Electrical Power and Energy Systems 97 (2018) 106–119

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Multi-agent based coordinated protection systems for distribution feeder T

fault diagnosis and reconfiguration
⁎
M.S. Rahman , N. Isherwood, A.M.T. Oo
School of Engineering, Deakin University, Geelong, VIC 3216, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: This paper presents an innovative peer-to-peer agent-based protection system for fault diagnosis and automatic
Distribution feeder power restoration in power distribution networks. In practice, modern digital relays play a vital role for handling
Multi-agent system faults by allowing different switching configurations of opening and closing the circuit breakers in power net-
Protection system works. An effective fault detection and diagnosis along with backup protection is essential in order to save the
Restoration
systems from collapse and reconfigure them properly after an outage. In the proposed agent-based protection
Communication
scheme, a hierarchy of two categories of agents- (i) Relay agent and (ii) Configuration agent is designed where
the agent communication plays an important role to effectively diagnose the fault conditions and thereby, re-
store the system through proper relay coordination. An open ring feeder distribution system is considered to
evaluate the performance of the proposed protection scheme. Furthermore, the impacts of both communication
failure and physical protection misoperation are also addressed in this paper through effective backup protection
provided by the coordination of agents.

1. Introduction impedance (PAPSII)-based wide-area backup protection scheme is

Power distribution systems consist of numerous protection relays at
different locations which are pre-programmed to protect the system
from various faults and contingencies. The coordination of protection
relays during faults is a tedious and time-consuming task as it is always
desirable to minimize the number of switching operations to restore the
power to unserved loads. Moreover, the traditional relays are not suf-
ficiently intelligent to handle the rapidly changing conditions since the
distribution systems exhibit random usage patterns due to the consumer
demands. As a result, sometimes traditional relays fail to handle the
faults quickly or falsely trip a healthy line from the system during a
restoration process which may lead to a complete or partial blackout in
power systems. Therefore, a better coordination among the protection
devices is very important to detect faults and take corrective actions to
preserve the system reliability and provide acceptable performance by
maintaining a proper reconfiguration after service restoration.
A lot of effort has been made to design the protection automation
infrastructure and computers are used at different levels of the power
networks to provide assistance to that protection automation tools
[1,2]. Recently, a number of knowledge-based intelligent fault diag-
nosis and reconfiguration techniques have been proposed for improving
the automation of protection tools for restoration solution [3–8]. In [9],
a new intelligent phase angle of positive-sequence integrated
proposed to provide effective fault diagnosis in distribution systems. In
[10], a frequency selectivity-based remedial action scheme (RAS) along
with backup protection is proposed for overcurrent protection schemes
in distribution networks. In [5,11], an optimized estimation-algorithm
is proposed for adaptive backup protection scheme to diagnose faults in
distribution feeders. However, most of the approaches discussed above
are based on limited communication facilities which are susceptible to
single point failures and obviously require high communication cap-
ability [12].
Currently there is a lack of application of knowledge-based artificial
intelligence technique applications in power system protection design
[13]. Recent research indicate that few agent-based technologies have
been proposed for designing protection automation tools in power
distribution networks [14]. In [15], a fuzzy clustering algorithm within
a multi-agent framework is proposed where the information of the
space relative distance between the on-line sample data and the cluster
center are utilized to coordinate the relay operation during faults. In
[16,17], an autonomous agent-based switching control algorithm is
proposed for fault diagnosis and power restoration solution in dis-
tribution networks. In [18], a multi-agent based self-healing algorithm
is proposed for distribution network automation to solve the service
restoration problems. In [2], a group of decision support agents is
proposed for distribution network fault diagnosis and restoration

⁎
Corresponding author.
E-mail address: md.s.rahman@deakin.edu.au (M.S. Rahman).

http://dx.doi.org/10.1016/j.ijepes.2017.10.031
Received 18 March 2017; Received in revised form 9 September 2017; Accepted 22 October 2017
Available online 04 January 2018
0142-0615/ © 2017 Elsevier Ltd. All rights reserved.
M.S. Rahman et al.
strategy where the unified data platform is built to manage the fault
data, information and system models. In [19], an agent-based co-si-
mulation platform is developed for fault diagnosis and reconfiguration
of distribution networks where a high-level architecture (HLA) is used
for managing the information exchange among agents. Although the
aforementioned approaches are surely functional to solve relevant
problems, still better coordination among the protection relays is ne-
cessary to handle the situation when there is a physical failure.
In [20], it is reported that an agent-based protection system can reduce
the operating time of the backup protection systems to react when the
primary relay fails as compared to the traditional protection systems. In
[21,22], an adaptive multi-agent based protection system is proposed for
distribution networks to provide a pair-to-pair relay agent coordination for
backup protection. In [23], a multi-agent strategy along with program-
mable logic controller (PLC) and field bus technology is proposed to adopt
the information interaction among multiple agents to improve the re-
sponse time of the backup protection schemes. In [24], a communication
assisted multi-agent based adaptive over-current protection scheme is
proposed in which the reliable communication among agents increases the
potential of the backup protection systems. Although the agent-based
backup protection systems work well if any breaker fails to trip during
faults, these methods are unable to handle the situations if there is a
communication failure in the systems.
In order to overcome the aforementioned limitations and technical
challenges, a new agent-based protection coordination model is proposed
without changing the basic decision making logic of the actual digital relays.
In this paper, two types of agents are designed such as- (i) relay agent (RA)
and (ii) configuration agent (CA) to improve the performance of the digital
relay operation and their interactions with power systems via multi-agent
communication. The communication of pair-to-pair agent facilitates the
information exchange between the RAs and the CA which enables the relays
to respond quickly to the changing system conditions by adopting new
prevailing conditions. In the designed scheme, the agents interact with each
other through local as well as neighboring information exchange. Moreover,
the information sharing can also reduce the frequent false tripping and
prevent cascading failures. The proposed scheme also provides an effective
platform for backup protection to reduce the fault clearance time in case of
any failure in physical breakers or communication links.
Multi-agent communication layer
Configuration
Agent
RA-1 RA-2 RA-3
Data acquisition layer (PMU)
CB CB CB
Node 1 Node 2
TR-1
Load-1
DG-1
Physical power grid model
107
Electrical Power and Energy Systems 97 (2018) 106–119
In this paper, an open ring distribution feeder is considered for simu-
lation purposes and several case studies are demonstrated to evaluate the
effectiveness of the proposed method. The simulation results show that the
agents can determine their activity to flexibly manage the network re-
configuration process through proper fault diagnosis. Also, a comparative
study with an existing method is also presented to justify the effectiveness
of the proposed approach. However, the major contributions of this paper
are outlined below which indicate the novelty of the proposed method:
(1) The multi-agent interaction allows to determine the relay control
operation to precisely identify the faults as well as to take proper
actions to restore the power system with minimal impact to the rest
of the system.
(2) The appropriate information exchange among the agents requires
only a minimal number of switching operation which improves the
resiliency and reliability of the distribution systems.
(3) Each set of agents uses local as well as neighboring information
which in turn primarily reduces the computational burdens and also
makes the breaker operation highly efficient and accurate to diag-
nose the fault conditions.
2. Agent-based protection systems
In recent years, a number of intelligent protection analysis tools have
been developed by the research community to meet the requirements of
comprehensive and automated protection relay coordination for the utility
companies. A multi-agent system (MAS) is widely recognized as an ef-
fective real-time platform which adds an extra value to the automation of
protection systems [12]. The modular architecture of the MAS provides
agents to interact with each other through neighboring information ex-
change to achieve a global goal while satisfying the local objectives.
In the designed protection framework, the RAs are designed to co-
ordinate the relay operation and control the corresponding circuit
breakers. It is assumed that, each RA has complete access to the phasor
measurement unit (PMU) information where the synchrophasor current
information provide an accurate detection of faults in the system. The RAs
simultaneously send and accept requests and share information to the
neighboring RAs via information exchange to rapidly determine the fault
Fig. 1. Interaction between multi-agent system and power
system.
RA-n
CB
Node n
TR-n
DG-n
M.S. Rahman et al. Electrical Power and Energy Systems 97 (2018) 106–119

location and remove that fault from the system. On the other hand, the CA 3.1. Fault detection and isolation
configures the relay operation via information exchange with the RAs to
determine the new system configuration to restore the network and pro- Each intelligent RA works according to the algorithm developed for
vide backup protection. In the agent communication framework, all controlling the relays and associated CBs. At every time-step, they read
messages received by an agent are acknowledged with a reply. In the the local synchrophasor current value from the PMUs and take appro-
present implementation, it is assumed that if a message isn't acknowledged priate actions based on the set of rules depicted in Algorithm 1 to
within two simulation cycles, the receiving agent has failed to cooperate. control the corresponding CBs. In order to determine the pickup current
The agent hierarchy and information flow among the RAs and CA to in- for threshold setting (Ts) of each relay, the RA allows maximum load
teract with the physical power grid model can be seen in Fig. 1 and the current for each load by summing all the loads (from first load (FL) to
detail functionalities of each agent are discussed in the following sections. the last load (LL)) downstream to the sequence of breakers. The pickup
current (ITH) of the relay generally lies in between the maximum load
3. Role of RA currents (ILoadmax) and the minimum fault currents (Ifaultmin) of the
system [25] to ensure that the relays do not trip unnecessarily and
In this paper, each relay in the distribution network is regarded as protection system operates for low as well as high fault current values.
an individual RA which is designed to control the corresponding CBs via Therefore, the relay pickup current can be written as [26]:
breaker command signals. All RAs have their own individual IDs where
ILoadmax < ITH ⩽ I faultmin (1)
they run to detect faults, determine the breaker failures due to mis-
operation and perform backup protection through information ex- Now, if FL > LL, there are no loads in between the segments and in
change. PMUs are used to capture the continuous measurement of such a condition, the threshold (Ts) is set to default. However, if
synchrophasor currents and RAs utilize these information to determine FL < LL, the threshold (Ts) is set to the summation of the maximum
the threshold settings of each relay to control the corresponding CBs. In load current for all loads between FL and LL. Therefore, the overall
order to simplify the relay operation, the RAs are divided into two threshold setting can be obtained as:
subcategories with unique agent IDs- upstream RA (UA) and down-
stream RA (DA) as shown in Fig. 2 in which it can be seen that the UA is ⎧ default;FL > LL
closer to the power generation source whereas its neighboring DA is far ⎪ LL
Ts =
away from the same power source. The algorithm for the high level ⎨ ∑ ILoadmax;FL < LL
⎪ i = FL (2)
logic operation of RA is summarized in Algorithm 1 and the details of ⎩
the functions of RA are discussed in the following subsections. However, the decision about the relay setting with corresponding
Algorithm 1 (High level logical algorithm for RA activity). breaker control state is a complex feature which often requires to
identify the active relays. The following expression shows the re-
1. Start lationship between the fault status flag (Fs), current through the relay
2. Update settings → Threshold setting (Ts), IDs of UA and DA, (IR) and the threshold value of current (ITH) determined by the RA:
protection location
0;0 < IR ⩽ ITH
3. Update relay status → Active or Inactive Fs = ⎧
4. Identify → Active RAs ⎩1;ITH < IR < ∞
⎨ (3)
5. Observe relay current (IR) ← PMU measurements The RAs use the synchrophasor current measurement to determine
6. Loop: whether the flow of current through the relay is within the threshold
7. if ITH < IR < ∞ then ranges or not. From the above expression, it can be seen that Fs will be
8. message.equal → “Faultstatus” ‘0’ when the current through the relay flows below the threshold limit,
9. Report fault location (segment) → CA i.e., (IR ≤ ITH) whereas it will be ‘1’ when a fault occurs in the system as
10. Set Fs = 1 and |z|c = 0 the current through the relay exceeds the threshold limit, i.e.,
11. Identify → immediate UA of each DA and send fault status to (IR > ITH). However, once the fault is identified, the information about
corresponding UA the fault location is sent to the CA for system reconfiguration. It is noted
12. Clear fault and wait for fault clearing time that, steps 13–23 in Algorithm 1 regards to the backup protection
13. if Breaker failed to remove fault then capability of agents.
14. Report failed breaker ID → CA Depending upon the fault status, RAs apply a set of logics according
15. Set → UA ← UA + 1 to the Algorithm 1 to control the corresponding CBs and the breaker
16. message.equal → “BackupRequest” control flags provided from each RA can be written as:
17. Send backup request → UA (with threshold settings)
18. if Request acknowledged by UA then 1;FS = 0
zc = ⎧
19. Wait until 0 < IR ≤ ITH ⎩ 0;FS = 1
⎨ (4)
20. Report fault clear → CA
From the above expressions, it is seen that the |z|c will be ‘1’ to close
21. else
the CB if the fault status is ‘0’ whereas it will be ‘0’ to open the CB if the
22. Set → present UA to immediate next UA
fault status is ‘1’.
23. end
24. goto step 13
25. else
Power flow
26. 0 < IR ≤ ITH
27. Set → Fs = 0 and |z|c = 1 CB CB
28. Send fault status → UA
Load
29. end DG Relay
UA DA
30. goto loop agent
31. end
32. close Communication link
Fig. 2. Upstream and downstream relay agents.

108
M.S. Rahman et al. Electrical Power and Energy Systems 97 (2018) 106–119

Fig. 3. Singe line diagram of open ring distribution feeder.

L-1 L-2
B B B B B CB5
TR-1 CB1 CB2 CB3 CB4
DG-1 L-3
Load 1 Load 2
B B B CB6
L-4
Subsystem-1 CB8 CB7
Load 4 Load 3
B CB9
Open ring
point L-5
B CB10
CB14
L-6 L-7
B B B B
Subsystem-2
CB11 CB12 CB13
Load 5 Load 6
L-10 L-9 Load 7
B B B B B B
L-8
TR-2 CB20 CB19 CB18 CB17 CB16 CB15
DG-2 Load 9 Load 8

3.2. Communication and coordination among RAs
If a DA detects a fault in the system, it will send the fault status
information (“FaultStatus” message) to its immediate UA once per cycle.
While at the same time, the UA waits for the message to arrive from its
DA so it can determine whether the fault is within its protection seg-
ment. If the fault occurs, it will send an additional fault status messages
to its next neighboring agent. However, if the fault is not detected by
that agent, the RAs determine that the fault is within the segment of
these two agents. A remarkable point is that, this feature can prevent
the possibility of cascading failures and limit the chances of false trip-
ping which is a key advantage of the proposed scheme.
Since there is a continual flow of information between the UA and
DA, the RAs update their fault status and the breaker control logic ac-
cording to the synchrophasor information. The messaging structure for
RA-to-RA communication within the MAS framework is given below:
Message: “FaultOnSegment”
Contents: The segment has failed
Actions: Fs → either 0 (normal state) or 1 (fault detected)
Response: “FaultOnSegmentReceived”
3.3. Backup protection by RA
this information to the CA and communicates to its immediate neigh-
boring UA (UA + 1) by sending “BackupRequest” message to take the role
of backup protection to remove the fault from the system. However, if the
fault is cleared, the RA reports this information to the CA.
Now, if a RA doesn't receive a “BackupRequestReceived” acknowl-
edgment reply from its neighborhood UA within a time frame, it is then
recognized this condition as a communication failure or agent mal-
function. In such case, it communicates to the next UA by sending the
“BackupRequest” message which allows the system to cope with both
agent malfunctions and communication failures.
4. Role of CA
In this paper, the well-known floor function approach [27] is con-
sidered to determine the rules for reconfiguring the distribution networks
with a minimal number of UA and DA operation. Since the distribution
system is supplied from individual DG sources, let the minimum segment
of the distribution system is a and the breaker (for both UA and DA)
number is b, then the generalized formulation can be written as:
UA = DA = b ± 1
b±1
FL = floor (x ) → ⌊x ⌋;where x = 2
a±1
LL = floor (x ) → ⌊x ⌋;where x = 2

In this paper, the designed agent-based protection scheme is able to
provide backup protection in case if any RA or physical CB fails to operate.
In order to coordinate the primary-backup relay pairs, all RAs (upstream
and downstream agents) can play the dual role of both primary and
backup protection by sending “BackupRequest” message to their immediate
UA if necessary. A primary RA attempts to remove a fault by opening its
corresponding CB and waits until the fault clearance time to expire.
Moreover, this RA also checks whether the current flows through the
primary relay has returned to a level below its threshold setting or not. If
the fault is not cleared and still exists after a certain time period, the RA
assumes that the corresponding breaker is not operating properly perhaps
due to physical failure or misoperation. In such a condition, the RA sends
Using the above formulation, the CA reconfigures the distribution
network in case of faults, physical breaker failure and communication
failure. The breaker number can be more or less than the open point of the
distribution system depending upon the placement of the DG sources
whether they are connected to the start point or end point of the open ring
distribution feeders. This determines the initial system configuration
through the CA. However, the configuration of the system due to the failed
segments and failed breakers can be updated according to the condition of
the fault status flag and the increment of either maximum or minimum
source segments. The DG source configuration can also be updated ac-
cording to the position of DG sources within the open ring feeders.
However, the algorithm developed for the generalized high level logical

109
M.S. Rahman et al. Electrical Power and Energy Systems 97 (2018) 106–119

operation of CA to find a suitable restoration solution for reconfiguring the 5. Interaction between CA and RAs
distribution network is summarized in Algorithm 2.
The interaction between the RAs and the CA determines an effective
Algorithm 2 (High level logical algorithm for CA activity).
reconfiguration of the distribution feeders when fault occurs and backup
protection requires. If a RA determines that it has a fault on its protection
1. Start
segment, it will send a “FaultOnSegment” message along with its protection
2. Record open point ← no. of CB/2
segment ID to the CA which records that information in the segment list and
3. List a ← SegmentNum
acknowledge this by sending a reply message “FaultOnSegmentReceived”.
4. List b ← BreakerNum (corresponding breakers)
Moreover, if any CB is not working properly due to physical malfunctions,
5. List Smax ← MaxSourceSegment
the corresponding RA will send a “BreakerFailed” message to the CA which
6. List Smin ← MinSourceSegment
records the individual breaker ID or status as non-operational to the breaker
Determine initial system configuration
list and sends a reply message “BreakerFailedReceived”. However, the CA
7. Require
uses all the information when it reconfigures the system and sends a
(i) For all b < open point (consider DG source at the start point of
“BreakerConfiguration” message to both UA and DA of the faulted segment to
open ring feeder)
restore power to unserved loads to ensure the energy supply to as many
(ii) For all b > open point (consider DG source at the end point of
loads as possible. The RA will also respond to the message being sent from
open ring feeder)
the CA with a “BreakerConfigurationReceived” message to confirm the receipt
Update configuration due to failed segments
of the message.
8. Set a → 1
Now, if a RA determines that the fault is cleared by either itself or its
9. while a < NumberofSegments do
immediate upstream agent after it requested for backup protection, the RA
10. if Fs = 1 and a < Smax then
will inform this condition to the CA. In such a condition, the CA receives a
11. Smax = a
“FaultCleared” message from the corresponding RAs and consequently,
12. else
sends a “FaultClearedReceived” acknowledgement message to the corre-
13. Fs = 1 and a ≥ Smin
sponding RAs. At the same time, the CA determines the open point of the
14. Smin = a + 1
distribution feeders and reconfigures the distribution network topology by
15. end
sending messages to the corresponding RAs. During the entire process, the
16. end while
agents perform the following messaging structure for reconfiguration se-
Update configuration due to failed breakers
quence.
17. Set b → 1
18. while b < NumberofBreakers do
Message: “FaultOnSegment”
19. if Fs = 1 and b − 1 < Smax then
Contents: Segment ID
20. Smax = b − 1
Response: “FaultOnSegmentReceived”
21. else
Message: “FaultCleared”
22. Fs = 1 and b ≥ Smin
Contents: The senders agent number
23. Smin = b + 1
Actions: Fs → either 0 (normal state) or 1 (fault detected)
24. end
Response: “FaultClearedReceived”
25. end while
Message: “BreakerConfiguration”
Update DG source configuration
Contents: Breaker status
26. Require
Response: “BreakerConfigurationReceived”
(iii) For all b < Smax (consider DG source at the start point of
Message: “BreakerFailed”
open ring feeder)
Contents: Sender ID
(iv) For all b > Smin (consider DG source at the end point of open
Response: “BreakerFailedReceived”
ring feeder)
High level logic for system reconfiguration
b±1
27. FL = floor(x) → FL = ⌊x ⌋ where x = 2
a±1 Table 1
28. LL = floor(x) → LL = ⌊x ⌋ where x = 2 Breaker current measurement with margin of relay settings.
29. if FL > LL then (Consider DG source at the start point of open ring
feeder) CB Loads Peak current Threshold current Margin (A) Margin (%)
(A) (A)
30. Set Ts → default (There are no loads in faulted segment)
31. Set corresponding b → Inactive 1 1,2,3,4 194.2 300 105.8 35.3
32. else 2 1,2,3,4 194.2 300 105.8 35.3
L
33. Set Ts → Ts = ∑i =LFL ILoad (There are loads in faulted segment) 3 2,3,4 145.6 225 79.5 35.3
4 2,3,4 145.6 225 79.4 35.3
34. Set corresponding b → Active
5 3,4 97.1 150 52.9 35.3
35. if b + 1 > NumOfBreakers then 6 3,4 96.9 150 53.1 35.4
36. Set UA → 0 7 4 48.6 75 26.4 35.2
37. else if b ± 1 = = open breakers then 8 4 48.4 75 26.6 35.5
38. Set DA → 0 9 – 0.0 1.0 1.0 100
10 – 0.0 1.0 1.0 100
39. else
11 5 44.0 65 21.0 32.3
40. Set UA = DA → b ± 1 12 5 44.2 65 20.8 32.0
41. end 13 6,5 88.2 130 41.8 32.2
42. Set a = b − 1 14 6,5 88.4 130 41.6 32.0
15 7,6,5 132.4 195 62.6 32.1
43. end
16 7,6,5 132.4 195 62.6 32.1
44. Update Inactive agents 17 8,7,6,5 176.6 260 83.4 32.1
45. Close 18 8,7,6,5 176.6 260 83.4 32.1
46. Send inactive configuration → corresponding RAs 19 9,8,7,6,5 220.9 325 104.1 32.0
20 9,8,7,6,5 220.9 325 104.1 32.0

110
M.S. Rahman et al.
Process Layer
Interfacing Layer
MACSimJX
Measured signals
MACSim Client
Physical Open Ring
Real-time control
Distribution Feeder
commands
MACSim Server
Fig. 4. Interaction of physical power system with multi-agent system.
Fig. 5. (a) Current through CB7 and (b) corresponding CB status during fault in subsystem-1 without time delay.
Fig. 6. (a) Current through CB8 and (b) corresponding CB status during fault in subsystem-1 without time delay.
6. Open ring distribution feeder
In this paper, a 60 Hz medium voltage (MV) open ring distribution
feeder as shown in Fig. 3 is considered. The system is divided into two
subsystems with a total load demand of 3.01 MW and 1.232 MVAr. The
DG units in both subsystems operate at 33 kV and each subsystem is
supplied from a separate 4.1 MVA 33/11 kV transformer to reduce 33
kV voltage level to a nominal voltage of 11 kV. At normal operating
condition, subsystem-1 is supplied from DG-1 and subsystem-2 is sup-
plied from DG-2. The distribution feeder also consists of twenty CBs,
nine residential loads, and ten power distribution lines where the
overhead power lines are modeled as a single Π section with a length of
0.5 km each. The resistance, inductance, and capacitance of each dis-
tribution line is considered as 0.2568 Ω⧹km, 2.0 mH⧹km, and 8.6 nF/
km, respectively. It is assumed that, each of the subsystems has PMUs at
each bus which provide the continuous measurement of the synchro-
phasor voltage and injected current.
However, the subsystem-1 consists of four loads where the real
power demand of each load is 362.5 kW and reactive power demand is
119 kVAr. The nominal RMS current (measured from the PMUs) for
each load in subsystem-1 is 34.7 A whereas the nominal peak current is
49.1 A. On the other hand, the subsystem-2 consists of five loads where
111
Electrical Power and Energy Systems 97 (2018) 106–119
Multi-agent Layer
Multi-agent system
CA
Real-time sampling
signals
RA RA
Real-time decision
signals
the real power demand of each load is 312 kW and reactive power
demand for the same is 151.2 kVAr. The nominal RMS current mea-
sured from the PMUs for each load in subsystem-2 is 31.5 A whereas the
nominal peak current is 44.5 A. the power factor for all loads is con-
sidered as 0.95. In this paper, the margin of the relay settings along
with percentage margin for each relay associated with the corre-
sponding loads can be determined as:
M argin (A) = ITH −Iabs
M argin (A)
M argin (%) = ITH
× 100 (5)
where Iabs is the absolute value of the peak current measurement. At
standard configuration of the distribution network, the threshold set-
tings for each of the relays can be seen in Table 1. From Table 1, it can
be seen that the settings of the threshold values are at least 35% and
32% greater than the peak currents in subsystem-1 and subsystem-2,
respectively. As a result, the system will not trip unnecessarily since
there is enough margin considered within the relay settings. From
Table 1, it can be also seen that there is a negligible current flow
through CBs 9 and 10 which confirms that the open point of the system
is operating at the default configuration and the DG sources are isolated
from each other.
M.S. Rahman et al. Electrical Power and Energy Systems 97 (2018) 106–119

Table 2 middleware as client whereas the real-time output control signals from
Breaker current measurement with threshold for fault condition. the JADE layer are sent back to the physical system through MACSim
middleware as server. The agents within the JADE environment are
CB Absolute peak current (A)
capable of communicating with each other via real-time information
Pre-fault During fault Post-fault exchange. All the simulations are performed on a 100 µs resolution
which translates to a 10 kHz sampling time.
1 194 9235 160
2 194 9235 157
3 145 9223 109 7.2. Simulation results
4 145 9223 102
5 97 9216 55
6 97 9216 49
The following cases are considered to evaluate the potential of the
7 48 9216 0 proposed approach:
8 48 28 0
9 0.0 0 48
• Performance evaluation for fault in subsystem-1 without time delay
10 0.0 0 51
• Performance evaluation for fault in subsystem-2 with time delay
• Performance evaluation when a physical CB fails to operate
11 44 44 93
12 44 44 95
13
14
87
87
87
87
138
140
• Performance evaluation during communication failure
15 131 131 183
16 131 131 184 7.2.1. Performance evaluation for fault in subsystem-1 without time delay
17 175 175 227 In this case study, a short-circuit fault is applied at line 4 in sub-
18 175 175 230 system-1 at 1.0 s. Since the power is flowing from DG-1 to supply the
19 219 219 273
loads in subsystem-1, the fault current flows through all RAs 1–7 except
20 219 219 275
RA8. However, RAs 1–6 will not take any action as they all have cor-
responding DA which detects the fault accordingly. Since RA8 is a
7. Performance evaluation and results downstream agent of RA7 and it does not sense the fault, RA7 identifies
the fault at line 4. As the current through CB7 exceeds the threshold
The performance of the proposed agent-based protection scheme is value as seen in Fig. 5(a), it initiates a control signal |z|c = 0 to open
evaluated on the open ring distribution feeder as seen in Fig. 3 under the corresponding CB7 at 1.001 s as seen in Fig. 5(b).
various case studies. The RAs are capable of detecting and removing Now, once the fault has been cleared by the RA7, the CB8 is tripped
faults from the system, handling various failure modes and auto- by RA8 at 1.018 s as shown in Fig. 6(a) to isolate line 4 as well as load 4
matically restoring power to the un-faulted sections in between 18 ms from the system. The current through the corresponding CB8 is also
and 22 ms by coordinating with the CA. The proposed framework is also shown in Fig. 6(b). The fault clearance time is dependent on the time
compared with an existing Alternative Transients Program (ATP) si- the fault occurs as the breaker will only open as the current passes
mulation method presented in [28]. The detail description of the si- through zero which may cause a half mains cycle variance. The absolute
mulation middleware platform is discussed in the following section. peak current measurements for each CBs at pre-fault, during fault and
post-fault periods are tabulated in Table 2. From Table 2, it can be seen
that during fault the currents through CBs 1–7 are increased by ap-
7.1. Simulation middleware proximately 47 times the pre-fault values while CB8 has a decrease in
current magnitude as 28 A before a high frequency spike appears at
In this paper, the required multi-agent system is developed by using 1.018 s as seen in Fig. 6(b). Since agents restore power to load 4 from
the Java Agent Development Framework (JADE) platform and the test DG-2, an extra current flows into the load 4 from DG-2 can be seen in
distribution feeder is modeled and simulated by using MATLAB/ Table 2. As a result, currents through CBs 9–20 increase in subsystem-2
Simulink SimPowerSystems Toolbox. Both the JADE and MATLAB/ at post-fault period whereas there are decrease in currents through CBs
Simulink run on a same computer to allow the agents in the MAS layer 1–8 in subsystem-1.
to gather readings (PMU measurements) and control (open/close Now, while coordinating the CB operation, the corresponding RA7
command) the breakers in the Simulink process layer. These two soft- and RA8 send a “FaultOnSegment” message to the CA to inform about
ware platforms are linked together through the MACSimJX middleware the fault event. After receiving information from RAs, the CA re-
to enable an interaction of the physical power system and its integrated configures the distribution system by exchanging information with both
multi-agent communication layer as shown in Fig. 4 to form an RA9 and RA10 which initiate their control signal |z|c = 1 to close their
equivalent industry standard smart grid communication protocol. In corresponding CB9 and CB10 as seen in Fig. 7(a) and (b), respectively.
Fig. 4, it can be seen that the measurement data from the physical The currents through the corresponding CB9 and CB10 are also shown
power system are transferred to the JADE layer through MACSim in Fig. 8(a) and (b), respectively. In such a condition, the open point in

Fig. 7. Breaker status for (a) CB9 and (b) CB10 during fault in subsystem-1 without time delay.

112
M.S. Rahman et al.
Fig. 8. Current through (a) CB9 and (b) CB10 during fault in subsystem-1 without time delay.
the distribution feeder is line 4 while load 4 is being unserved from 1.0
s to 1.018 s before the restoration process. The corresponding voltage
and current for load 4 are shown in Fig. 9(a) and (b), respectively, in
which it is seen that there is a significant amount of high frequency
noise present in the voltage measurement due to the current harmonics
interacting with the impedance present in the power sources. However,
the 18 ms power interruption is slightly more than one mains cycle at
60 Hz. From Fig. 9(b) it can be seen that at pre-fault condition, the peak
current of load 4 is 48.5 A whereas at the post-fault period, it has
dropped slightly to 47.8 A due to an extra loading on DG-2 as the
distribution system is reconfigured by the agents.
During the system reconfiguration, the power demand at load 4 is
362.5 kW. The output power from both DG units are shown in
Fig. 10(a) and (b) in which it can be seen that DG-1 has lost 348 kW real
power and 121 kVAr reactive power whereas DG-2 has gained 335 kW
real power and 120 kVAr reactive power. The load on DG-2 at nominal
rating is expected to be slightly less than that of DG-1 since a number of
distribution lines are present between load 4 and DG-2 which cause an
increase in voltage drop within the system.
The communication flow diagram of all agents for this case study is
shown in Fig. 11 in which it can be seen that the CA communicates with
all corresponding RA7, RA8, RA9 and RA10 via different message
schemes. All the agents are performed their tasks properly in terms of
fault detection, breaker control and system reconfiguration.
113
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 9. (a) Voltage and (b) current at load
point 4 during fault in subsystem-1 without
time delay.
Fig. 10. Active and reactive power output from
(a) DG-1 and (b) DG-2 during fault in subsystem-1
without time delay.
7.2.2. Performance evaluation for fault in subsystem-2 with time delay
In this scenario, another fault is applied in subsystem-2 at 6.0 s in
line 7. In order to demonstrate a real world scenario, an intentional
time delay has been introduced in the agent communication paths.
Furthermore, the results are also compared with the method presented
in [28]. A 5 ms time delay is also used for RAs to extract data from the
PMU data acquisition layer. The RAs 14–20 detect the fault since the
power is fed from DG-2. In such a condition, RAs 15–20 do not act as all
have respective DA which also detects that fault. Since the fault exists in
line 7 between RA 13 and RA 14, both RAs send a control signal |z|c = 0
to their corresponding CBs to remove the fault. CBs 9–13 are opened
with control signal |z|c = 0 at 6.02 s once they received commands
from the corresponding RAs to remove all loads that have no source of
power. In this scenario, an additional time delay of 10 ms is added to
the RA-to-RA communication channel by allowing the RAs for a
“BreakerMessageDelayTime” message to respond. Fig. 12(a) and (b)
shows the fault current through the RAs 13–14, respectively, in which it
can be seen that even if with the time delay, the proposed system
performs much better with less oscillations. Moreover, the agents can
quickly remove the fault from the system in 25 ms whereas the pro-
tection systems with the method reported in [28] requires 0.25 s to
remove the fault from the system.
However at the onset of the fault at line 7, the RAs 13–14 inform the
event to the CA and a separate “ConfigMessageDelay” message with time
M.S. Rahman et al.
Fig. 11. Multi-agent communication flow diagram.
of 2 full mains cycles is added to the communication path between
them. The CA has been modified with delay messages
“MaxBreakerReactionTime” when reconfiguring the system by allowing
a little extra time for the inactive RAs to open the corresponding CBs
before the active RAs close their CBs. In such a condition, RAs 8–15 are
reconfigured as inactive agents by opening their CBs to remove all lines
as well as loads which cannot be supplied due to the upstream faults
and the open point becomes from line 4 to load 7. The voltage profile of
load 5 and load 6 are also affected due to the faults which are shown in
Fig. 13(a) and (b), respectively. It can be observed that, the load
114
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 12. Current through (a) CB16 and (b) CB17
during fault in subsystem-2 with time delay.
voltages tend towards zero (as the fault occurs between load 6 and DG-
2) and a voltage rise appears in load 5 and load 6 which is significant
when the existing method [28] is used. This voltage rise occurs due to
the removal of the fault and loads from the DG sources. Now, when the
fault is removed by RAs 13–14, the power delivery from DG-2 has
dropped by 925 kW and 417 kVAr which are shown in Fig. 14. From the
simulation results obtained in this section, it can be seen that the pro-
posed scheme performs better with 10 times faster than the exiting
method presented in [28].
M.S. Rahman et al.
Fig. 14. Active and reactive power delivery from DG-2 during fault in subsystem-2 with
time delay.
7.2.3. Performance evaluation when a physical CB fails to operate
In this scenario, a case of physical breaker failure is considered to
see the effectiveness of the backup protection provided by the proposed
agent-based scheme. In this case study, CB16 is intentionally disabled
and a short-circuit fault is applied at line 8 in subsystem-2 at 1.0 s. It is
assumed that CB16 is unable to operate physically perhaps due to
protection malfunction.
Now when the fault occurs, RAs 16–20 detect that fault and since
RA15 does not seen this fault, RA16 recognize this fault on the line
segment 8 in subsystem-2. Although RA16 initiates a control signal |z|c
= 0 as shown in Fig. 15(a) to control the CB16, it fails to open and
clears the fault due to the physical malfunction. After the fault clearing
time has expired, RA16 checks whether the current has fallen below the
threshold value or not. Since CB16 is unable to operate and the fault
current still persists, RA16 sends a “BackupRequest” message to its
neighboring upstream RA17 which immediately changes its control
signal from |z|c = 1 to |z|c = 0 as shown in Fig. 15(b) to open its
corresponding CB17 at 1.017 s to remove the fault. At the same time,
RA16 also sends both “FaultOnSegment” and “BreakerFailed” messages to
the CA. However, once the fault is cleared, both RA16 and RA17 send
“FaultCleared” message to the CA which determines the new system
configuration taking into account all the faulted segments and RAs to
reconfigure the system. The CA sends “BreakerConfiguration” message to
115
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 13. Voltage profile for (a) Load 5 and (b)
Load 6 during fault in subsystem-2 with time
delay.
all RAs to set them as active agents to reconfigure the network and the
corresponding outputs from CB9 and CB10 are shown in
Fig. 16(a) and (b), respectively, which indicate the restoration of power
to the unserved loads. Moreover in Fig. 17(a) and (b), it can be seen that
there is no current flowing through CB9 and CB10, respectively, until
the fault is cleared and system is reconfigured at 1.02 s.
In this case study, the open point in the distribution feeder is line 8
and load 8. Although agents can provide backup protection for the
entire distribution system, power cannot be restored to supply load 8 as
seen in Fig. 18(a) and (b) since the CB16 has failed to operate physi-
cally. In such a condition, loads 5, 6, and 7 are being supplied from DG-
2 rather than DG-1 which requires overall 22 ms for successful co-
ordination to restore power to the loads. Fig. 19(a) and (b) shows that
when the fault is cleared there is no current flowing through CB15 and
CB17, respectively, from 1.02 ms since they tripped properly to remove
the fault.
The output power delivered from both DG-1 and DG-2 are shown in
Fig. 20(a) and (b), respectively, in which it can be seen that the power
demand in subsystem-2 is decreased by 1200 kW real power and 596
kVAr reactive power at post-fault period due to the reduction of loads 5,
6, 7, and 8. It can be also seen that the power demand in subsystem-1 is
increased by 854 kW real power and 451 kVAr reactive power since
loads 5, 6 and 7 are being supplied from DG-1 rather than DG-2. Due to
the successful multi-agent interaction all agents are performing their
specific tasks in a smarter way to provide the backup protection. From
this case study it is obvious that though there is a physical failure on the
primary protection system, the corresponding RA of the affected CB is
able to determine the issue and requests backup protection from the
neighboring RAs.
7.2.4. Performance evaluation during communication failure
In this case study, a failure in the communication link is considered
(perhaps due to cyber attack or security breach in the database) among
the RAs in order to observe the potential of the proposed approach. It is
noted that, agents use the information from other agents to decide
whether the communication link is functional or not. A short-circuit
fault is also applied at line 9 in subsystem-2 at 1.0 s which is supposed
to be cleared by RA17 and RA18. It is considered that, CB18 is unable to
operate physically and in such a condition, RA18 sends a
“BackupRequest” message to its upstream RA19 to take the role for
Fig. 15. Breaker status for (a) CB16 and (b)
CB17 during physical CB failure.
M.S. Rahman et al.
clearing the fault.
In order to simulate an event of communication failure, the com-
munication link between RA18 and RA19 is intentionally disabled. As a
result, RA19 will not acknowledge the message received from RA18
after a certain cycle (10 ms is considered). Since there is no information
exchange between RA18 and RA19 within a specified time frame in the
same subsystem, this event is recognized as a failure in the commu-
nication link between them. Since RA18 fails to receive a
“BackupRequestReceived” within the timeout period, it determines its
next upstream RA20 and subsequently, sends a “BackupRequest” mes-
sage to it. Due to the failures in both physical breaker and the com-
munication link between RA18 and RA19, both CB18 and CB19 do not
116
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 16. Breaker status for (a) CB9 and (b)
CB10 during physical CB failure.
Fig. 17. Current through (a) CB9 and (b) CB10
during physical CB failure.
Fig. 18. (a) Voltage and (b) current at load point
8 during physical CB failure.
Fig. 19. Current through (a) CB15 and (b) CB17
during physical CB failure.
act properly as seen in Fig. 21(a) and (b), respectively. During the
faulted condition, RA18 and RA20 take proper action to open their
corresponding CBs to remove the fault from the system while at the
same time both RAs also communicate with the CA which determines
the new system configuration. In such a case in order to remove the
fault, RA20 and RA17 send breaker control signal to their corre-
sponding CBs at 1.018 s and 1.021 s as shown in Fig. 22(a) and (b),
respectively. The current through the corresponding CB20 and CB17
are also shown in Fig. 23(a) and (b), respectively.
At the same time, RA9 and RA10 close their corresponding CBs as
they are now configured as active agents through the CA. From
Fig. 24(a) and (b), it is seen that there is no current flows through CB9
M.S. Rahman et al.
and CB10 until the fault is cleared and the system is reconfigured at
1.021 s. In this case, the open points in the distribution feeder are line
9, load 9 and line 10, and loads 5, 6, 7 and 8 are being supplied from
DG-2. From Fig. 25(a), it can be seen that the active power demand
from DG-2 is decreased by 1510 kW real power and 602 kVAr reactive
power as all of loads in subsystem-2 are supplied from DG-1 instead of
DG-2 after the fault is cleared by the proposed agent-based protection
system. From Fig. 25(b), it can be also seen that the power demand on
DG-1 is increased by 1125 kW real power and 599 kVAr reactive power
since loads 5, 6, 7 and 8 are being supplied from DG-1 in subsystem-1.
Figures Fig. 26(a) and (b) show that load 9 becomes unserved after
117
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 20. Real and reactive power output from (a)
DG-1 and (b) DG-2 during physical CB failure.
Fig. 21. Breaker status for (a) CB18 and (b) CB19
during communication failure.
Fig. 22. Breaker status for (a) CB20 and (b) CB17
during communication failure.
Fig. 23. Current through (a) CB20 and (b) CB17
during communication failure.
1.0 s which cannot be restored as it is disconnected from both DGs since
both CB18 and CB19 failed to open due to physical as well as com-
munication failures. In this case study, the fault clearance and the
power restoration time for the system is obtained as 21.2 ms after the
fault occurred. From the case study, it is obvious that the proposed
approach can work effectively when there is a communication failure.
8. Conclusion and future works
In this paper, a multi-agent framework is proposed for the dis-
tribution feeder protection systems. The proposed solution is compliant
M.S. Rahman et al.
with state-of-the-art standards using the smart grid communication
protocol. Each agent is designed to comply with the requirement of the
protection relay which handles the system reconfiguration and backup
protection through information exchange. The information exchange
through agent collaboration improves the reliability of the system and
reduces the fault clearing time in the system. Various case studies are
considered to investigate the performance of the proposed scheme. The
results obtained from the simulation studies indicate that the agent-
based intelligent relaying scheme is highly reliable and accurate with a
response time of 18–22 ms to clear the fault and restore the power to as
many loads as possible. The simulation results also show the robustness
of the proposed approach properly manage a condition when the pri-
mary breaker fails to remove the fault. Moreover, the proposed scheme
also able to handle the fault situation when there are failures in primary
breakers and the communication link through backup protection
strategy.Although the proposed scheme covers some major issues of a
typical practical protection systems, it does not consider the high im-
pedance fault case which is also worst in distribution networks and one
of the main reasons of powerline bushfire. Therefore, the future scope
of this work will consider such case to comply with the practical cir-
cumstances.
References
[1] Lim S. A service interruption free testing methodology for IEDs in IEC 61850-based
substation automation systems. Int J Electr Power Energy Syst 2017;87:65–76.
118
Electrical Power and Energy Systems 97 (2018) 106–119
Fig. 24. Current through (a) CB9 and (b) CB10
during communication failure.
Fig. 25. Real and reactive power output from (a)
DG-1 and (b) DG-2 during communication failure.
Fig. 26. (a) Voltage and (b) current at load point
9 during communication failure.
[2] Liu Y, Sun P, Wang C. Group decision support system for backbone-network re-
configuration. Int J Electr Power Energy Syst 2015;71:391–402.
[3] Chen C-R, Chang C-J, Lee C-H. A time-driven and event-driven approach for sub-
station feeder incident analysis. Int J Electr Power Energy Syst 2016;74:9–15.
[4] Guardado RA, Guardado JL. A PMU model for wide-area protection in ATP/EMTP.
IEEE Trans Power Delivery 2016;31(4):1953–60.
[5] Sultana S, Roy PK. Oppositional krill herd algorithm for optimal location of capa-
citor with reconfiguration in radial distribution system. Int J Electr Power Energy
Syst 2016;74:78–90.
[6] Mahboubi-Moghaddam E, Narimani MR, Khooban MH, Azizivahed A, shari MJ.
Multi-objective distribution feeder reconfiguration to improve transient stability,
and minimize power loss and operation cost using an enhanced evolutionary al-
gorithm at the presence of distributed generations. Int J Electr Power Energy
Systems 2016;76:35–43.
[7] Jamali S, Bahmanyar A. A new fault location method for distribution networks
using sparse measurements. Int J Electr Power Energy Syst 2016;81:459–68.
[8] Rahman MS, Pota HR. Agent-based smart grid protection and security. In: Hossain
MJ, Mahmud MA, editors. Renewable energy integration: challenges and solutions.
Springer-Verlag: Berlin Heidelberg; 2014. p. 383–409.
[9] Jena MK, Samantaray SR, Panigrahi BK. A new wide-area backup protection scheme
for series-compensated transmission system. IEEE Syst J 2015(99):1–11.
[10] Cintuglu MH, Ma T, Mohammed O. Protection of autonomous microgrids using
agent-based distributed communication. IEEE Trans Power Delivery
2017;32:351–60.
[11] Pang N, Liu H, Huang S, Meng J. Emergency rush repair task scheduling of dis-
tribution networks in large-scale blackouts. Int J Electr Power Energy Syst
2016;82:373–81.
[12] Rahman MS, Mahmud MA, Oo AMT, Pota HR. Multi-agent approach for enhancing
security of protection schemes in cyber-physical energy systems. IEEE Trans Ind Inf
2017;13(2):1–12.
[13] Rahman MS, Mahmud MA, Pota HR, Hossain MJ. A multi-agent approach for en-
hancing transient stability of smart grids. Int J Electr Power Energy Syst
2015;67:488–500.
[14] Rahman MS, Pota HR, Mahmud MA, Hossain MJ. A decentralised multi-agent
M.S. Rahman et al.
approach to enhance the stability of smart microgrids with renewable energy. Int J
Sustain Energy 2016;35(5):429–42.
[15] Wang Y, Wei G, Yang H, Chen H, Ouyang Z, Hill DJ. Novel protection scheme of
single-phase earth fault for radial distribution systems with distributed generators.
IEEE Trans Power Delivery 2016;PP(99). 1-1.
[16] Tarhunia NG, Elkalashyb NI, Kawadyb TA, Lehtonenc M. Autonomous control
strategy for fault management in distribution networks. Electr Power Syst Res
2015;121:252–9.
[17] Elkalashy NI, Lehtonen M. Decentralized earth fault selectivity using transient front
slopes for unearthed mv networks. Electr Power Syst Res 2014;63:908–16.
[18] Eriksson M, Armendariz M, Vasilenko OO, Saleem A, Nordström L. Multiagent-
based distribution automation solution for self-healing grids. IEEE Trans Ind Electr
2015;62(4):2620–8.
[19] Albagli AN, Falcão DM, de Rezende JF. Smart grid framework co-simulation using
HLA architecture. Electr Power Syst Res 2016;130:22–33.
[20] Wang XR, Hopkinson KM, Thorp JS, Giovanini R, Birman K, Coury D. Developing an
agent-based backup protection system for transmission networks. Power Syst
Commun Infrastruct Future. 2002. p. 1–6.
[21] Wan H, Li KK, Wong KP. An adaptive multiagent approach to protection relay
119
Electrical Power and Energy Systems 97 (2018) 106–119
coordination with distributed generators in industrial power distribution system.
IEEE Trans Ind Appl 2010;46(5):2118–24.
[22] Lwin M, Padullaparti HV, Santoso S. Model-based relaying supervision for mitiga-
tion of cascading outages. IEEE Power Energy Soc Gen Meet (PESGM). 2016. p. 1–5.
[23] Rui Q, Qian S. Research on realization of over-current protection based on multi-
agent strategy. Appl Mech Mater 2014;602–605:3009–12.
[24] Osman AH, Hassan MS, Sulaiman M. Communication-based adaptive protection for
distribution systems penetrated with distributed generators. Electr Power
Components Syst 2015;43(5):556–65.
[25] Rahman MS, Mahmud MA, Pota HR, Hossain MJ, Orchi TF. Distributed multi-agent-
based protection scheme for transient stability enhancement in power systems. Int J
Emerg Electr Power Syst 2015;16(2):117–29.
[26] Almas MS, Leelaruji R, Vanfretti L. Over-current relay model implementation for
real time simulation amp; hardware-in-the-loop (HI L) validation. 38th Annual Conf
IEEE Ind Electron Soc. 2012. p. 4789–96.
[27] URL http://mathworld.wolfram.com/FloorFunction.html.
[28] Yanxia Chen, Xianggen Yin, Zhe Zhang, Deshu Chen, Xiangjun Zeng. Research on
protective relaying systems based on multi-agent system. Int Conf Power Syst Tech.
2002. p. 661–5.