2394 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO.

5, OCTOBER 2018

Distributed Intelligent System for

Self-Healing in Smart Grids
Bruno Silva Torres , Lucas Roberto Ferreira, and Alexandre Rasi Aoki , Senior Member, IEEE

Abstract—More and more consumers of electricity demand ser- the flow of electricity through the network [3]. This concept
vices with the highest quality standards. One of these standards also exists for the distribution systems (DS), which occurs due
that have been required is the continuity of services as well as the to faults or non-scheduled interruptions of energy supply [4].
speedy recovery after a power outage. With the advent of smart
grids, the problems of distribution systems are attacked and sys- Automatic actions can help the DS to minimize the damage that
temic reliability indexes are improved. It occurs as a result of the power interruptions can cause in these systems. These automatic
automation that has been taking place in these systems. This paper actions in DS are exactly the focus of this work.
aims to present a contribution in the area of self-healing distribu- To keep the duration of an interruption under the supervision
tion networks in the event of a permanent short-circuit wherein a of a regulatory agency, indexes of reliability [5] are created,
systemic reconfiguration is necessary. The proposed strategy, based
on distributed intelligence, allows the signals to propagate between which try to reflect the average values of these disruptions to
the recovery switches, without these switches requiring a large consumers, both in the number of interruptions and in the aver-
range of communication. It occurs because the sign of a switch age time of interruption. Standards including maximum values
broadcasts on the network through communication between ad- of these indexes are established and continually revised down-
jacent switches to establish an optimized switching solution. This
ward to increase the quality of services provided by the dealers.
strategy has been tested in a real network and in a large distribution
network for simple, sequential, and multiple short-circuits. However, historically, the values of these indexes could only
Index Terms—Self-healing, distributed control, smart grids,
be effectively reduced with the use of new equipment for the
production-rule system. DS. Thus, auto reclosers started to have some electronic parts
and new coupled functions, which together with the automation
I. INTRODUCTION of the new DS switches allow this problem to be reformulated.
HE increasingly growing demands of society for a power Now, DS switches may also be used conveniently operating
T supply more reliable and with more quality have pushed
power utilities to establish new operating procedures, new au-
through computer programs, being open or closed remotely,
without the need for local human intervention, reducing the
tomation schemes, and more flexible infrastructures. Smart grids duration of an outage.
appear in this context, to meet consumers’ demands more effi- This also was only possible because the electronic devices,
ciently, with high degrees of continuity and quality [1]. Smart available in DS, can provide data to a central system of analysis
grid infrastructures have allowed an existing area in the studies that infers where the source of the outage is. This central system
of electric systems, called systemic restoration [2], which could collects data from the DS, performs a diagnosis, and prepares a
be increased by the introduction of other possibilities for energy plan of action, which is composed of a set of rapid switching in
supply and the automation of electrical networks. an orderly manner [6]. Then, communication signals are raised
The advent of smart grids provided an integration of infor- from the central system to the switches that perform the actions
mation and communication technology with the electric system, commanded (opening or closing). This new procedure of self-
resulting in increased efficiency and reliability and stability of healing has decreased the values of the indexes of continuity.
the system. Self-healing is one of the features of smart grids, The development of electronic technologies continues, and
i.e., a smart automation switching scheme intended to restore the DS is still growing in scale and complexity. At this point,
there is a need to have not only a central system of analysis,
Manuscript received September 3, 2017; revised March 24, 2018; accepted but also that this analysis can be distributed by DS’s equipment.
May 18, 2018. Date of publication June 8, 2018; date of current version Septem- This development requires new algorithms to be incorporated
ber 25, 2018. Paper no. TPWRD-01060-2017. (Corresponding author: Alexan-
dre Rasi Aoki.)
into network equipment, so that they can perform these actions.
B. S. Torres is with the Lactec Institutes, Curitiba 81530-180, Brazil, and It is exactly at this point that this job fits. This paper proposes
also with the Gnarus Institute, Itajuba 37500-052, Brazil (e-mail:, bs_torres@ algorithms wherein the equipment of a DS may notice a de-
hotmail.com).
A. R. Aoki is with the Lactec Institutes, Curitiba 81530-180, Brazil, and
fect, remove the affected area, and produce a restoration plan,
also with the Federal University of Paraná, Curitiba 80060-000, Brazil (e-mail:, without the need of a central system, using only the distributed
[email protected]). intelligence in the switches.
L. R. Ferreira is with the Siemens, Curitiba 80020-030, Brazil, and also
with the Federal University of Paraná, Curitiba 80060-000, Brazil (e-mail:,
Thus, the research problem can be formulated as follows:
[email protected]). the production of algorithms to run on-board the equipment
Color versions of one or more of the figures in this paper are available online (normally) distributed in typical DS, i.e., with radial features
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPWRD.2018.2845695
and distributed generation and voltage regulators [7].

0885-8977 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 TORRES et al.: DISTRIBUTED INTELLIGENT SYSTEM FOR SELF-HEALING IN SMART GRIDS 2395

Fig. 2. Flowchart with the three steps of the self-healing process. Note:
Fig. 1. Hypothetical DS with its initial configuration, pre-fault values, and Sections II-D, II-E, and II-F show these steps in details.
short-circuit location.

algorithm is based. “Group” is a topological set formed by a

The scientific literature is lavish in strategies, methodologies, segment of supply feeder (named here as “branch”) plus one or
procedures, and algorithms for self-healing [5], [8]–[10]. Most more switches. Thus, in any radial distribution network, each
of them involve numerical techniques [11]–[14], meta-heuristics branch belongs to only one group, substation switches belong
[15]–[17], and intelligent systems [6], [18]–[21], trying to get a to only one group, and switches located in the network always
maximization (minimization or whatever the case may be) of a belong to two groups. This division between substation switches
given function-global goal, which can be the maximization of and (network) switches is important because the proposed algo-
revenue, maximization of the quality of service, minimization rithms for each one are different.
of load shedding, or minimization of de-energizes area. Aiming The distribution network shown in Fig. 1 has 14 groups. In
to this new smart grid scenario, there are methodologies in- fact, the number of groups is the same as the number of branches.
cluding distributed energy resources (renewable energy sources
and storage) and microgrids [22]–[24]. For this, the system is B. Logic of the Proposed Algorithm
seen to be a unified and centralized way, which is oriented for a
distributed solution, much closer to the multi-agent techniques The proposed algorithm uses only a distributed solution with-
[25], [26]. out the need of a central system for command and observation.
The solutions are performed on each switch individually, but
II. SELF-HEALING ALGORITHM PROPOSED the final solution for a specific situation is unique. The logic
Before presenting the proposed self-healing algorithm, a the- of the process can be divided into three main steps: (a) de-
oretical distribution network with all the characteristics of real tection of the short circuit, (b) location and isolation of short
networks is introduced in Fig. 1 to illustrate the concepts of this circuit, and (c) restoration of the branch without energy (and
algorithm. The normal configuration of this distribution network also without fault). Fig. 2 shows this flowchart. In the first part,
is composed of one distribution substation (blue rectangle) with the switch detects and computes the number of abnormalities
5 circuits, 5 substation switches, 16 network switches (7 nor- (short-circuit current or under-voltage depending on the loca-
mally open and 9 normally closed), and 14 branches (distribu- tion of each switch, substation, or network, respectively) and
tion network segment between one or more switches). In this when this number reaches a predetermined value, the switch
example, the short circuit is permanent and it is located in the opens, and the algorithm passes to the next step. In the second
branch between switches S13, S15, and S16. step, location and isolation of the short circuit, the algorithm
For simplicity, in this analysis, the same nominal capacity for checks if the short circuit is in the branch of its group: (a) If the
all switches and all branches is adopted to be equal to 300 A. answer is “yes”, the switch locks and terminates its participation
The values in Fig. 1 present the pre-fault conditions. The values in the process of self-healing, or (b) if the answer is “no”, the
inside the parentheses are the load in the branch (i.e., the current algorithm passes to the next step. In this third step, the switch
that feeds the loads that are in the branch - this value is not checks when and whether it takes part in the process of network
measured), whereas the values inside the brackets are the switch restoration. This process is done in a decentralized manner and
loading (the measured value, they expressed the current through the algorithm can be embedded in any automated switch without
the switch). These values are used in the restoration. All the the need for complex or speedy communication systems.
values of loads in this paper are expressed in current (A). The proposed algorithm allows each switching decision to be
taken by restricting communication between adjacent switches
only, into a mesh communication infrastructure. Even so, the
A. Concept of Group
operating conditions of the entire network required for the solu-
Before presenting the proposed algorithm, it is important to tion of the problem are observed, taking into consideration their
discuss the concept of group (or block or team) on which this pre-fault conditions, operating limits, and systemic availability.

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 2396 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

In the end, the set of individualized switching forms a systemic At the end of this step, with the permanent short-circuit at the
integrated restoration plan, where the change of position of a location indicated in Fig. 1, the substation switch S14 already
switch depends on the operating conditions observed by other detected the short circuit, tried the number N of attempts to
switches, often far away from it electrically and geographically. reclose, and was not successful to open the breaker, eliminating
This depth of search is achieved because, during the process of the short circuit. The network switch S15 realized the short
restoration, several transfers of data are carried out, setting up a circuit and also checked N + 1 sags. Network switches S9, S16,
comparison of the various search trees, where the nodes are the S21, and S13 verify N + 1 sags and did not realize the short-
switches and the branches of the search tree are the branches of circuit current. Thus, all these network switches pass to the next
the distribution network. step of the algorithm. The other switches continue with its initial
state unchanged, not coming out of this step.
C. Definition of Pre-Inserted Data and Measured Values in the
Switches E. Step of Permanent Short-Circuit Isolation
Each switch of the network must have pre-inserted its own In this second part of the process of self-healing, the goal is to
operation algorithm, namely, for substation switches or network detect in which branch (or group) the short circuit is occurring.
switch, and the following pre-inserted data: normal switch po- This is done by rules that verify the internal status of the counters
sition (“normally open” or “normally closed”), the information of the same group. For that, it is used the record of overcurrent
about its group(s) where it belongs (i.e., the adjacent switches), of one switch and the same record of the other switches of the
and smaller nominal capacity among the switches of each group same group.
(usually expressed in Ampère). If two switches belonging to a same group have overcurrent
The measured values for the substation switch are the branch record, this is because the short-circuit current is “passed” by
current and the number of its own openings, whereas the network this group (branch), and it is possible to infer that the short
switches measure the current through the switch and the sags in circuit is not in this branch of the network. However, if only one
each of its terminals. switch presents this record in its group, the short-circuit current
In terms of communication, the switches transfer the value entered but not got out of this branch, thus it is possible to infer
of the remaining capacity (smaller nominal capacity minus the that the short circuit is in this branch of the network. What’s
current through the switch) in its group and await, when appro- more, all switches of this group must remain open and locked
priate, for the return signs to continue the implementation of its in this position; otherwise, in the restoration of the network, the
algorithm, as seen below in the steps of the self-healing process. short circuit signals back to be supplied.
Thus, the rule of detection of the short-circuit location can be
described as follows: “if only one switch of the group presents
D. Step of Permanent Short-Circuit Detection overcurrent record, then the short circuit is located in this branch
In the short-circuit detection step, the substation switches of the network; otherwise (i.e., no switch in one group or two
are responsible for checking if the short circuit has transient switches in the same group have overcurrent record), the short
or permanent origin through its opening. For that, in this first circuit is not located in this branch.
part of this step, there is a short-circuit current counter (opening For the implementation of this operative situation of the short-
counter) in these switches. Thus, the following rule is proposed: circuit location, the switches of the same group should com-
Rule 1: If the overcurrent relay detects an occurrence, check municate among them (in fact, there exists a communication
the openings: if opening counter < N + 1, then [1. open switch, between adjacent switches only). This is an important feature
2. record the event of overcurrent, 3. increment the opening of the proposed algorithm: the necessity of communication be-
counter, and 4. launch an internal timer for a new switch closing]; tween the adjacent switches only because the communication
if opening counter = N + 1, then [1. open switch], where N is capacity can be reduced. Thus, the rule that will be used for the
the preset number of attempts to restart the power utility usually substation switches can be written as follows:
uses (normally N = 3). Rule 2: If there is an overcurrent record AND no other switch
For network switches, there are two sensors that exist: one for of its group has overcurrent record, then [1. end restore logic,
overcurrent and another to voltage sag. When there is a short- 2. set status “locked” switch, and 3. wait until the switch is
circuit current, the overcurrent sensor is actioned; when a sag externally released].
is detected, the sag counter is incremented. Then, when the sag For the network switch, there are two possible ways depend-
counter achieves an N + 1 number of sags, the execution of the ing on whether the switch recorded (or not) the overcurrent. If
algorithm should proceed. Thus, the following rules should be the switch recorded the overcurrent, the rule is similar to the
followed in this order: substation switches (Rule C); however, if it did not register the
Rule A: If the overcurrent sensor detects an occurrence, then overcurrent, it is necessary to check how many adjacent switches
[1. register in the overcurrent record]. recorded overcurrent. If only one switch recorded the overcur-
Rule B: If the sag sensor detects an occurrence, then check rent, it is because the short circuit occurred into this group and
the sag counter: if sag counter < N + 1, then [1. increment the switch must be locked (Rule D). There are two other pos-
the sag counter]; if sag counter = N + 1, then [1. continue the sibilities: no switch recorded the overcurrent in the group or
algorithm]. two switches recorded the overcurrent in the group. In the first

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 TORRES et al.: DISTRIBUTED INTELLIGENT SYSTEM FOR SELF-HEALING IN SMART GRIDS 2397

case, the short-circuit current did not pass through this group; same and the processing time continues within the same order
whereas, in the second case, the short-circuit current passed by of magnitude of the search algorithms of degree 1 (within the
this group, but the defect was not located in this group. In both same group) but searching in all branches of interest for an
cases, the switch should not be operated by these rules. optimized solution.
Rule C: If there is an overcurrent record AND no other switch Thus, a more complex logic is based on the idea that a net-
of its group has overcurrent record, then [1. end restore logic, work switch to contribute with the restoration process must be
2. set status “locked” switch, and 3. wait until the switch is of one of two types: (a) possess voltage in just one of its termi-
externally released]. nals or (b) have data transferred to another switch. The switch
Rule D: If there is no overcurrent record AND only one switch with voltage in just one of its terminals is open and ready to
of its group has overcurrent record, then [1. end restore logic, close the feed with energy its other branch. The switch that had
2. set status “locked” switch, and 3. wait until the switch is transferred data is a switch without voltage in its two terminals
externally released]. but “represents” another switch with only one terminal voltage.
Therefore, continuing the example, the group that contains The transferred data define the optimization criteria adopted,
the short circuit is formed by switches S13, S15, and S16 and which can be, for example, the remaining capacity on the cir-
its connector branch. With these rules, the substation switch cuit, i.e., any difference between the nominal capacity and the
S14 verifies that the network switch S15 also noted the short- loading, calculated for each branch and switch involved in the
circuit current, so S14 is able to continue the algorithm. Switch solution. In the example, if this remaining capacity on switch S9
S15 verifies that, by its group with S14, it might be able to is 200 A (= 300–100, it means that its nominal capacity minus
close; however, by its group with S13 and S16 (which did not the pre-fault loading), when this value is transferred to switch
observe the short-circuit current), it must open and locked in this S17, this value should be reduced in the pre-fault loading branch
position. Switches S13 and S16 did not observe the short-circuit between switches S9, S16, and S17, which, in the example, is
current, but as the switch S15 observed these switches should 30. Thus, switch S17 is able to represent switch S9 in a dispute
also open and locked. Switches S17 to S21 did not observe the with a transferred value of 170 A.
short-circuit current and any switches of their groups, and they A similar fact occurs with the transfer between the following
are ready to continue to the next step of the algorithm. switches: from S18 (270 A) to S20 (230 A) and from S21
Observe that the final configuration after this step presents (140 A) to S19 (120 A) and S20 (120 A). Thus, there is a
the switch S14 open, and switches S13, S15, and S16 are open dispute on switch S20 because it has data transferred from S18
and locked. and S21.
A dispute is made in two parts using the transferred data from
one group trying to advance to the other group. These two parts
F. Step of Restoration are the switch conquest and the conquest of the next branch.
In this step, the switches are seeking to energize the major Many disputes can occur during the self-healing process. The
number of branches without power to get the better balance of order of the disputes is irrelevant for the final reconfiguration
the new load distribution according to the loading conditions solution. This order depends on the internal processing of infor-
and availability pre-fault. The switches, which passed by the mation of each switch and of its communication system. Using
previous step, are able to contribute in the process of restoring the illustrative example, one of the first disputes occurs in switch
the network. In the example, the switches are S9, S14, and S17 S20, which has transferred values from S18 and S21 (230 and
to S21. 120 A, respectively).
The rule for the substation switches is simple: if this switch The first part of the dispute, the switch conquest, compares
can help with the restoration process, it must be closed (Rule 3). the values of the transferred data for the switch. In our example,
In the example, the switch S14 is closed to supply energy to the the value 230 A (transferred by S18) is bigger than 120 A
branch between S14 and S15 (which is open and locked). (transferred by S21) and wins the first part of the dispute. In
Rule 3: If the substation switch is not locked, then it must be the second part of the dispute, switch S18 (Circuit D) tries to
closed. supply energy to the branch between switches S19, S20, and
For network switches, the restoration actions are more com- S21. To accomplish this goal, the transferred capacity in switch
plex and involve communication between them. As the com- S20 must be bigger than the capacity to supply of switch S21.
munication occurs only with adjacent switches (i.e., the same After supplying the branch between S18 and S20, the capacity
group), the ability to search on a better solution can be re- of S20 to supply this new branch is 230 A, whereas the capacity
duced. However, in the proposed algorithm, a viable solution to supply this branch of switch S21 is 140 A. As the capacity to
to this problem is to create a new logic for the switches, which supply of S20 is bigger than S21, the branch between switches
involves information data exchanges among groups using the S19, S20, and S21 can be supplied by Circuit D. However, it does
switches that belong to these two adjacent groups, without any not occur immediately because S20 verifies that it can transfer
improvement or extension of the communication system. its data to switch S19, and new disputes continue to occur.
It is important to notice that the proposed solution has a The problem now is to know when this process finalizes and
more complex algorithm of data exchange. The complexity of the final reconfiguration is achieved. It occurs when the switch
this logic does not refer to its execution but is related to its that transferred data receives a return signal from the switch that
conception, making the processor able to continue to be the received the transferred data. Although it does not receive the

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 2398 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

execution of the algorithm for this group and verify the

other group; (b) If no, then move to action 2.
2) Check if there are other switches in the same group with
voltage in just one of its terminals: (a) If yes, then verify
if the capacity of the other switch is the lower than its
own load limit (establishment of a dispute) and then: (i)
If yes, then move to the next action; (ii) If no, finalize
the execution of the algorithm; (b) If no, then move to
action 3.
3) Check if there is for a switch without voltage in its ter-
minals: (a) If yes, then transfer data to this switch and
wait for the return signal when all signs return to close the
switch; (b) If no, then close the switch.
Fig. 3. Hypothetical DS with the final reconfiguration. Rule F: If the switch had transferred data, then
1) Verify if the switch that transferred data belongs to this
exact number of return signals equal to the transferred data, the group and then (a) If yes, finalize the execution of the
switch remains in the same status. The return signal is emitted algorithm for this group and verify the other group; (b) If
when the dispute is finished. In some cases, where its transferred no, then move to action 2.
data are defeated in a dispute, the switch never receives a return 2) Verify if there are switches on the same group that have
data. It is the case of switch S21, which is open and continues to voltage in just one of its terminals and then (a) If yes, then
open, because other switches (and circuits) are better to supply establish a dispute and then (i) If it wins all disputes, then
energy than it. move to action 3; (ii) If not, finalize the execution of the
Just to exemplify a final dispute, observe the dispute in switch algorithm; (b) If no, move to action 3.
S17. It can receive data from switches S9 and S19. The capacities 3) Verify if there are switches on the same group that there
of these switches are 200 and 210 A (= 230–20, which means the is no voltage in its terminals and then (a) If yes, then
capacity of S20 minus the load of branch between S19, S20, and verify if this switch has transferred data, (i) If yes, then
S21), respectively. Transferring data from the switches S9 and establish a dispute and then: (j1) If it wins all two parts of
S19 to S17, the values received are 170 and 185 A, respectively. the disputes, then transfer data to these switches; (j2) If it
The first dispute is the switch conquest. As 185 A is bigger than loses all parts of one dispute, then finalize the execution
170 A, this first part of the dispute is won by S19. However, of the algorithm; (j3) If it wins the first part of a dispute
when this switch tries to conquest the next branch in the second and loses the second part, then open the switch and send
part of the dispute, it losses because 185 A is not bigger than a return signal; (ii) If no, then transfer data to this switch
200 A (capacity of S9). At this moment, as there are no more and wait for the return signal; when all signs return, send
switches to transfer data, the algorithm of S17 opens the switch a return signal to the switch that received the data; (b) If
and sends return signals to switches S9 and S19, which have no, then send a return signal to the switch that received
transferred data. the data.
On the one hand, when S9 receives the return signal, this
switch compares to the number of transferred data. As this num-
G. Need for Timings in Practical Implementations
ber is equal, the algorithm of S9 closes the switch. On the other
hand, when S19 receives the return signal, it compares to the The major difference among centralized control system and
number of transferred data. As this number is equal, S19 sends distributed systems is the possibility to schedule each action of
a return signal to S20 (which received the transferred data). each device in a common timeline. In centralized systems, it is
When S20 receives this signal, it makes the same comparison possible to order an action before (or after) another action. In
and sends a return signal to S18. When S18 receives this signal distributed systems, this type of order is hardy to do or even
from S20, it provides the comparison to the number of trans- impossible. For that reason, one possible solution is to provide
ferred data. As it is equal, this switch closes. Fig. 3 shows the each control algorithm with timers (temporizations); it means,
final reconfiguration of the network, where the branch between time lapses where the control of the device waits before taking
S9, S16, and S17 is supplied to Circuit E due to the closing of an action. In the proposed algorithm, three timers are important
switch S9 and the branches between S18 to S20, S19 to S21, to be included for a better performance of the algorithm, which
and S17 to S19 are supplied to Circuit D due to the closing of are located in rules 1, 3 and F.
switch S18. In Rule 1, a timer must exist among each opening of the
Thus, the rules governing this step in the process of self- substation switch during the verification if the short-circuit is
healing are presented below: permanent or not. It happens because, for the extinction of a
Rule E: If the switch has voltage in just one of its terminals, short-circuit, a certain time without any type of energy must
then occur. This time depends on the characteristics of the network,
1) Check if there is another switch in the same group with the soil, the existing distributed generation, and also the presence
voltage of two terminals: (a) If yes, then finalize the of large motors.

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 TORRES et al.: DISTRIBUTED INTELLIGENT SYSTEM FOR SELF-HEALING IN SMART GRIDS 2399

proposed algorithm, switches CR12 and CR13 transfer data to

switch C21, settling a dispute on this switch. The values of
these data are 313 and 195 A, respectively. It turns out that
switch CR12 wins both parties of the dispute, making branches
Alim12 and Alim13 to be supplied by Circuit 2 by the closing
of switch CR12.
Now, by applying a short circuit in branch Alim21 to this new
configuration, switches DJ2 and C12 are opened and locked to
isolate the short circuit, and branches Alim12, Alim13, Alim22,
and Alim23 get without energy. Then, switch CR13 transfers
data to switch C21, whereas switches CR23 and CR24 transfer
Fig. 4. Real DS used for testing. data to switches CR12 and C22, settling a dispute on these
switches. The transferred data are 248 and 206 A, respectively.
In Rule 3, a timer is necessary to give time for the other
It turns out that switch CR23 wins both parties of the dispute
switches complete their isolation step in the algorithm; it means,
and transfers data to C21, with a value of 89 A (= 505–257–
for the network switches have time to open and lock, isolating
41–46–72), which already has a value of 193 A that has been
the short-circuit.
transferred by CR13. The dispute is won in two their parts by
In Rule F, a timer must be implemented before the opening
switch CR13. In this way, switch C21 transfers data to switch
of the switch and sending the return signal. The reason for
CR12, with a value equal to 121 A; however, switch CR12
that is to wait for all other possible disputes (transferred data)
already has data from CR23 (161 A). The dispute occurs with
which can be arrived in this switch. Depending on the network
a split decision in CR12, i.e., transferred data wins one part
configuration, the switch processors, and the communication
of the dispute. Therefore, switch CR12 opens and sends return
system, the transferred data for disputes can arrive in different
signals back to CR23 and C21. Switch CR23, which has hosted
times and orders. In such way, to ensure that all data arrived
the return signal from C22, closes, energizing branches Alim22
(that is, that all disputes have occurred) the easiest way is to set
and Alim23. Switch C21 sends a return signal to CR13, which
a temporization in the switch.
closes, energizing branches Alim12 and Alim13.
For all these situations, the setting of the timer in some sec-
This example can be used for a comment of great importance
onds is sufficient.
in distributed systems. With the proposed approach, the algo-
rithms are processed individually without a central control, and
III. TEST ON A REAL DISTRIBUTION NETWORK there does not exist any type of control over what action is taken
A. Real Distribution Network Configuration before. In the implementation above, after the first match in
switch CR12, data are transferred to CR23 and CR24, won by
The real distribution network [27] used in this illustrative nu-
CR23, and it transfers data to switch C21. However, the imple-
merical example is located at Rio de Janeiro, the second largest
mentation of algorithm on switch C21 was faster, which seems
Brazilian city (Fig. 4). This network is composed of 2 substa-
to be the case, as no dispute is settled on it (only one transfer),
tions, 5 circuits, 5 substation switches (DJ), 15 branches (Alim),
and switch C21 would be the switch that would transfer the data
10 normally closed network switches (C), and 6 normally open
to CR12, eliminating the dispute that occurred in C21 and pass-
network switches (CR); and with 493 distribution transform-
ing it directly to the second dispute in CR12, obviously with the
ers (13.8 kV/220–127 V), which, for the sake of simplicity, are
same final result. Thus, the proposed algorithm is independent
not represented in the figure. The total load of this network is
of the runtime of each switch processor and even by several
equal to 28 MVA and maximum rated capacity of loading for
paths always arrives at the same result, optimized locally.
all branches is 505 A. The values inside the parentheses are
the load in the branch (this value is not measured), whereas the
values inside the brackets are the switch loading (the measured C. Tests With Simultaneous Short-Circuits
value). The results of representative tests for sequential or si-
This section presents two tests conducted with simultaneous
multaneous short-circuits are carried in the next sections. Single
short-circuits applied in the distribution network shown in Fig. 4
short-circuits are not presented because they are a sub-case of
using the base-case data: (a) short circuits in branches Alim21
sequential short-circuit.
and Alim31 and (b) an internal short-circuit in Substation 2,
opening and locking all the breakers of this substation.
B. Tests With Sequential Short-Circuits
In the first test, short circuits in branches Alim21 and Alim31,
Sequential short-circuits are a sequence of single short- switches DJ2, C12, DJ3, and C13 are opened and locked to
circuits, one after the other. The studies presented in this section isolate the short circuits and fought to isolate the short circuits,
begin with a short circuit in the branch Alim11 and then short and branches Alim22, Alim23, Alim32, and Alim33 get without
circuits are tested in Alim21, Alim23 and Alim43 branches. energy. Switches CR12, CR13, CR24, CR34, and CR35 are
When a short circuit occurs in branch Alim11, switches DJ1 able to start the restoration process. Switches CR12 and CR24
and C11 are opened and locked to isolate the short circuit, and transfer their data to switch C22, with values of 178 A (= 219–
branches Alim12 and Alim13 get without energy. Using the 41) and 165 A (= 206–41), respectively. A dispute occurs in

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 2400 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

switch C22, where CR12 wins. As switch C22 has no one to TABLE I
MAXIMUM LOADING CAPACITY FOR EACH TESTED APPROACH
transfer its own data, it sends a return signal to CR12, which
closes feeding branches Alim22 and Alim23. At the same time,
switches CR13, CR34, and CR35 transfer their data to switch
C23.The two first transfers are running a dispute and the winner
has a new dispute against the third transfer. The transferred
values are 89 A (= 248–159), 47 A (= 206–159), and 195 A
(= 260–65), respectively. After the two disputes, the decision is
split among switches CR13 and CR35. Thus, switch C23 opens
and sends return signals to CR13 and CR35, which close and
cases from all network were supplied; while in the second ap-
energize branches Alim33 and Alim32, respectively, by Circuits
proach, the distributed CBR, for each switch, only its own cases
1 and 5.
were supplied.
In the second test, there is a short circuit in Substation 2, open-
The comparison of the results is made numerically, through
ing and locking all the substation switches and depriving the
comparison of the results of tests, and conceptually, through
network to be supplied for this substation. Therefore, branches
other points relating to the mathematical methods and opera-
Alim31, Alim32, Alim33, Alim41, Alim42, Alim43, Alim51,
tional characteristics.
Alim52, and Alim53 are without energy. Switches CR13, CR23,
and CR24 are able to assist in the self-healing process. Switches
CR13 and CR23 transfer data to C23, with values of 60 A (= A. Comparison Using Real Distribution System
219–159) and 216 A (= 375–159), respectively. A dispute is For numeric comparison, an extreme case with complex so-
established in C23, and the winner is switch CR23. As switch lution is chosen to be presented. In this case, the substation 2
C23 has no other switch to transfer its data, it sends a return of Fig. 4 presents a blackout, thus its energy is lost, and all
signal to CR23, which closes and energizes branches Alim31, substation switches (DJ3, DJ4 and DJ5) open and lock. Other
Alim32, and Alim33 by closed switches C13 and C23. What’s comparisons have been made, but for non-extreme cases, and
more, to feed branches Alim32 and Alim33, they energize one the differences between the solutions of the proposed algorithm
of the poles of switches CR35 and CR34, respectively, and they and the others are not significant.
can contribute to the energization of the other branches. Switch In terms of switching, the solution of the proposed distributed
CR35 transfers its data to C25 and this, in turn, transfers data algorithm is the closing of switches CR23, CR24, and CR35. For
to C15, which not having one to transfer data to sends a return the distributed CBR approach, the solution is closing the follow-
signal to C25 that then transmits its return signal to CR35, which ing switches: CR23, CR24, CR34 and CR35; and opening the
closes, energizing branches Alim51, Alim52, and Alim53. switch C24; while the centralized approaches (GA, PSO, HPSO,
Switch CR34 transfers data to C14 and C24 and then switch and centralized CBR) presents a solution where the switches
CR24 also transfers data to switch C24, settling a dispute. The CR13, CR24 and CR35 must be closed. Table I shows the branch
result of the dispute is divided; thus, switch C24 opens, and with greater loading in each circuit after each reconfiguration.
switch CR24 energizes branch Alim43, whereas switch CR34 It is observed that overload occurs in all solutions (because the
energizes branches Alim41 and Alim42.Thus, the restoration limit of the extensions is 505 A); however, this overload was
process is complete even with overloaded in the circuits of already provided for, as it is an extreme case of self-healing.
Substation 1. Simple rules may exist to disallow overloads or just The solution presented for distributed CBR presents a worse
certain maximum values, but this is intrinsically linked to the division of the loading among the two remained circuits than
operational philosophy of the distribution utility. Many times, the proposed approach; while the solution proposed by the cen-
overloads are accepted, especially in extreme cases, as in the tralized method presents the best global load division. It is pre-
case of this example. dictable that the centralized approaches present better solutions
in extreme cases, because one of the judging criteria of fitness
function is the overload of the branches, producing the restored
IV. COMPARISONS WITH OTHER TYPES OF IMPLEMENTATIONS network (after the switching) with a better division of load-
The comparison between the results obtained using the pro- ing between their circuits. This criterion was not present in the
posed self-healing approach and the other existing methodolo- proposed algorithm.
gies is made through the real distribution system presented in However, this example has been exactly chosen because it
the previous section and the large distribution network shown is possible to present a feature of the proposed algorithm. The
in [28]. In this comparison, different approaches, such as ge- difference between the answers from the proposed algorithm and
netic algorithm (GA), particle swarm optimization (PSO), hy- the centralized algorithms is which circuit 1 or 2 will supply
brid particle swarm optimization (HPSO), and case-based rea- energy to Circuits 3 and 5. Circuit 4 is supplied to Circuit 2
soning (CBR) are used, based on the approaches represented in through switch CR24 for in both answers. In the centralized
[28]–[30], respectively. GA, PSO and HPSO were implemented solution, Circuits 3 and 5 are supplied by Circuit 1, the switch
in a centralized self-healing structure; while the CBR were CR13 must be closed; however, if these circuits are supplied
implemented in two different approaches: centralized and dis- by Circuit 2, the switch CR23 must be closed. The best global
tributed. In the first approach, the centralized CBR, restoration solution is to close switch CR13 because the Circuit 2 will be

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 TORRES et al.: DISTRIBUTED INTELLIGENT SYSTEM FOR SELF-HEALING IN SMART GRIDS 2401

TABLE II
SOLUTIONS FOR TESTED APPROACHES

Fig. 5. Large DS used for testing [25].

tion, the search space, and the number of solutions; as a result,

supplied by Circuit 4. However, in the proposed algorithm, when making it difficult to find the optimal solution. It is composed by
a dispute occurs between CR13 and CR23, the capacity at this three main circuits, which, for the purposes of explanation, are
moment of the switch CR23 is bigger than the switch CR13, called right, middle and left circuits, which have the following
and for this reason CR23 wins the dispute locally. initial branches 1–91, 1–122 and 1–189, respectively. In that
In the conceptual comparison, it can be noticed that the cen- paper with results, the fitness function is composed by several
tralized self-healing methods aim a global optimization of the items, such as: unsupplied loads, overload factor, number of
network after their restoration; while the proposed distributed transmission lines with overload, and number of created loops
algorithm can perform a local optimization. Due to this fact, it (because, each time a loop is created, a switch must be open
seems that centralized systems present a better practical result. to eliminate it). Table II presents some solutions for different
However, this is not necessarily true, because they can introduce short-circuit locations using the proposed method, distributed
switching that the operator would not do in his/her daily oper- CBR, GA, PSO and HPSO. The solutions using centralized
ation. An operator will never command an action in a healthy CBR is not presented because this approach is not appropriated
part of the network only to provide a better loading division. for large networks due to the enormous number of possible ex-
This action will occur only if the affected part of the network amples. Even the approach proposed in [30] can’t handle with
needs this help. This fact can be exemplified with the use of the this extremely large situation.
results when a short-circuit occurs in the branch ALIM52. Observing Table II, four cases were selected. The two first
By using the proposed distributed algorithm, the substation cases are extremes, as with the opening of the switches in the
switch DJ5 opens to eliminate the short-circuit; then the switches initial branches 1–122 and 1–91, the energy in all circuits are
C15 and C25 open and lock to isolate the branch with the short- lost; while in the last two cases, partial blackouts are presented.
circuit. Finally, the proposed algorithm closes the switches DJ5 Note: C: x-y – O: i-j means: the switch between the buses x
and CR35, supplying energy to branches ALIM51 and ALIM53, and y must be closed; and the switch between buses i and j must
respectively. The same answer is obtained using the distributed be open.
and centralized CBR approaches. However, by the other cen- In the first case, the middle circuit lost all energy. At this
tralized methods (GA, PSO and HPSO), the switching solution moment, the proposed algorithm is executed by the follow-
is closing the switches DJ5, CR35 and CR25 and opening the ing switches 2–19 and 19–57, trying to supply energy by the
switch C23. right circuit; and the switches 20–56, 126–127, and 126–177,
The centralized meta-heuristic methods have the possibility supplying by the left circuit. Immediately, disputes are estab-
to run an optimal power-flow, and to use its solution to provide lished between 2–19 and 19–57 and between 126–127, and
a better load distribution for all network. This fact is important, 126–177, being the winners the switches in the branches 2–19
for instance, for preparing the network for a load increase or a and 126–177. The execution of the algorithm continues in the
new contingency. Despite this fact, in practice, operators will other switches of the middle circuit where the transferred data of
never adopt the solution provided for this last example. Other switch 126–127 is defeated, and the data transferred of switches
switching only would be carried out in practice if the switching 2–19 and 126–177 divide the middle circuit in the switch of the
of DJ5 and CR35 produced some kind of problem in the network, branch 117–121, which opens. In this case, part of the energy of
which is not the case. the middle circuit is supplied to the right circuit, and the other
part is supplied to the left circuit. This solution, also achieved by
B. Comparison Using Large Distribution Network
PSO and HPSO, presents the best loading for the two remaining
In this comparison, the modified large network and the results circuits.
presented in [28] are used. This modified large network, shown In the second case, the short-circuit occurs in the branch 1–
in Fig. 5, has some extra auxiliary switches from the original 91. The left circuit has some auxiliary switches located in the
network [31], increasing the complexity of the system opera- branches 2–19, 19–57 and 58–76, which are linked with the

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 2402 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

middle circuit only. It means, by using the proposed algorithm, [2] T. Sakagushi and K. Matsumoto, “Development of a knowledge based
only these switches are triggered to execute the restoration. system for power system restoration,” IEEE Trans. Power App. Syst.,
vol. PAS-102, no. 2, pp. 320–329, Feb. 1983.
Disputes occur in the switches internally of the right circuit [3] K. Moslehi and R. Kumar, “Vision for a self-healing power Grid,” ABB
with the transferred data from these switches and the winner is Rev., vol. 4, pp. 21–25, 2006.
the switch 2–19. This is not the best loading general solution, [4] R. E. Brown and T. M. Taylor, “Modeling the impact of substations
on distribution reliability,” IEEE Trans. Power Syst., vol. 14, no. 1,
which has the following switching: closing of the switches 2–19 pp. 349–354, Feb. 1999.
and 20–56, and opening of switch 20–48. In this solution, the [5] K. Moslehi and R. Kumar, “A reliability perspective of the smart grid,”
middle circuit assumes the load of right circuit, but part of the IEEE Trans. Smart Grid, vol. 1, no. 1, pp. 57–64, Jun. 2010.
[6] G. Lambert-Torres, G. M. Ribeiro, C. I. A. Costa, A. P. Alves da
middle circuit load is transferred to the left circuit. As mentioned Silva, and V. H. Quintana, “Knowledge engineering tool for training
previously, this type of solution can be more appropriated for the Power-Substation operators,” IEEE Trans. Power Syst., vol. 12, no. 2,
network loading, but it is never executed by operators without a pp. 694–699, Apr. 1997.
[7] V. C. do Nascimento, G. Lambert-Torres, C. I. de Almeida Costa, and L.
specific reason (such as: branch overloads or sag bus voltages). E. Borges da Silva, “Control model for distributed generation and network
In the third case, the short-circuit occurs in the branch 9–10, automation for microgrids operation,” Elect. Power Syst. Res., vol. 127,
and the following switches are triggered: 2–9, 19–57 and 58–76. no. 11, pp. 151–159, 2015.
[8] M. Amin, “Toward self-healing energy infrastructure systems,” IEEE
Several disputes occur with transferred data and the winner is the Comput. Appl. Power, vol. 14, no. 1, pp. 20–28, Jan. 2001.
switch 2–9. In this case, this solution is also the solution with [9] K. Moslehi et al., “Control approach for Self-Healing power Systems: A
the best global distribution loading. Finally, when the short- conceptual overview,” presented at the Electricity Transmission in Dereg-
ulated Markets: Challenges, Opportunities, and Necessary R&D, Carnegie
circuit occurs in branch 12–201, the following switches are Mellon University, Pittsburgh, PA, USA, Dec. 2004.
triggered: 2–9, 19–57 and 58–76. Several disputes occur with [10] V. V. Vadlamudi and R. Karki, “Reliability-based appraisal of Smart Grid
the transferred data and the data transferred of switches 2–9 challenges and realization,” in Proc. IEEE Power Energy Soc. Gen. Meet-
ing, San Diego, CA, USA, 2012, pp. 1–7.
and 19–57 divide part of the right circuit in the switch of the [11] A. Baggini, D. Chapman, and F. Buratti, “Reliability of electricity supply:
branch 60–62, which opens. This solution has also the best Structure,” in Handbook of Power Quality, A. Baggini, Ed. New York,
global distribution loading. NY, USA: Wiley, 2008, pp. 383–402.
[12] A. A. Chowdhury and D. Koval, Power Distribution System Reliability:
Practical Methods and Applications. New York, NY, USA: IEEE Press,
2011.
V. CONCLUSION [13] C. W. Taylor, D. C. Erickson, K. E. Martin, R. E. Wilson, and V. Venkata-
subramanian, “WACS-Wide-Area stability and voltage control System:
In this paper, self-healing distributed algorithms have been R&D and online demonstration,” Proc. IEEE, vol. 93, no. 5, pp. 892–906,
proposed. These algorithms are based on rules developed for May 2005.
[14] B. Chen, C. Chen, J. Wang, and K. L. Butler-Purry, “Sequential service
two sets of switches: substation and located in the network. restoration for unbalanced distribution systems and microgrids,” IEEE
The implementation in each switch is made without any kind of Trans. Power Syst., vol. 33, no. 2, pp. 1507–1520, Mar. 2018.
centralized control, allowing the switching decision to be taken [15] M. R. AlRashidi and M. E. El-Hawary, “A survey of particle swarm
optimization applications in electric power systems,” IEEE Trans. Evol.
by restricting communication only between adjacent switches. Comput., vol. 13, no. 4, pp. 913–918, Aug. 2009.
Even so, the operating conditions of the entire network required [16] V. Miranda and N. Fonseca, “EPSO - best-of-two-worlds meta-heuristic
for the solution of the problem are observed, taking into consid- applied to power system problems,” in Proc. Congr. Evol. Comput.,
Honolulu, HI, USA, 2002, pp. 1080–1085.
eration their pre-fault conditions. [17] W. C. Wu and M. S. Tsai, “Application of enhanced integer coded particle
The set of individualized switching creates in the end an swarm optimization for distribution system feeder reconfiguration,” IEEE
integrated self-healing plan that is built with few communica- Trans. Power Syst., vol. 26, no. 3, pp. 1591–1599, Aug. 2011.
[18] P. Mitra and G. K. Venayagamoorthy, “Implementation of an intelligent re-
tion features and intensive exchanges of data between adjacent configuration algorithm for an electric Ship’s power system,” IEEE Trans.
switches only. This self-healing plan is dynamically established Ind. Appl., vol. 47, no. 5, pp. 2292–2300, Sep./Oct. 2011.
and depends on the chosen optimization criteria. In the examples [19] T. Nagata and H. Sasaki, “An efficient algorithm for distribution network
restoration,” in Proc. PES Summer Meeting, 2001, pp. 54–59.
presented in this paper, the nominal capacity is available using [20] A. M. Tahboub, V. R. Pandi, and H. H. Zeineldin, “Distribution sys-
values of the pre-condition loads. In the proposed algorithms, tem reconfiguration for annual energy loss reduction considering variable
disputes establish winners among transferred data and conquest distributed generation profiles,” IEEE Trans. Power Del., vol. 30, no. 4,
pp. 1677–1685, Aug. 2015.
a new branch for its own circuit. [21] E. Drayer, N. Kechagia, J. Hegemann, M. Braun, M. Gabel, and R. Caire,
The proposed algorithm has been tested through computer “Distributed Self-Healing for distribution Grids with evolving search
simulations, in a real distribution system and in a large distribu- space,” IEEE Trans. Power Del., vol. 33, no. 4, pp. 1755–1764, Aug.
2018.
tion network, with its full restoration for various categories of [22] S. A. Arefifar, Y. Abdel-Rady, I. Mohamed, and T. H. M. EL-Fouly,
short circuits. Sequential and simultaneous short-circuits have “Comprehensive operational planning framework for Self-Healing control
been tested. Even in severe conditions of systemic shutdowns, actions in smart distribution Grids,” IEEE Trans. Power Syst., vol. 28,
no. 4, pp. 4192–4200, Nov. 2013.
the proposed self-healing algorithms create a restoration plan of [23] Z. Wang and J. Wang, “Self-Healing resilient distribution systems based
switching that allows the entire DS in fail were again energized. on sectionalization into microgrids,” IEEE Trans. Power Syst., vol. 30,
no. 6, pp. 3139–3149, Nov. 2015.
[24] Z. Wang, B. Chen, J. Wang, and C. Chen, “Networked microgrids for
Self-Healing power systems,” IEEE Trans. Smart Grid, vol. 7, no. 1,
REFERENCES pp. 310–319, Jan. 2016.
[25] A. Zidan and E. F. El-Saadany, “A cooperative multiagent framework
[1] M. Amin, “National infrastructures as complex interactive networks,” in
for Self-Healing mechanisms in distribution systems,” IEEE Trans. Smart
Automatic, Control and Complexity: An Integrated Approach, T. Samad
Grid, vol. 3, no. 3, pp. 1525–1539, Sep. 2012.
and J. Weyrauch, Eds. New York, NY, USA: Wiley, 2000, pp. 263–286.

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.
 TORRES et al.: DISTRIBUTED INTELLIGENT SYSTEM FOR SELF-HEALING IN SMART GRIDS 2403

[26] J. B. Leite and J. R. S. Mantovani, “Development of a Self-Healing strategy Lucas Roberto Ferreira received the B.S. degree in
with multiagent systems for distribution networks,” IEEE Trans. Smart automation and control engineering from the Catholic
Grid, vol. 8, no. 5, pp. 2198–2206, Sep. 2017. University of Parana, Curitiba, Brazil, in 2012, the
[27] L. C. Siebert, A. R. Aoki, A. R. R. de Souza, L. R. Ferreira, A. F. Bonelli, M.Sc. degree in electrical engineering from the
and F. de O Toledo, “Deterministic versus probabilistic approaches to Federal University of Paraná, Curitiba, Brazil, in
self-healing in smart Grid,” in Proc. 22nd Int. Conf. Exhib. Elect. Distrib., 2015. He is currently working toward the Ph.D. de-
Stockholm, Sweden, 2013, pp. 1–4. gree with the Department of Electrical Engineering,
[28] G. Lambert-Torres, H. G. Martins, M. P. Coutinho, C. P. Salomon, F. M. the Federal University of Parana. He is currently with
Matsunaga, and R. A. Carminati, “Comparison between PSO and GA in the Siemens, Curitiba, Brazil, as a Software Devel-
system restoration solution,” in Proc. 15th Int. Conf. Intell. Syst. Appl. opment Engineer. His research areas include the ma-
Power Syst., Curitiba, Brazil, 2009, pp. 1–6. chine learning applied in smart grids, self-healing,
[29] A. A. A. Esmin, G. Lambert-Torres, and A. C. Z. de Souza, “A hybrid load shedding, and power system operation.
particle swarm optimization applied to loss power minimization,” IEEE
Trans. Power Syst., vol. 20, no. 2, pp. 859–866, May 2005.
[30] G. Lambert-Torres et al., “New technology applied to a restoration of
electrical power substation,” in Proc. PES TD Asia Conf. Expo., 2009,
pp. 1–4.
[31] I. J. Ramirez-Rosado and J. L. Bernal-Agustin, “Genetic algorithms ap-
plied to the design of large power distribution systems,” IEEE Trans.
Power Syst., vol. 13, no. 2, pp. 696–703, May 1998. Alexandre Rasi Aoki (M’12–SM’17) received the
B.S., M.Sc., and Ph.D. degrees in electrical engineer-
ing from the Federal University of Itajubá, Itajuba,
Bruno Silva Torres received the B.S. degree in elec- Brazil, in 1996, 1999, and 2003, respectively. He was
trical engineering from the Federal University of a Senior Researcher and a Manager with the Insti-
Itajubá, Itajubá, Brazil, in 2009, and the M.Sc. degree tute of Technology for Development, Lactec Insti-
in development of technology from the Lactec Insti- tutes, Curitiba, Brazil, between 2003 and 2017. He
tutes, Curitiba, Brazil, in 2016. He was an Engineer was a Professor with the Development of Technology
with Eltman Engineering and Systems in 2010, with Master Degree Program, Lactec Institutes between
Schneider Electric Brazil between 2010 and 2015, 2004 and 2017 and a Coordinator between 2010 and
and also with Schneider Electric Argentina between 2014. Since 2008, he has been a Professor with the
2015 and 2018. He is currently a Researcher with Federal University of Paraná, Curitiba, Brazil. His research areas include plan-
Gnarus Institute, Itajuba, Brazil, developing R&D ning, operation studies for distribution systems, smart grid including renewable
projects for Brazilian power industries. His research energy sources, and storage systems. He serves on a number of committees
areas include the analysis and operation of complex power systems, adaptive related to distributed generation and smart grids including the IEEE and the
protection, and smart grids. CIGRÉ.

Authorized licensed use limited to: UNIV ESTADUAL PAULISTA JULIO DE MESQUITA FILHO. Downloaded on September 27,2022 at 17:25:02 UTC from IEEE Xplore. Restrictions apply.