Content uploaded by Mansour Selseleh Jonban
Author content
All content in this area was uploaded by Mansour Selseleh Jonban on Mar 27, 2024
Content may be subject to copyright.
Autonomous Energy Management System with
Self-Healing Capabilities for green buildings (Microgrids)
Mansour Selseleh Jonbana, Luis Romerala, Adel Akbarimajdb, Zunaib Alic,
Seyedeh Samaneh Ghazimirsaeidd, Mousa Marzbandc,e, Ghanim Putrusc
aMCIA Center, Electronic Engineering Department, Universitat Politecnica de Catalunya, Terrassa, Spain
bElectrical Engineering Department, Faculty of Engineering, University of Mohaghegh Ardabili, Ardabil,
Iran
cNorthumbria University, Electrical Power and Control Systems Research Group, Ellison Place NE1 8ST,
Newcastle upon Tyne, United Kingdom
dSchool of the Built Environment, 4th Floor, Maxwell Building Room 712 (THINKlab), University of
Salford, Salford, M5 4WT, United kingdom
eCenter of research excellence in renewable energy and power systems, King Abdulaziz University, Jeddah,
Saudi Arabia
Abstract
Nowadays, distributed energy sources are widely used to supply demand in micro
grids (MGs) especially in the green buildings. Despite increasing number of inter-
mittent distributed energy resources, MGs have essential influence on decreasing
use of conventional generation, but it give rise to new challenges in terms of en-
ergy management, stability and reliability of system. Multi-agent systems (MASs)
as distributed smart units have been widely used recently. However, control and im-
plementation of these smart structures enhance existing challenges because these
units need a framework that, at first, guarantees operation of MASs and then in-
sures management in generation and demand side. This paper presents a smart
control and energy management of a DC microgrid that split the demand among
several generators. An energy management system (EMS) based on multi-agent
system (MAS) controller is developed to manage energy, control the voltage and
create balance between supply and demand in the system with the aim of support-
ing reliability characteristic. In the proposed approach, a self-healing hierarchical
algorithm is implemented to control interaction of agents and also guarantees re-
Email address: mansour.selseleh.jonban@upc.edu Corresponding author (Mansour
Selseleh Jonban)
Preprint submitted to Journal of Building Engineering September 8, 2020
liability of smart control system under faults. Theoretical analysis and simulation
results for a practical model demonstrate that the proposed technique provides a
robust and stable control of a microgrid.
Keywords: Energy management system, multi-agent system, self-healing,
subsumption architecture, microgrid, green building.
2
Nomenclature
Acronyms
CANopen Controller Area Network-open
CEMS Centralized energy management system
DEMS Distributed energy management system
EMS Energy management system
ESS Energy storage system
MAS Multi-agent system
PV Photovoltaic
RES Renewable energy system
SC Super-capacitor
Indices
∆V Maximum oscillation voltage at the DC bus
ηEfficiency of converter
tcController’s delay
tdThe time delay for the power electronic devices
tmDelay in communication port
tpDelay in processing of agent
tsDelay in measurement sensor
UDuty cycle
Parameters
∆t Total delay in the system
iBat Injected current of battery
ich
bat,idis
bat Charging/discharging current of battery
igrid,ipv ,isc Injected current of grid/PV/SC
ImMeasured current
iPV Injected current of PV
isc Injected current of SC
ich
sc ,idis
sc Charging/Discharging current of SC
itot Total injected current to the DC bus
L Inductance
mBinary variable, m=1 boostmode, m=0 Buck mode
PBat,Pgrid ,
Ppv,PSC
Injected Power of battery/grid/PV/SC
Pch Charging power
Pdis Discharging power
3
RResistance
VBat,Vpv ,Vsc,
Vbus,Vgrid
Output voltage of battery/PV/SC /DC bus/grid
VmMeasured voltage
Vnom Nominal voltage of DC bus
Variables
Cbus Capacitor of DC bus
i0
Bat Injected current by battery converter
i0
grid Injected current by grid converter
i0
PV Injected current by PV converter
Iref Current reference (current set point)
i0
sc Injected current by SC converter
Vref Voltage reference (voltage set point)
1. Introduction
Utilization of renewable energy systems (RES) such as wind and solar power is
increasing due to their environmental friendly and cost-effective operation. How-
ever, intermittent generation from RES and unpredictable changes in demand pro-
files cause concerns regarding supply-demand balance and the stability and relia-
bility of power systems. For example, in a building with engaged renewable en-
ergy resources that could be considered as a micro grid, if additional infrastructure
added; voltage stability and reliability of system could be significant as well as sup-
porting supply-demand balance [1, 2]. Energy Storage systems (ESSs) could be
used in such system to guarantee these features [3]. However, by aggregating ESS
(usually of small scale) in a micro-grid, power control and energy management be-
come more challenging. MASs are introduced to solve such control problems in
EMSs, where individual system components are distributed and control is decen-
tralized and autonomous [4–7]. Each component acts as an agent and operates in
a dynamic environment, where it has the freedom to either join or leave the system
whenever required. The agents in MASs are able to perceive environmental changes
4
and decide based on their own decision structure and change the environment with
proper actions [8]. Consequently, the overall challenge and objective in the EMS is
divided into a set of small tasks that each part could be controlled separately [9].
It is important to mention that, by autonomous and decentralized control, the re-
sponse time and control delays are minimized, thus increasing the reliability of grid
connected RES [10].
The energy management system for distributed energy sources can be divided
into centralized energy management system (CEMS) and distributed energy man-
agement system (DEMS) [11–13]. In both cases, the aim is to achieve a balance
between supply and demand; the difference lies in the control architecture and
decision type [14–18]. In CEMSs, associated controllable devices are directly con-
nected to a central control unit, so that it efficiently monitors the whole system
taking into consideration the various factors such as the energy balance and cost
functions [19–24]. The control unit receives measured variable and sends suitable
control signals based on certain restrictions and set points [25–28]. In the DEMS,
on the other hand, system deals with different decision makers and use distributed
processing and control units for regulating the entire system [29]. Each part in the
DEMS has its own decision unit and based on a specific predefined control structure,
it participates in the process of EMS [30–32]. In order to implement the DEMS, the
MAS can be used to provide maximum independence to the individual energy sys-
tems. In fact, each decision part, acting as an agent, communicates and offers a ro-
bust control over the distributed energy sources [33]. This type of control presents
many benefits as compared to the CEMS, such as, the scalability and redundancy
[34]. The following shortcomings have been recognized in previous studies relating
to energy management:
•Non-existence of an online optimum algorithm for minimizing cost func-
tion and maximizing benefits to the whole power system in the smart
buildings [35, 36]. In fact most of presented approaches in this field
rely on scheduling over day-ahead basis [37, 38].
•Absence of an approach for controlling smart buildings in a scenario
when failure appears in the decision part. Generally, smart building
5
control is focused on generation and demand side without considering
any failure in the smart control units [37–39]. However, the stability and
reliable operation in smart buildings mainly depend on the performance
of decision parts, and thus, require attention.
•In the online smart control of green buildings, EMS with self-healing
capability has not been investigated to enable control of devices and, in
addition, the communication is used to only send signal to the devices
and is not used when a collapse appears [40, 41].
In this paper, an active load is supplied by multi-type distributed generators that
are connected to the grid. Battery and super-capacitor (SC) are considered as ESSs
to store extra energy in the microgrid. In order to control and management energy
in the microgrid, MAS will be implemented over the system wherein any generator
will be considered as an agent which can make its own decision based on different
conditions. In this study, a hierarchical algorithm which is modified subsumption
architecture is used as a framework to control and allocate tasks among agents in
the system. In designed framework, any agent will be located separately into layers
and operates based on system’s situation by taking into account its own constrain.
Lower layer has high priority and can dismiss priority of its higher layer. In order
to have cooperation in the system, agents have interact through a CAN bus.
In this research, the aim is to fix the voltage of DC bus and supply demand. At
any time, only one agent controls the voltage in system and shares current based
on its own constrain. This request is applied by a shared signal through the com-
munication port. In most previous research (mentioned earlier), the objective is to
enable energy management in a system without considering any fault in the man-
agement unit [37, 38]. This may threaten the reliability and stability of smart grids.
In this paper, a new energy management system is developed and applied to a sys-
tem similar to that presented in [42, 43]and results are compared especially when
decision part faced with failure. Results show that the proposed approach is more
fault tolerant as compared to the presented method in [42, 43], where the whole
system would collapse if CEMS faced a failure or agent carries token fails, respec-
tively. In the proposed mechanism, when the agent is in normal operation, sets in a
6
layer of subsumption architecture and by sending 5 bits signal via the communica-
tion bus, clarifies its cooperation in the system. When it faced with failure, its signal
will not be received by others and hierarchical algorithm reconfigured to heal its
operation. In order to demonstrate the novelty of proposed approach, contributions
of our work are compared with similar works from the literature, as given in Table
1. The main contribution of this research can be summarized as follows:
•An online smart energy management framework without any pre-processing
and scheduling day-ahead is implemented to allocate task in the MAS.
The proposed method monitors online the situation of system and sends
suitable set points to the converters.
•In order to enhance the reliability of smart control, a self-healing algo-
rithm is used to control and manage interaction of agents. The proposed
algorithm is auto reconfigurable in case a failure appears in the agent. In
fact, it is the advantage of the proposed approach that will be discussed
and compared.
•In order to implement self-healing algorithm, subsumption architecture,
designed by Brooks [44], is modified and a low bandwidth communica-
tion is implemented to overcome rigid separation among the layers.
The rest of paper is organized as follows. In section 2, the structure of the
proposed system and control are explained. The energy management and commu-
nication mechanisms are discussed in section 3. Finally, results of simulation and
conclusions are presented in section 4 and 5, respectively.
2. Description, modelling and control structure of the micro-grid
This section presents the description and details for the micro-grid model con-
sidered for our study.
2.1. Electrical Network
A simple model of DC micro-grid is depicted in Figure 1, which is similar to the
system in [42, 43]. This electrical system contains a Photovoltaic (PV), battery, SC
and gird those are supplying a DC active load through a DC bus.
7
Table 1: Comparison of novelty and contributions of the proposed model with similar works
Ref Online MAS Scheduling Emission Power Commu- Self-
decision day-ahead sharing nication healing
[13,17,18]×Ø Ø Ø Ø × ×
[14]Ø Ø ×Ø×Ø×
[15,32]Ø× × Ø×Ø×
[20,23]×Ø Ø Ø ×Ø×
[21,27]× × Ø Ø Ø Ø ×
[25]×Ø Ø Ø × × ×
[26]× × Ø Ø Ø × ×
[29,33]Ø Ø Ø Ø ×Ø×
[30]× × Ø Ø ×Ø×
[31]×Ø Ø Ø Ø Ø ×
[32]Ø× × Ø×Ø×
[34]Ø Ø ×Ø Ø Ø ×
[35]Ø Ø ×Ø×Ø×
[36,37]Ø Ø ×Ø Ø × ×
Our work Ø Ø ×Ø Ø Ø Ø
2.2. Mathematical Model
The mathematical formulation of the system is presented in this subsection.
Equations (1)-(4) describe the average model of converters which is obtained by
applying the Kirchhoff rules on circuit shown in Figure 2 [45].
disc/dt =1/L1∗(Vsc −U12 Vbus −R1isc)(1)
di0
Bat/dt =1/L2∗(U43 VBat −Vbus −R2i0
Bat)(2)
di0
PV /dt =1/L3∗(U5VPV −Vbus −R3i0
PV )(3)
di0
grid/dt =1/L4∗(U6Vgrid −Vbus −R4i0
grid)(4)
8
where, Rand Lare resistance and inductance of converters, respectively. Uis the
duty cycle, U12 and U34 are defined as the common control input of bidirectional
converters and calculated using Eq.(5) and (6), respectively.
U12 =m(1−U1) + (1−m)U2(5)
U43 =m(1−U4) + (1−m)U3(6)
where, mis binary variable defined as following:
m=
1if Boostmode
0if Buckmode
(7)
The voltage at DC bus is equal to the sum of injected currents to the DC bus and
is given by Eq.(8.
Vbus(t) = 1/Cbus Zitot dt +Vbus(t0)(8)
where, Cbus is the capacity of capacitor connected to the bus. Equation (8) can be
rewritten as Eq. (9).
Vbus(t) − Vbus (t0) = 1/Cbus Zitotaldt (9)
The total injected current itotal to the DC bus can be calculated by Eq. (10.
itot =Xi=i0
sc
dis −i0
sc
ch +i0
Bat
dis −i0
Bat
ch +i0
PV +i0
grid (10)
where, Piis the sum of injected current to the DC bus by individual converters
and it can be calculated by Eq. (11.
itot =X
i0
scdis =η1Pdis
sc /Vbus =η1Vscidis
sc /Vbus
i0
scch =Pch
sc /η1Vbus =Vscich
sc /η1Vbus
i0
Batdis =η2Pdis
Bat/Vbus =η2VBat idis
Bat/Vbus
i0
Batch =Pch
Bat/η2Vbus =VBat ich
Bat/η2Vbus
i0
PV =η3Ppv/Vbus =η3VP V iPV /Vbus
i0
grid =η4Pgrid/Vbus =η4Vgrid igrid/Vbus
(11)
Consequently, the total injected current in Eq. (10) can be expressed as Eq. (12).
9
Figure 1: Simplified Model of a DC green building (microgrid)
Figure 2: Realisation mode of Microgrid
10
itot(t) = 1/Vbus η1Vscidis
sc (t) − Vscich
sc (t)/η1+η2VBatidis
Bat(t)
−VBatich
Bat(t)/η2+η3VPV iP V (t) + η4Vgridigrid(t)(12)
By substituting Eq. (12) in Eq. (9), it can be rewritten.
V−bus(t) − Vbus(to) = 1/Cbus Vbus Zη1Vscidis
sc (t) − Vscich
sc (t)/η1
+η2VBatidis
Bat(t) − VBat ich
Bat(t)/η2
+η3VPV iPV (t) + η4Vgrid igrid(t)dt (13)
Equation (13) shows voltage oscillation on the DC side, thus, a proper control
strategy is needed to avoid voltage fluctuations at the DC bus. The corresponding
control strategy is applied to the system by transmitting set points through the indi-
vidual agents. In order to calculate capacity of capacitance, in DC bus, the following
voltage constraint can be assumed:
Vnom −∆V 6Vbus 6Vnom +∆V (14)
where Vnom is the nominal voltage and ∆V signifies the maximum oscillation volt-
age at the DC bus that can be controlled by the DC bus capacitor. Equation (15)
shows the relationship between the current and voltage of capacitor.
dV(t) = 1/Cbus ∗itot(t)dt (15)
Equation (15) can further be modified to find the optimal value of capacitance
for a certain value of ∆, as
Cbus =Itot∆t/∆V (16)
The ∆t in Eq. (16) represents the total time delay in the system and calculated by
Eq. (17).
∆t =td+tp+tm+ts+tc(17)
where tdis the time delay for the power electronic devices such as the delay in
switching of converters, tpis related to the delay involved in the processing of
individual agent, tmindicates the delay in sending and receiving messages on a
11
communication port, tsrepresents the measurement delay caused by sensors and
tcis time delay associated with the controllers. For instance, in the system with
0.2 ms as a total time delay and 20 A as total current; if 2% voltage oscillation is
desired, a capacitor with 2mF should be used.
2.3. Control structure
Figure 3 shows the control structure of converters which are used to control the
voltage and current [46]. Each element in this topology is controlled by an agent.
In this control structure, agent based on measured value that could be output volt-
age of PV panel or SOC of SC or battery, and also communication signal that comes
from other agents, generates suitable set points. These set points include mode
selector, voltage and current reference. Mode selector is used to determine that
a converter is in voltage mode or in current mode. Voltage mode is considered to
maintain output voltage of converter in constant value; consequently, voltage of DC
bus will be preserved in desired value. Current mode is expected to supply shared
current by other agents. It is clear that one agent can be used to regulate the volt-
age and others just participate in the current control. In order to enable a smooth
switching, to maximize the optimum use of resources and to help protecting storage
devices, agents share current in the system. In this control structure, error signal
that is difference between Voltage/current reference, Vref (or Iref), and measured
voltage/current, Vm(or Im), applied to the P.I control. The out put of the P.I control
is applied to the pulse width modulation (PWM) in order to generate proper signal
for switching Masfet in the converter.
3. The proposed Energy Management System
3.1. Decision algorithm
As it was mentioned that there are two decision-making structures in order to
control the agents in multi-agent systems; the CEMS and the DEMS. In this study, the
DEMS is used to manage and control the energy among various available sources,
where each energy source is controlled through an agent. The first step for the
12
Figure 3: The control structure for a dc-dc converter
agents is to check the available power from the generator. Subsequently, based on
the hierarchal algorithm given in Figure 4, agents will choose the control mode that
is either voltage mode or current mode. If the voltage control mode is chosen, agent
will control DC bus voltage and can share current based on its current sharing layer
(Section 3.4). In this case, shared current is supplied by another one (as will be
discussed in Section 3.2).
3.2. Priority mechanism
In order to manage and control MASs, it is essential to utilize a robust intelligent
algorithm to allocate tasks between agents. Subsumption architecture is proposed
as a reactive structure to organize and define principle among agents. In this archi-
tecture, any agent has priority in the system. Each agent takes place in one layer
and the lowest layer has the highest priority. Brooks developed this algorithm in
1986 to control robot’s behaviours, wherein any behaviour is organized in a layered
architecture based on a priority as shown in Figure 5 [44]. For example, layer 1 has
priority over layer 2 and 3 but if layer 0 is activated, it can be disregarded. With
this approach, the main challenge of energy management in a system is divided into
a set of simple challenges which can be implemented by some simple if and then
rules.
13
Figure 4: The flowchart of agent’s decision
14
Figure 5: Brooks’s Subsumption architecture for our case study
Subsumption architecture designed by Brooks includes a hierarchy of compe-
tence for layers. Accordingly, a layer without any communication can override the
remaining of its higher layers at any time and control the system as long as it is
necessary [47]. It means that there is a rigid separation among the layers and it
is impossible for an agent to disregard this hierarchy any time even if a failure ex-
ists in the decision part. Thus, when a fault occurs in a decision layer, the entire
system collapses. In order to have a robust control on the system, in this study, a
mechanism similar to the classic subsumption will be implemented but there is a
communication between agents that enables it to change priorities. This architec-
ture is developed to control the voltage of DC bus.
As the main objective of energy management is to reduce the use of conventional
energy sources, thus, the external grid is located in the highest layer. Despite an
active load in the system, SC is considered in the lowest layer in order to store energy
when the load is in generation mode and supply power as soon as the load turns to
demand mode. Hence, SC has given the highest priority in suggested architecture
by allocating it the lowest layer. PV is next agent that has high priority to make
room for energy saving and maximum exploitation of its energy. Finally, battery is
located in layer 2 and when SC and PV are unable to control the voltage, it could be
controlled by the battery. As it is obvious, grid has the lowest priority and when SC,
15
PV, and battery could not control the voltage of DC bus, the grid will be switched in
for maintaining the system operation.
In the proposed approach, there is a distributed decision unit where agents make
their own decisions and carry out tasks independently, and in comparison with ref.
[42], communication between agents makes system robust. In order to better un-
derstand, assume in [42], agent A is in voltage mode, if a fault occurs in the central
unit, the entire system will collapse, because all the set points are appointed by
the central unit. However, in the proposed mechanism, when an agent collapses,
system is reconfigured and rest of system controls and manages energy.
3.3. Communication mechanism
To have proper reaction in the system, status of agent is transferred by a com-
munication link. Consequently, the communication between agents is explained.
3.3.1. Communication protocol
In order to use multi-agent systems, especially in DEMSs as depicted in Figure 6,
it is essential to have frequent communication among the agents. For this purpose,
Controller Area Network-open (CANopen) can be used as a communication proto-
col to transmit instructions between the agents [48]. The CAN bus is a fast field
bus control system having the possibility to transmit instructions in 0.2ms between
various CAN stations and is used for decentralization, intelligence and network con-
trol [49, 50]. Despite of CANbus does not need to support high data traffic in this
application; it has been selected due to its wide availability and comparatively low
price.
3.3.2. Digital message
In the proposed approach, agents communicate through CAN bus and transmit
5 bits to clarify their states. By using this data, each agent can have appropriate
reaction in the system. This data as status signal is defined as follows:
Bit 20shows current control mode of agents. If it is 1, that means agent just
supplies shared current. For example, in Figure 6, agent 2 supplies requested cur-
rent by agent 1. Bit 21shows voltage control mode and when its value is 1, it
16
means agent is involved in controlling the voltage of DC bus. For instance, agent
1 in Figure 6 is in voltage control mode. Bit 22and 23show the level of current
sharing (refer to Table 2). When bit 22is 1, it means that agent shares current as
defined by level 1, and when bit 23is 1 it implies that level 2 of the current sharing
is activated. Further, when bit 22and 23is activated simultaneously, it means that
current is shared based on level 3. For example, in Figure 6, agent 1 shares layer
2 of its Subsumption architecture with others. In next subsection, Subsumption
architecture for current sharing will be explained.
Bit 24shows failure status of the agent. If it is 1, agent is in normal performance
and when it is 0, agent is under fault. For example, in Figure 6, agent 3 is collapsed
and others are operating in normal state.
3.4. Current sharing mechanism
For optimal use of RESs and to eliminate voltage ripple on the switching, Brooks’
Subsumption mechanism is implemented to share current between the available
sources. Thus, when agent wants to share current, it sends a binary code through
the CAN bus wherein agent clarifies about the shared current. Based on priority and
capability, other agents can reply to the request. In order to decrease transmitted
data in sharing, maximum three levels for each agent are considered. In this case
only the agent that is on voltage control mode can make the request. The complete
mechanism of current sharing is explained in the following sections.
3.4.1. Current sharing in SC
Assume SC is in voltage control mode, therefore, it can share current based
on its charge according to Subsumption architecture shown in Figure 7 (a). The
following set of instructions will be passed over the layers.
1. Layer 0: SC will set bit 22 if its SOC is in level 1.
2. Layer 1: When its charge reaches to level 2, it triggers the bit 23.
3. Layer 2: SC activates bit 22and 23when its SOC is in level 3.
17
Figure 6: Structure of communication between agents
18
Figure 7: The Subsumption architecture for current sharing by (a) the SC agent, (b) the PV agent and
(c) by the battery agent
3.4.2. Current sharing in PV
Similar to SC, current sharing in PV is carried out by a hierarchy approach. As it
is shown in figure 7(b), when PV agent controls voltage of DC bus; it shares current
in two states.
1. Layer 0: When terminal voltage of PV is nominal, it activates bit 22.
2. Layer 1: When terminal voltage of PV is dropped from nominal value, bit 23
is set by PV agent until it loses the voltage control.
3.4.3. Current sharing in battery
According to hierarchy algorithm, battery controls voltage when SC and PV are
out of service. In this case, if battery has enough stored charge, it will supply to
fulfil the demand and share current in three levels with grid as shown in Figure 7
(c).
1. Layer 0: When SOC of battery is level 1, battery agent triggers bit 22.
2. Layer 1: Battery agent sets bit 23when its charge is in level 2.
3. Layer 2: When its charge drops to level 3, before moving out of the voltage
control, battery agent activates bits 22and 23.
When an agent is unable to supply shared current, it informs the other by acti-
vating bits 20and 21.
19
4. Results and Discussion
The system as shown in figure 1, includes an active load and a grid connected
PV with two storage devices. The details are:
•1 kW PV generator with a voltage of 190 V
•The electrical grid with a voltage of 380 V (phase to phase)
•SC with a capacity of 10 F having a nominal voltage of 75 V
•Two batteries each having a voltage of 144 V and a capacity of 60 AH
All the elements are connected to an active load through a 100 V DC bus. The
active load can be operated as a motor or generator. A current rectifier and 230/100
V buck converter is used to connect the grid to the DC bus. The SC and batteries
are connected to the load through bidirectional converter and buck converter, re-
spectively.
At first, the proposed EMS is implemented on a simulation model with a scenario
of 30 s while all the converters act with real behaviour. The converters are modelled
to present real behaviour and not just as a voltage source converter. For this reason,
inevitable oscillation appears in profiles. In the second case, a scenario of 10 s is
set up to consider operation of EMS when a failure occurs in a decision part.
4.1. Result of Energy Management
To implement the proposed mechanism on the system, some parameters have
been defined as shown in Table 2. In order to have a suitable comparison, these
parameters are obtained from [42]. Initially, it is assumed that SC and battery is
fully charged and all the agents are in normal mode, and therefore, bit 24 for all of
them is set to 1.
Control voltage by SC: In the start, all the agents check for the available power
and based on Subsumption architecture, SC agent overrides all requests and takes
control of DC bus and supplies the energy demand. Consequently, it sets bit 21and,
sends [10110]as signal to the communication port. It means that SC agent is in
normal mode and shares current with level 1 of its current constraint as shown in
figure 8. At t=1.3s, its charge is dropped to 85% (Figure 9) and SC enters to second
20
Table 2: The level of constraints for agents and respective shared currents.
Agent Level of constraints Constraints Shared current (A)
Level 1 85% <SOCSC 0
SC Level 2 70 <SOCsc 685% 6
Level 3 55% <SOCSC 670% 12
Level 1 VPV nom >190 0
PV
Level 2 VPV nom<190 15
Level 1 13% 6SOCBat 0
Battery Level 2 10% 6SOCBat<13% 8
Level 3 7%<SOCBat<10 16
level of current constrain, so activates 23bit, [11010], to request for sharing current.
Based on section 3.4, PV has high priority to share current with SC. Subsequently,
PV agent accepts to supply shared current, by activating bit 20 ([10001]). The PV
supplies 6A of demand as shown in Figure 10. At t=3.7s, as the SOC of SC drops
below 70%. PV increase the share to 12 A. Therefore, the communication signal
by SC is changed to [11110], and PV supplies a total of 12 A. If PV cannot supply
shared current, battery and grid can do it based on their priority.
Control voltage by battery: When charge of SC drops to 55% (t=7.9 s), SC
agent based on its voltage constraint in Subsumption architecture cannot deliver to
fulfil the demand. So, the voltage control transferred completely to the agent placed
in the next layer (which was supposed to be PV). However, at t=7.9 s, terminal
voltage of PV is also dropped such that it is out of service (Figure 11). Thus, based
on figure 12, battery has enough charge to supply the requested demand, and by
sending [10110]as a data signal, it controls the DC bus. Further, based on the SOC
of battery, the battery agent shares current in three levels with the grid agent, as
shown in Figure 13 and Figure 14.
Control voltage by grid: After cooperating in three levels of current share by
battery as shown in figure 14, at t=11 s the grid takes the control of voltage at DC
21
Figure 8: The injected current by SC under different time periods of decision making
Figure 9: The state of charge of SC
Figure 10: The injected currents by PV under different time sections of the decision making
22
Figure 11: The evolution of the PV output voltage
Figure 12: The state of charge of battery
Figure 13: The injected current by battery under different time span of decision making
23
Figure 14: The injected current by grid under different time section of decision making
bus by sending [10010]over the communication port. If any agent with high prior-
ity exists with sufficient power to share, it can override the grid’s request. However,
the grid has the ability to keep it for a long time without sharing current until one
of the agents with high priority overrides its precedence.
As there is an active load in the system and thus, it can provide energy in gen-
erator mode (see Figure 15). At t=13 s, load generates power and SC stores it.
Therefore, SC takes voltage control until t=15.6s. During this time, the terminal
voltage of PV is also increased. The energy generated by PV is absorbed by the bat-
tery since it is already discharged. When the SOC of battery is reached to 100%,
it is disconnected from the PV. Once the load finished generating energy, the SC is
charged enough to keep the control of DC bus by sending request [10110]. The
other agents cannot override its competence because it is in the lowest layer with
high priority.
The SC agent during voltage control of DC bus shares current with PV in two
steps by sending signals [11010]and [11110]over the communication network.
After the SOC of SC drops to a critical percentage, the PV with high priority based
on hierarchical algorithm takes voltage control of DC bus. It is obvious that when
PV controls the voltage of DC bus, others can just participate in sharing the current.
It is worth mentioning that only the agent with high priority (here SC) can take
voltage control back. The PV agent will supply load until there is sunshine. To
24
Figure 15: The waveform for the load current variation
Figure 16: The voltage at the DC bus
25
Figure 17: The injected current of SC, PV, battery and grid under different time periods of fault tolerance
test
conclude, all the interactions among the agents are assigned by a communication
signal. Agents just by reading the 5 bits signal can sense variations in the system and
take a proper control action. Figure 16 shows how agents control DC bus without
any fluctuation in switching.
4.2. Result of Fault Tolerance
In this section, the system is simulated for 10s in order to evaluate the operation
of EMS in a fault-facing at the decision part. As mentioned in Section 4.1, agents
transfer 5 bits to the communication port wherein bit 24describes about the failure
status of an agent. Assuming at the beginning that the battery agent (decision part
of battery) suffered from a breakdown. Subsequently, the signal from the battery
over the data bus will be [00000]. In this case, hierarchy in layers is reconfigured
only with sub-behaviours of SC, PV, and grid agents without considering the battery
agent. This is one of the main advantages of MASs known as fault tolerance and
high reliability. It enables a system to continue its operation while even some parts
stop working.
At first, SC agent completely controls the voltage and after sharing 6 A and 12
26
Figure 18: The voltage of the DC bus during fault tolerance test
A by sending [11010]and [11110]to the data bus, respectively, PV agent takes
the control of DC bus voltage, as shown in figure 17. The PV supplies demand un-
til t=6.3 s, this time it shares current by sending signal [11011]to the data bus.
Although battery has high the priority rather than the grid, however, the detected
signal from battery on the communication port is [00000], due to its fault. Con-
sequently, the grid agent supplies the shared current. At t=6.9 s, the grid controls
voltage of DC bus by sending [10010]signal. This situation stays until one of the
agents is able to supply the demand.
The battery failure is resolved at t=9 s, so it sets 24bit to 1. By this signal, the
configuration among the agents is changed to 4 agents and the voltage control of
DC bus is handed over to the battery agent. Figure 18 shows the changes in the
DC bus voltage. It is obvious that despite of failure in battery agent, there is no
oscillation in the voltage profile and energy management is carried out smoothly
among the generators.
5. Conclusions
In this paper, an online multi-agent based DEMS has been presented to control
the voltage of DC bus and supply-demand balance in a micro-grid. In the proposed
system, agents are able to control voltage and current of the connected device by
27
using a decision structure. The modified subsumption architecture has been pre-
sented to simplify the interaction of agents connected in a framework. The proposed
framework uses layered structure for controlling MAS. Furthermore, in order to en-
hance the reliability of decision structure and as well as stability in the microgrid,
low bandwidth communication bus has been created among layers to approve self-
healing capability. The communication between agents improves the reliability of
system, especially in mitigating the failure state by reconfiguring the sequence of
agents. The proposed control for voltage and power were validated for both normal
operation and failure state. Overall, the proposed framework presents the following
advantages:
1. Online smart control of microgrids without any prediction and scheduling
2. Flexible and easy implementation, even for large-scale systems
3. Require low communication bandwidth and low sized transmission data (5
bits)
4. High reliability and fault tolerance
5. Easy redevelopment by reassigning the priorities
The future work includes implementing a DEMS with capability of asynchronous
computing and scalability that could be used in smart buildings and as well in large
scale networks.
6. Acknowledgments
This research was supported by the Catalan Agència de Gestiód’Ajuts Universi-
taris i deRecerca, under the AGAUR 2017 SGR-00967 Research Project and by the
British council under grant contract No: IND/CONT/GA/18-19/22.
References
[1]S. Khemakhem, M. Rekik, L. Krichen, A collaborative energy management
among plug-in electric vehicle, smart homes and neighbors’ interaction for
28
residential power load profile smoothing, Journal of Building Engineering 27
(2020) 100976.
[2]V. Aryanpur, M. S. Atabaki, M. Marzband, P. Siano, K. Ghayoumi, An overview
of energy planning in iran and transition pathways towards sustainable elec-
tricity supply sector, Renewable and Sustainable Energy Reviews 112 (2019)
58–74.
[3]S. Aznavi, P. Fajri, R. Sabzehgar, A. Asrari, Optimal management of residen-
tial energy storage systems in presence of intermittencies, Journal of Building
Engineering 29 (2020) 101149.
[4]H. R. Gholinejad, A. Loni, J. Adabi, M. Marzband, A hierarchical energy man-
agement system for multiple home energy hubs in neighborhood grids, Jour-
nal of Building Engineering 28 (2020) 101028.
[5]P. Chen, Y. Wu, L. Zou, Distributive pv trading market in china: A design of
multi-agent-based model and its forecast analysis, Energy 185 (2019) 423–
436.
[6]R. K. Chauhan, K. Chauhan, Building automation system for grid-connected
home to optimize energy consumption and electricity bill, Journal of Building
Engineering 21 (2019) 409–420.
[7]L. Ju, Q. Zhang, Z. Tan, W. Wang, H. Xin, Z. Zhang, Multi-agent-system-based
coupling control optimization model for micro-grid group intelligent schedul-
ing considering autonomy-cooperative operation strategy, Energy 157 (2018)
1035–1052.
[8]J. Ferber, G. Weiss, Multi-agent systems: an introduction to distributed artifi-
cial intelligence, Vol. 1, Addison-Wesley Reading, 1999.
[9]J. B. Almada, R. P. S. Leão, R. F. Sampaio, G. C. Barroso, A centralized and
heuristic approach for energy management of an ac microgrid, Renewable
and Sustainable Energy 60 (2016) 2884–2894.
29
[10]T. Wang, D. O’Neill, H. Kamath, Dynamic control and optimization of dis-
tributed energy resources in a microgrid, IEEE Transactions on Smart Grid,
6 (6) (2015) 2884–2894.
[11]M. S. S. Danish, T. Senjyu, A. M. Ibrahimi, M. Ahmadi, A. M. Howlader, A man-
aged framework for energy-efficient building, Journal of Building Engineering
21 (2019) 120–128.
[12]H. Zhou, T. Bhattacharya, D. Tran, Composite energy storage system involving
battery and ultracapacitor with dynamic energy management in microgrid
applications, IEEE Transactions on Power Electronics 26 (3) (2011) 923–930.
[13]M. Latif, A. Nasir, Decentralized stochastic control for building energy and
comfort management, Journal of Building Engineering 24 (2019) 100739.
[14]B. Li, R. Roche, D. Paire, A. Miraoui, A price decision approach for multi-
ple multi-energy-supply microgrids considering demand response, Energy 167
(2019) 117–135.
[15]M. Nazari-Heris, M. A. Mirzaei, B. Mohammadi-Ivatloo, M. Marzband,
S. Asadi, Economic-environmental effect of power to gas technology in cou-
pled electricity and gas systems with price-responsive shiftable loads, Journal
of Cleaner Production 244 (2020) 118769.
[16]A. Q. Santos, R. M. Monaro, D. V. Coury, M. Oleskovicz, A new real-time multi-
agent system for under frequency load shedding in a smart grid context, Elec-
tric Power Systems Research 174 (2019) 105851.
[17]D. Stimoniaris, T. Kollatou, D. Tsiamitros, M. Zehir, A. Batman, M. Bagriyanik,
A. Ozdemir, E. Dialynas, Demand-side management by integrating bus com-
munication technologies into smart grids, Electric Power Systems Research
136 (2016) 251–261.
[18]M. Marzband, E. Yousefnejad, A. Sumper, J. L. Domínguez-García, Real time
experimental implementation of optimum energy management system in
30
standalone microgrid by using multi-layer ant colony optimization, Interna-
tional Journal of Electrical Power & Energy Systems 75 (2016) 265–274.
[19]D. E. Olivares, C. A. Cañizares, M. Kazerani, A centralized energy management
system for isolated microgrids, IEEE Transactions on smart grid 5 (4) (2014)
1864–1875.
[20]M. Marzband, M. Ghadimi, A. Sumper, J. L. Domínguez-García, Experimental
validation of a real-time energy management system using multi-period grav-
itational search algorithm for microgrids in islanded mode, Applied energy
128 (2014) 164–174.
[21]M. Jadidbonab, B. Mohammadi-Ivatloo, M. Marzband, P. Siano, Short-term
self-scheduling of virtual energy hub plant within thermal energy market,
IEEE Transactions on Industrial Electronics (2020) 1–1.
[22]M. A. Mirzaei, A. Sadeghi-Yazdankhah, B. Mohammadi-Ivatloo, M. Marzband,
M. Shafie-khah, J. ao P.S. Catalão, Integration of emerging resources in igdt-
based robust scheduling of combined power and natural gas systems consid-
ering flexible ramping products, Energy 189 (2019) 116195.
[23]M. Pourakbari-Kasmaei, M. Lehtonen, M. Fotuhi-Firuzabad, M. Marzband,
J. R. S. Mantovani, Optimal power flow problem considering multiple-fuel
options and disjoint operating zones: A solver-friendly MINLP model, Inter-
national Journal of Electrical Power Energy Systems 113 (2019) 45–55.
[24]N. Gholizadeh, G. Gharehpetian, M. Abedi, H. Nafisi, M. Marzband, An inno-
vative energy management framework for cooperative operation management
of electricity and natural gas demands, Energy Conversion and Management
200 (2019) 112069.
[25]C. Wang, Y. Liu, X. Li, L. Guo, L. Qiao, H. Lu, Energy management system
for stand-alone diesel-wind-biomass microgrid with energy storage system,
Energy 97 (2016) 90–104.
31
[26]R. Das, Y. Wang, G. Putrus, R. Kotter, M. Marzband, B. Herteleer, J. Warmer-
dam, Multi-objective techno-economic-environmental optimisation of electric
vehicle for energy services, Applied Energy 257 (2020) 113965.
[27]M. Hong, X. Yu, N.-P. Yu, K. A. Loparo, An energy scheduling algorithm sup-
porting power quality management in commercial building microgrids, IEEE
Transactions on Smart Grid 7 (2) (2014) 1044–1056.
[28]A. Mondal, S. Misra, M. S. Obaidat, Distributed home energy management
system with storage in smart grid using game theory, IEEE Systems Journal
11 (3) (2015) 1857–1866.
[29]Y. Xu, W. Zhang, G. Hug, S. Kar, Z. Li, Cooperative control of distributed energy
storage systems in a microgrid, IEEE Transactions on smart grid 6 (1) (2014)
238–248.
[30]B. M. Radhakrishnan, D. Srinivasan, A multi-agent based distributed energy
management scheme for smart grid applications, Energy 103 (2016) 192–204.
[31]Q. Shafiee, J. M. Guerrero, J. C. Vasquez, Distributed secondary control for is-
landed microgrids—a novel approach, IEEE Transactions on power electronics
29 (2) (2013) 1018–1031.
[32]Y. F. Eddy, H. B. Gooi, S. X. Chen, Multi-agent system for distributed man-
agement of microgrids, IEEE Transactions on power systems 30 (1) (2014)
24–34.
[33]A. Kulmala, M. Alonso, S. Repo, H. Amaris, A. Moreno, J. Mehmedalic, Z. Al-
Jassim, Hierarchical and distributed control concept for distribution network
congestion management, IET Generation, Transmission & Distribution 11 (3)
(2017) 665–675.
[34]I.-Y. Chung, S.-J. O. Cheol-HeeYoo, Distributed intelligent microgrid control
using mul-ti-agent systems, Engineering 5 (01) (2013) 1.
32
[35]H. J. Monfared, A. Ghasemi, A. Loni, M. Marzband, A hybrid price-based de-
mand response program for the residential micro-grid, Energy 185 (2019)
274–285.
[36]M. Marzband, S. S. Ghazimirsaeid, H. Uppal, T. Fernando, A real-time evalua-
tion of energy management systems for smart hybrid home microgrids, Elec-
tric Power Systems Research 143 (2017) 624–633.
[37]T. Logenthiran, R. T. Naayagi, W. L. Woo, V.-T. Phan, K. Abidi, Intelligent con-
trol system for microgrids using multiagent system, IEEE Journal of Emerging
and Selected Topics in Power Electronics 3 (4) (2015) 1036–1045.
[38]G. Artale, A. Cataliotti, V. Cosentino, D. Di Cara, S. Guaiana, N. Panzavecchia,
G. Tinè, Real-time power flow monitoring and control system for microgrids
integration in islanded scenarios, IEEE Transactions on Industry Applications
55 (6) (2019) 7186–7197.
[39]H. Fallahzadeh-Abarghouei, S. Hasanvand, A. Nikoobakht, M. Doostizadeh,
Decentralized and hierarchical voltage management of renewable energy re-
sources in distribution smart grid, International Journal of Electrical Power &
Energy Systems 100 (2018) 117–128.
[40]M. S. Golsorkhi, D. J. Hill, H. R. Karshenas, Distributed voltage control and
power management of networked microgrids, IEEE Journal of Emerging and
Selected Topics in Power Electronics 6 (4) (2017) 1892–1902.
[41]M. R. B. Khan, R. Jidin, J. Pasupuleti, Multi-agent based distributed control
architecture for microgrid energy management and optimization, Energy Con-
version and Management 112 (2016) 288–307.
[42]M. Selseleh Jonban, A. Akbarimajd, J. Javidan, Intelligent fault tolerant en-
ergy management system with layered architecture for a photovoltaic power
plant, Journal of Solar Energy Engineering 137 (1) (2015).
[43]J. Lagorse, D. Paire, A. Miraoui, A multi-agent system for energy management
of distributed power sources, Renewable energy 35 (1) (2010) 174–82.
33
[44]R. Brooks, A robust layered control system for a mobile robot, IEEE journal
on robotics and automation 2 (1) (1986) 14–23.
[45]A. Tahri, H. El Fadil, F. Belhaj, K. Gaouzi, A. Rachid, F. Giri, F. Chaoui, Manage-
ment of fuel cell power and supercapacitor state-of-charge for electric vehicles,
Electric Power Systems Research 160 (2018) 89–98.
[46]R. Teodorescu, M. Liserre, P. Rodriguez, Grid converters for photovoltaic and
wind power systems, Vol. 29, John Wiley & Sons, 2011.
[47]G. Schouten, J. Steckel, A biomimetic radar system for autonomous naviga-
tion, IEEE Transactions on Robotics 35 (3) (2019) 539–548.
[48]H. Boterenbrood, Canopen high-level protocol for can-bus, Nikhef, Amster-
dam 20 (2000).
[49]M. Esro, A. A. Basari, S. Kumar, M. Sadhiqin, Z. Syariff, Controller area net-
work (can) application in security system, World Academy of Science, Engi-
neering and Technology 35 (2009).
[50]Y. Zhu, F. Zhuo, L. Xiong, Communication platform for energy management
system in a master-slave control structure microgrid, in: Proceedings of The
7th International Power Electronics and Motion Control Conference, Vol. 1,
IEEE, 2012, pp. 141–145.
34
