Editor of IEEE