Electric Power Systems Research 188 (2020) 106525

Contents lists available at ScienceDirect

Electric Power Systems Research

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

A multi-agent-based integrated self-healing and adaptive protection system T

for power distribution systems with distributed generation
⁎
Felipe C. Sampaio ,a, Ruth P.S. Leãob, Raimundo F. Sampaiob, Lucas S. Melob,
Giovanni C. Barrosoc
a
Department of Electrical Engineering, University of Fortaleza, Washington Soares Av., 1321, Fortaleza, 60811-905, Ceará,Brazil
b
Department of Electrical Engineering, Federal University of Ceará, Campus Pici, 60440-554 Fortaleza, Ceará,Brazil
c
Department of Physics, Federal University of Ceará, Campus Pici, Fortaleza 60440-554, Ceará,Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: The development of self-healing systems in the context of power distribution networks has increased sig-
Adaptive protection nificantly over the last years. However, such feature causes changes in the power grid structure, which may
Self-healing affect the coordination of protection systems. This work proposes a self-healing system (SHS) integrated with an
Distributed generation adaptive protection system (APS) based on multi-agent systems (MAS). This approach aims at minimizing ad-
Distribution networks
verse impacts caused by the network restoration on the protection systems of distribution networks that contain
Multi-agent systems
distributed generation (DG). The SHS uses negotiation among agents to maximize the number of restored loads,
while the agents of the APS reset the protection settings after changes occur in the network topology and state of
DGs. The APS is capable of changing the current setting group and, if necessary, calculating new settings that can
be sent to the relays to ensure the accurate coordination of protection systems. The proposed schemes are
evaluated in a test workbench using commercial digital relays in different scenarios comprising DG penetration
levels and network topology changes. The results clearly demonstrate that this strategy is fast and reliable when
enhancing and providing robustness to the self-healing procedure.

1. Introduction automatically updating the relay settings. The work developed in [10]
proposes an APS for microgrids, which selects the appropriate protec-
Self-healing systems (SHSs) have been successfully applied to the tion settings group (SG) of each relay based on the microgrid connec-
improvement of reliability in power distribution networks [1]. Al- tion mode and respective topology. The APS in [12] monitors the grid
though this solution provides many benefits to both the grid and cus- and selects the most appropriate SG from a list of pre-optimized con-
tomers, it may also cause severe impacts on the grid operation. figurations for each relay. Although the authors in [8] discuss the
When applied to distribution networks, SHSs affect the protection possibility of using online adaptive settings, the introduced APS cal-
system operation, since the grid topology reconfiguration can modify culates optimal SGs for relays using nonlinear programming. A clus-
the power flow direction, the protection zones of relays, and the fault tering technique is applied in [14] to obtain SGs that ensure the pro-
levels. Additionally, the increasing presence of distributed generation tection coordination for all restoration conditions of the electrical
(DG) makes the protection system operation an even more challenging network. All aforementioned approaches are related to online selection
task [2–6]. of offline-recorded SGs. However, in order to provide coordination for
Adaptive protection systems (APSs) have drawn the interest of many all possible network configurations, relays with a large number of SGs
researchers in the last few years due to the influence of DG units on the are required. Therefore, the protection coordination becomes nearly
protection of power distribution networks. Some proposed APSs employ unfeasible in networks with many DG units and several distinct possible
directional overcurrent protection due to its capacity of detecting the configurations owing to the limited number of available SGs.
current flow direction [7–9]. On the other hand, aiming at simplicity Another class of APS is based on the online calculation of new
and cost reduction, other works have introduced adaptive non-direc- protection settings instead of simply changing the active SG adaptively.
tional overcurrent protection schemes [10–15]. The authors in [7] propose an APS for microgrids, which calculates new
Various APSs found in the literature present distinct approaches for settings according to the microgrid operation mode (islanded/grid-

⁎
Corresponding author.
E-mail address: [email protected] (F.C. Sampaio).

https://doi.org/10.1016/j.epsr.2020.106525
Received 4 May 2019; Received in revised form 7 July 2020; Accepted 7 July 2020
Available online 13 July 2020
0378-7796/ © 2020 Elsevier B.V. All rights reserved.
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

connected) and the DG connection status (on/off). The APS in [11] uses
a rule-based algorithm to calculate new relay settings in response to
significant changes in the network. The APS in [16] defines new set-
tings for the protection relays according to the control mode of wind
turbine generators connected to the grid. The approach in [13] is
capable of obtaining modified settings for the relays based on online
estimations of the Thévenin equivalent impedance. Nonlinear optimi-
zation solvers are used in [9] for the calculation of optimum settings
that are sent to the relays in response to significant changes in the grid.
The APS in [15] uses the relay internal logic to reset the pickup current
whenever a change occurs in the grid, thus eliminating the need for
communication with a central control system. The online calculation of
settings is an interesting procedure to provide coordination in-
dependently of the number of SGs, although it is more time consuming. Fig. 1. Distribution network (a) under normal operation and (b) after topology
In networks where several DG units and network topologies exist, this reconfiguration.
strategy aggregates higher computational time.
In this context, the self-healing system, the connection status of DG
units, and the protection system should act together to provide relia-
bility and security to power distribution systems. To the best of the
authors’ knowledge, there are no works reported in the literature so far
focused on the development of an integrated architecture encompassing
the APS, self-healing, and DG status monitoring.
In view of the concerns associated with the existing APSs, this work
proposes a novel multi-agent-based integrated self-healing and adaptive Fig. 2. Impact of the DG in the overcurrent protection.
protection scheme for power distribution systems with DG. The SHS
employs a negotiation strategy to maximize the number of loads that take into account the bidirectional current flow and multiple sources
can be restored. The adaptive protection system updates the SGs of the that contribute to the fault current [2]. The presence of DG units can
relays in response to the DG connection status and self-healing proce- also impair the protection system operation by increasing the operation
dures. The proposed system applies a novel adaptive protection strategy time of overcurrent relays, thus leading to miscoordination.
based on the selection of pre-recorded SGs. New settings are updated Fig. 2 represents part of a typical feeder with two protection relays
when the pre-recorded ones are not capable of ensuring the protection (R1 and R2) and a DG unit between them. A fault is assumed to occur
coordination. This strategy combines the advantages of fast operation downstream to the feeder, thus showing that the currents sensed by
and reliability, provided by the online selection of SGs and calculation each relay correspond to distinct values when the DG is on (Idg) and off
of new settings, respectively. (Ino − dg ).
The remainder of this paper is organized as follows. Section 2 Under the fault condition shown in Fig. 2, when the DG is on, the
analyzes the impacts of the network reconfiguration and DG operation current sensed by R2 is higher, while the one sensed by R1 is lower.
on the protection system. Section 3 describes the proposed multi-agent This situation denotes how the presence of DG can compromise the
system. Section 4 presents the methodology used in the tests and dis- operation of overcurrent relays, causing the so-called blinding protec-
cusses the obtained results. Finally, conclusions are given in Section 5. tion. A mathematical analysis of this problem is derived in [8], which
also comprises a thorough discussion about the major factors associated
2. Impacts of topology reconfiguration and DG on distribution with this phenomenon. If the protection system is set to the condition of
network protection systems the DG connected to the network, but the DG is disconnected, the op-
eration times of R1 and R2 will decrease and increase, respectively,
2.1. Impact of topology reconfiguration causing protection miscoordination. Conversely, the operation times of
R1 and R2 will increase and decrease, respectively. In both configura-
Aiming at the improvement of reliability in medium-voltage dis- tions, the protection system sensitivity is compromised when the DG
tribution network topologies, radial configuration with open tie- connection status changes.
switches located at adjacent feeders is often used. However, the dis- In order to overcome such situation, a feasible solution is to use
tribution network reconfiguration may affect the power flow direction digital relays with multiple SGs. However, it only works well in net-
and short-circuit level, with direct impact on the protection system. works with few DG units. In larger networks, the required number of
Fig. 1 (a) shows the power flow direction and the protection zone of SGs to deal with most of the possible scenarios is unfeasible considering
the relay associated with switch 3 in a radial distribution system con- the use of relays available commercially. Therefore, an interesting al-
sidering the normal operating status. ternative lies in the online calculation of new settings for each relay
The faulty sector in Fig. 1(b) is between switches 1 and 2, which was when the number of active SGs is not enough to ensure proper co-
isolated so that the branch between switches 2 and 3 could be restored ordination.
by closing switch 4. After the restoration, all currents sensed by the
relays change, since relay 3 is now protecting a different sector.
Therefore, the relay settings must be changed, otherwise the relay op- 3. Self-healing and adaptive protection multiagent system
eration time may increase, with consequent selectivity loss and pro-
tection miscoordination. Multi-Agent Systems (MASs) are composed of agents represented by
software and/or hardware entities with specific tasks, which are able to
2.2. Impact of distributed generation interact with each other to achieve a global goal [18]. As a prominent
technique for distributed intelligent control, MASs have been applied to
Protection systems in radial distribution networks are normally the solution of several problems in power systems, e.g., diagnosis pro-
designed to operate under the premise of unidirectional current flow blem, monitoring issues, restoration plan, market simulations, and au-
[17]. With the increasing penetration of DG, distribution networks must tomation [19].

2
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

start negotiation with the DCAs of other substations aiming to restore

the disconnected loads in sections that are not under fault.
The load negotiation process follows the FIPA-ContractNet inter-
action protocol standard [18]. The NA is responsible for sending pro-
posals to the DCAs in each substation that shares a tie-switch with the
feeder under fault. After receiving the proposals, the most suitable one
is implemented. This task is performed according to the maximum
number of loads that can be restored, as represented by the objective
function in (1):
N
max ∑ Wi
i=1 (1)

which is subject to the following constraints:

Ik ≤ Ikmax (2)

Vmin ≤ Vi ≤ Vmax (3)

S ≤ Smax (4)

where N = 1;... ; n is the set of buses connected to the system, Wi is the

load connected to bus i ∈ N, Ik is the current through branch k ∈ B, B is
the set of system branches, Vi is the rms voltage across the bus i; and S is
Fig. 3. MAS architecture. the loading of the substation that sent the proposal. Indices “max” and
“min” denote the maximum and minimum values allowable for each
The proposed self-healing and adaptive protection multi-agent variable, respectively.
system (SAPMS) is based on the integration of two MAS approaches: a If the NA receives more than one proposal capable of maximizing
self-healing system and an adaptive protection system. This arrange- the load to be restored, the one with the lowest loading associated with
ment aims at ensuring the protection coordination of distribution net- the substation transformer will be chosen. After that, the NA sends an
works in the presence of DG units (on/off condition) and under network acceptance message to the DCA of the chosen substation and notifies
topology reconfiguration for load service restoration. the DCAs of the rejected ones. The DCA that receives the acceptance
message requests the APA in its substation to reset the relay settings in
3.1. SAPMS Architecture the feeder where the restored load is supposed to be connected. The
DCA then sends a message to the DA to close the normally-open switch
The SAPMS was developed using the Python Agent Development so as to enable the load restoration.
framework (PADE) [20], which is an open-source platform conceived
for the implementation of MASs in power systems. The architecture 3.3. Adaptive protection scheme
employed in the SAPMS comprises five types of agents: the diagnostic
and configuration agent (DCA), the negotiation agent (NA), the adap- Whenever an APA is requested by a DCA to reset the relays, online
tive protection agent (APA), the distributed generation agent (DGA), load flow and short circuit calculations are performed based on system
and the device agent (DA). data and real-time monitoring information. Traditionally, distribution
According to Fig. 3, the proposed MAS is based on a distributed networks present large unsupervised areas. In this case, state estimation
multi-layer hierarchical architecture. The system is divided in three systems are required in order to identify the present operating state of
levels: device layer, control layer, and decision layer. The device layer the system [21]. Besides, the use of wide-area monitoring systems based
includes DAs and DGAs, which are physically located at the relays and on distribution-level phasor measurement units (D-PMUs) is expected
allow communication among the devices. Such agents are responsible to increase in the near future, thus contributing to the improvement of
for collecting information from relays and circuit breakers, while also the restoration process.
sending commands to their respective relays. After the load currents and the short-circuit contributions of all
The agents of the control and decision layers are physically located branches are calculated, the coordination of relays can be analyzed
at the substations. The control layer is composed of DCAs which are based on the active settings stored in the backup relay. Each feeder is
responsible for receiving information about the grid topology, also re- analyzed, starting from the closest relay to the substation, while the
quiring defined actions from NAs, APAs, DAs, and DGAs. The decision evaluation is carried out on each primary/backup (P/B) pair of relays
layer contains APAs and NAs. The APA is responsible for resetting the upstream to the relay at the tie-switch. The settings of the substation
relays when the DG connection status changes and topology re- relay are kept fixed and used as a reference for the remaining ones
configuration occurs. The NA is in charge of negotiating disconnected along the feeder.
loads with DCAs of other substations. For each pair of relays, the algorithm presented in Fig. 4 is executed,
where the coordination of each SG in the primary relay is tested. If a
3.2. Self-Healing scheme pre-recorded SG capable of providing coordination is found, a com-
mand is sent to activate it. Otherwise, new settings are calculated for
When a fault occurs in the network, an evaluation of the protection the time pickup current (TPC), instantaneous pickup current (IPC), and
operation begins for a proper restoration procedure. The DCA receives dial.
the trip messages of all relays related to the substation and evaluates This strategy enables pre-recorded SGs to be activated in the relays,
the protection coordination. In case of protection miscoordination, the thus reducing the time required to adjust the system protection. In si-
DCA sends a message to the DAs to isolate the fault and close the tuations where pre-recorded SGs do not provide adequate coordination,
switches that have been improperly open due to the miscoordination. the SAPMS calculates new settings, thus ensuring reliability to the
The DCA then sends a message to APA in its respective substation to protection system in a wide variety of possible cases. The algorithms for
reset the relays in the feeder under fault. Besides, they notify the NA to the calculation of settings are shown in Fig. 4.

3
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

Fig. 4. Flowchart of the adaptive protection algorithm.

4. Distribution network and case studies Table 1

Status of the DG units.
4.1. Test system Case DG1 Status DG2 Status

For the test case, real data of a medium-voltage distribution network A Connected Connected
is used, which is represented in Fig. 5. The 13.8 kV distribution network B Disconnected Connected
C Connected Disconnected
is powered by four substations (S1, S2, S3, and S4), with two 15 MW D Disconnected Disconnected
three-phase transformers connected in parallel for supplying S1. Each
one of the remaining substations employs a 10 MW transformer. The
network has seven feeders and 22 switches operated by relays, among Table 2
which six are normally-open switches (3, 5, 8, 13, 14, 17), while 16 Fault location, isolation and network reconfiguration.
switches are associated with load buses (A, B, C, ..., P). The network
Case Fault Open Closed Case Fault Open Closed
also includes two DG units (DG1 and DG2) rated at 5 MVA each. More Location Switches Switch Location Switches Switch
information about the system can be found in [22].
To evaluate the performance of the SAPMS, two different tests are 1 A 1, 2 3 7 J 15, 16 17
2 E 7, 19 8 8 D 6, 7 5
presented as follows, which consider: a) DG status change and network
3 D 6, 7 8 9 G 10, 11 14
reconfiguration; b) SAPMS operation in real-time condition. 4 H 11, 12 13 10 J 15, 16 14
5 G 10, 11 13 11 C 21, 4 3
6 K 16, 20 17 12 F 22, 9 8
4.2. DGs status change and network reconfiguration

In this test, the ability of the SAPMS to reduce the operation time of specified in Table 1. Then, the SAPMS was analyzed in different self-
the protection system was evaluated when compared with the con- healing scenarios as summarized in Table 2, also considering the DG
ventional protection system operation, with fixed protection settings status defined in Table 1.
and without SGs. In the analysis, it was assumed that each relay has up to three SGs.
This test includes two case studies. Initially, the SAPMS perfor- The first SG of each relay was adjusted for the condition in which the
mance was evaluated under different conditions of the DG status as network operates with the normal topology (Fig. 5) with the two DG
units on, i.e., case A in Table 1. The two remaining SGs were defined
taking into account the coordination of the protection system in case of
topological changes according to Table 2, where both DGs are on, i.e.,
case A in Table 1.
In addition, to evaluate the effect of DG penetration on the SAPMS
performance, each test was carried out considering three scenarios of
DG impedance: 0.15 pu, 0.3 pu, and 0.5 pu. The SGs of the relays were
calculated for each scenario.
Figs. 6 and 7 show the operation time of each primary protection
system involved in the situations given in Tables 1 and 2. The perfor-
Fig. 5. One-line diagram of the test system. mance in terms of operation time of the adaptive protection (AP)

4
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

Table 4
Monitoring of SGs.
Case DG Status Change Network Topology Change

0.15 pu 0.30 pu 0.50 pu 0.15 pu 0.30 pu 0.50 pu

ASG 7 (25%) 7 (25%) 7 (25%) 41 (33%) 35 (28%) 41 (33%)

MASG 6 (21%) 4 (39%) 3 (11%) 39 (31%) 44 (35%) 41 (33%)
NS 15 (54%) 17 (61%) 18 (64%) 45 (36%) 46 (37%) 43 (34%)

Fig. 6. Operation time of conventional and adaptive protection schemes for that of the CP system in some tested situations. Although the other two
different connection statuses of the DG units. remaining SGs were designed to deal with different operating condi-
tions, the limited number of SGs is not enough to cover all possible
changes in the network topology. In cases B, C and D, the operation
time of the CP system is increased, but the protection sensitivity was
recovered with the AP scheme. The highest reduction occurred in cases
with the higher impedance of the DG units, except for cases B and C
with ZDG = 0.5 p.u. As previously mentioned, the reduction of the short-
circuit level was not enough so that some relays could trip in the time
overcurrent region. However, due to protection zones changing, the
operation time was reduced in all cases.
Statistical figures of merit are shown in Table 4, which allow eval-
uating the coordination algorithm represented in Fig. 4 properly. This
Fig. 7. Operation time of conventional and adaptive protection schemes under analysis comprises the number of cases for which the active settings
network topology reconfiguration and change of the DG connection status. group (ASG) was capable of keeping coordination, as well as the ones
for which the modification of active settings group (MASG) and the
implementation of new settings (NS) recovered coordination.
system is compared with that of a conventional protection (CP) system
The strategy based on pre-recorded SGs was able to keep the pro-
with fixed settings for different DG statuses. The regions delimited by
tection system properly coordinated in most studied cases. However,
vertical dotted lines show the operation times of the primary protection
new relay settings were calculated in a significant number of cases. This
systems of the feeder relays for each case defined in Table 1.
occurred due to the limited number of available SGs and the complexity
In all cases, the SAPMS sets the relays to operate instantly, reducing
to define settings capable of protecting all likely topological changes
the operation time. Table 3 presents the percent reduction of the total
associated with the DG connection status.
operation time provided by the adaptive protection in each test case:
DG Status Change (DGSC) and Network Topology Change (NTC).
In general, cases B, C and D present longer operation times of the CP 4.3. SAPMS operation in real time
when compared with case A. This increase is due to the disconnection of
the DG units. Since the short-circuit currents are lower, the relays enter The SAPMS performance operating in real-time condition was
the time overcurrent region in the CP scenario. However, when the evaluated in the case of a self-healing procedure, where the protection
adaptive protection system is enabled, the relays operate in the region coordination is reset through the calculation of new settings. The test
associated with instantaneous overcurrent protection. case considers a fault at bus D. In this condition, the restoration system
In order to provide a fair comparison between the AP and CP sys- should isolate the section under fault and choose the best option to
tems, Table 3 shows the influence of the DG impedance. It is observed restore service to buses E and M.
that the higher the impedance, the longer the reduction of the total For this purpose, a testbed with a relay tester, three commercial
operation time provided by the AP system. It is also worth mentioning relays, a communication switch, and a personal computer was used as
that the total operation time corresponds to all primary protection re- shown in Fig. 8. The devices were connected in an Ethernet network
lays. The higher the DG impedance, the lower the short-circuit level through a communication switch. Relays R5, R6, and R19 represent the
and, as a result, the CP system takes longer to operate. Therefore, the relays associated with switches 5, 6, and 19, respectively. In the tests,
AP system is capable of further improving the protection system sen- the relays received new settings via Telnet protocol and were config-
sitivity even in scenarios where the DG participation is small. It is im- ured to exchange generic object oriented substation event (GOOSE)
portant to highlight the condition for which ZDG=0.5 pu in cases B and messages whenever a protection function is enabled; there is a change
C. Although the short-circuit level is decreased, the relay does not enter in the active SG; or new settings are received.
the time overcurrent region. Nevertheless, the worst case is verified The relay tester was used to emulate the distribution network by
when both DG units are off. injecting equivalent fault currents into the analog inputs of the relays,
In the case of network topology change, even in case A, where both which also receive on/off commands in their respective binary inputs.
DG units are on, the operation time of the AP approach is shorter than In addition, the relay tester was used to record the GOOSE messages
exchanged in the network.

Table 3
Reduction of total operating time.
Case DG Status Change Network Topology Change

A B C D A B C D

0.15 pu 0% 77.77% 59.98% 87.93% 39.73% 63.16% 71.18% 81.04%

0.30 pu 0% 78.07% 75.20% 91.27% 55.48% 71.37% 77.19% 87.83%
0.50 pu 0% 0% 0% 92.19% 65.58% 67.60% 71.78% 88.64%

5
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

Fig. 8. Test platform.

The SAPMS agents were embedded in the computer, including the

DAs, which are responsible for sending commands to the relays via
manufacturing message specification (MMS) and new settings via
Fig. 10. Messages sent by the relays and recorded during the test case.
Telnet protocol. The computer was also used for controlling the relay
tester via serial communication.
The communication among the agents during the test is illustrated and 7.16 s, respectively, as well as the command for changing the active
in Fig. 9. NA1 requests proposals for service restoration to the DCAs of SG at 8.88 s. As expected, they coincide with the instants at which the
substations 2, 3, and 4 (messages 4.1, 5.1, and 6.1, respectively). DCA4 signal from SG1 goes to zero, the signal from SG3 goes to 1, and relay
returns a message declining the request (message 6.2). Agents DCA2 R5 receives a command from agent DA6 for closing switch 5, respec-
and DCA3 return proposals for load restoration, informing the trans- tively. The new settings were recorded in the third SG and this is why a
former loading of their respective substations i.e. 78% and 96%, re- command was sent to change from SG1 to SG3.
spectively (messages 4.2 and 5.2). Since the restoration proposal of The time interval for the service restoration from sending the trip
DCA2 is associated with a lower transformer loading, NA1 sends a signal to the closing of switch 5 is 7.85 s. However, the interval required
message accepting the proposal (message 7). DCA1 is also informed to send the three new settings (TPC, IPC, and dial) and the command to
about the network topology change (Message 8). change the active SG is 5.27 s, which corresponds to 66.87% of the total
Upon acceptance of the proposal, DCA2 requests APA2 to adjust the restoration time. It is worth mentioning that the commands are sent
protection system before the reconfiguration occurs (message 9). concurrently to all involved relays. Thus, the number of relays will not
During the coordination analysis, APA2 could not find any appropriate influence the time interval required for communication.
SG in R19, thus assigning new settings to such relay. Finally, APA2 The cases that require the calculation of new protection settings
sends a message to DA5 requesting switch 5 to be closed, as the re- involve longer computational time, which can be regarded as worst
storation is successfully completed (message 10). case conditions. This is the case of the aforementioned test, whose re-
Fig. 10 shows the signals collected by the relay tester during the sults prove that the total restoration time of the protection system is
restoration process. The upper graph represents the fault current within the limit of 180 s, i.e., the maximum time interval that is often
through the relay R6. The next two curves show the GOOSE trip and considered for the automatic restoration of distribution systems[23].
start messages of the instantaneous overcurrent protection function
(50P) sent by R6. The settings curve shows the instants at which relay
R19 receives three new settings, i.e., TPC, IPC, and dial at 3.45 s, 5.59 s, 5. Conclusion

This work has presented an integrated SAPMS applied to medium-

voltage distribution networks with DG. The SAPMS is capable of im-
proving the self-healing procedure by taking into account the evalua-
tion of the system protection coordination.
The SAPMS was tested using data from a real distribution network
and commercial digital relays. Tests were performed in different sce-
narios involving the connection status of DG and topology re-
configuration. Additionally, the real-time operation of the SAPMS was
evaluated when sending new settings to the relays.
The tests have shown that the self-healing system is capable of
maximizing the load restoration considering the system operating
constraints. The adaptive protection system is capable of resetting the
protection system, also reducing the total operation time when com-
pared with the CP system. The importance of online calculation of new
settings has been demonstrated, thus ensuring superior performance
and accurate coordination of the AP system when compared with the
traditional strategy based on pre-recorded setting groups.
The SAPMS approach is capable of restoring disconnected loads
successfully, and when using commercial relays, it could provide new
Fig. 9. Communication diagram. settings to the relays in 7.85 s, which is a much shorter time interval

6
F.C. Sampaio, et al. Electric Power Systems Research 188 (2020) 106525

than that defined as short-term interruption according to standard IEC 1636–1644, https://doi.org/10.1109/TPWRD.2007.909115.
50160 [23]. [7] P. Mahat, Z. Chen, B. Bak-Jensen, C.L. Bak, A simple adaptive overcurrent protec-
tion of distribution systems with distributed generation, IEEE Trans. Smart Grid 2
(3) (2011) 428–437, https://doi.org/10.1109/TSG.2011.2149550.
CRediT authorship contribution statement [8] V.A. Papaspiliotopoulos, G.N. Korres, V.A. Kleftakis, N.D. Hatziargyriou, Hardware-
in-the-loop design and optimal setting of adaptive protection schemes for dis-
tribution systems with distributed generation, IEEE Trans. Smart Grid 32 (1) (2017)
Felipe C. Sampaio: Conceptualization, Methodology, Writing - 393–400, https://doi.org/10.1109/TPWRD.2015.2509784.
original draft. Ruth P.S. Leão: Data curation, Writing - original draft. [9] M.N. Alam, Adaptive protection coordination scheme using numerical directional
Raimundo F. Sampaio: Data curation, Writing - review & editing. overcurrent relays, IEEE Trans. Ind. Inf. 3203 (c) (2018) 1–10, https://doi.org/10.
1109/TII.2018.2834474.
Lucas S. Melo: Data curation, Writing - review & editing. Giovanni C. [10] H. Laaksonen, D. Ishchenko, A. Oudalov, Adaptive protection and microgrid control
Barroso: Data curation, Writing - review & editing. design for Hailuoto Island, IEEE Trans. Smart Grid 5 (3) (2014) 1486–1493, https://
doi.org/10.1109/TSG.2013.2287672.
[11] F. Coffele, C. Booth, A. Dysko, An adaptive overcurrent protection scheme for
Declaration of Competing Interest
distribution networks, IEEE Trans. Power Deliv. 30 (2) (2015) 561–568, https://
doi.org/10.1109/TPWRD.2013.2294879.
The authors declare that they have no known competing financial [12] M. Singh, T. Vishnuvardhan, S.G. Srivani, Adaptive protection coordination scheme
for power networks under penetration of distributed energy resources, IET Gener.
interests or personal relationships that could have appeared to influ- Transm. Distrib. (2016) 3919–3929, https://doi.org/10.1049/iet-gtd.2016.0614.
ence the work reported in this paper. [13] S. Shen, D. Lin, H. Wang, P. Hu, K. Jiang, D. Lin, B. He, An adaptive protection
scheme for distribution systems with DGs based on optimized Thevenin equivalent
parameters estimation, IEEE Trans. Power Deliv. 32 (1) (2017) 411–419, https://
Acknowledgements doi.org/10.1109/TPWRD.2015.2506155.
[14] M. Ojaghi, V. Mohammadi, Use of clustering to reduce the number of different
This study was financed in part by the Coordenação de setting groups for adaptive coordination of overcurrent relays, IEEE Trans. Power
Deliv. 33 (3) (2018) 1204–1212, https://doi.org/10.1109/TPWRD.2017.2749321.
Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - [15] R. Jain, D.L. Lubkeman, S.M. Lukic, Dynamic adaptive protection for distribution
Finance Code 001. systems in grid-connected and islanded modes, IEEE Trans. Power Deliv. 34 (1)
(2019) 281–289, https://doi.org/10.1109/TPWRD.2018.2884705.
[16] Z. Liu, C. Su, H.K. Høidalen, S. Member, Z. Chen, S. Member, A multiagent system-
References
based protection and control scheme for distribution system with distributed-gen-
eration integration, IEEE Trans. Power Deliv. 32 (1) (2017) 536–545.
[1] H. Farhangi, The path of the smart grid, IEEE Power Energy Mag. 8 (1) (2010) [17] J.M. Gers, E.J. Holmes, Protection of Electricity Distribution Networks, third ed.,
18–28, https://doi.org/10.1109/MPE.2009.934876. Institution of Engineering and Technology, 2011.
[2] B. Hussain, S. Sharkh, S. Hussain, M. Abusara, Integration of distributed generation [18] M.J. Wooldridge, An Introduction to MultiAgent Systems, second ed., Jonh Wiley &
into the grid: protection challenges and solutions, 10th IET International Sons Ltd, 2009.
Conference on Developments in Power System Protection (DPSP 2010). Managing [19] S.D.J. McArthur, E.M. Davidson, V.M. Catterson, A.L. Dimeas, N.D. Hatziargyriou,
the Change, (2010), p. 5, https://doi.org/10.1049/cp.2010.0347. F. Ponci, T. Funabashi, Multi-agent systems for power engineering applications -
[3] K. Kauhaniemi, L. Kumpulainen, Impact of distributed generation on the protection part i: concepts, approaches, and technical challenges, IEEE Trans. Power Syst. 22
of distirbution networks, Eighth IEE International Conference on Developments in (4) (2007) 1743–1752, https://doi.org/10.1109/TPWRS.2007.908471.
Power System Protection 2004 (mV), (2004), pp. 315–318, https://doi.org/10. [20] L.S. Melo, R.F. Sampaio, R.P.S. Leão, G.C. Barroso, J.R. Bezerra, Python-based
1049/cp:20040126. multi-agent platform for application on power grids, Int. Trans. Electr. Energy Syst.
[4] P.T. Manditereza, R. Bansal, Renewable distributed generation: the hidden chal- 29 (6) (2019) e12012, https://doi.org/10.1002/2050-7038.12012.
lenges - a review from the protection perspective, Renew. Sustain. Energy Rev. 58 [21] F. Ahmad, A. Rasool, E. Ozsoy, R. Sekar, A. Sabanovic, M. Elita, Distribution system
(2016) 1457–1465, https://doi.org/10.1016/j.rser.2015.12.276. state estimation-a step towards smart grid, Renew. Sustain. Energy Rev. 81 (2018)
[5] C.J. Mozina, Impact of green power distributed generation - traditional inter- 2659–2671, https://doi.org/10.1016/j.rser.2017.06.071.
connection methods to mitigate reliability issues, IEEE Ind. Appl. Mag. (2010) [22] MyGrid, Grid Examples. https://github.com/grei-ufc/grid-examples.
55–62. [23] H. Markiewicz, A. Klajn, Power quality application guide: voltage disturbances
[6] R.A. Walling, R. Saint, R.C. Dugan, J. Burke, L.A. Kojovic, Summary of distributed standard EN 50160 - voltage characteristics in public distribution systems, Copper
resources impact on power delivery systems, IEEE Trans. Power Deliv. 23 (3) (2008) Development Association IEE Endorsed Provider 5.4.2, (2004), pp. 4–11.