Electrical Power and Energy Systems 63 (2014) 787–805

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Fast fault section estimation in distribution control centers using

adaptive genetic algorithm
Fábio Bertequini Leão ⇑, Rodrigo A.F. Pereira, José R.S. Mantovani
Research Group in Electric Power System Planning, Department of Electrical Engineering, São Paulo State University – UNESP, Ilha Solteira 15385-000, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a novel mathematical model for fast fault section estimation in a Distribution Control
Received 4 July 2013 Center (DCC). The mathematical model is divided into two parts, namely: (1) a protection system oper-
Received in revised form 14 June 2014 ations model based on operator’s heuristic knowledge of the protection system performance and (2) an
Accepted 16 June 2014
optimization Unconstrained Binary Programming (UBP) model based on parsimonious covering theory.
Available online 17 July 2014
In order to solve the UBP model, an Adaptive Genetic Algorithm (AGA) using crossing over and mutation
rates that are automatically tuned in each generation is proposed. An Alarm Probabilistic Generator
Keywords:
Algorithm (APGA) is developed and a real four-interconnected distribution substation system is used
Distribution control centers
Fault section estimation
to test exhaustively the approach. Results show that the proposed methodology is capable of performing
Fault diagnosis fault section estimation in a very fast and reliable manner. Furthermore, the proposed methodology is a
Protective relaying powerful real-time fault diagnosis tool for application in future Distribution Control Centers.
Digital protection Ó 2014 Elsevier Ltd. All rights reserved.
Genetic algorithm

Introduction
The task of fault detection, isolation and network restoration
(FDIR) may be very complicated in a DCC environment mainly in
emergency conditions, where a large simultaneous flow of infor-
mation from several monitored substations may overwhelm the
operator. Also, the FDIR task becomes increasingly complex either
in multiple fault scenarios or in situations presenting protective
devices malfunction. Therefore, advanced Distribution Manage-
ment Systems (DMS) must have computational tools to provide
fast and accurate automated fault analysis in a DCC environment
to enhance the operator decision making process [1].
The first step in FDIR task is called fault section estimation or
fault diagnosis [2] and operators are required to use their
knowledge and expertise on the operation of protection system
in order to perform this task.
In order to aid substation operators, and due to the combinato-
rial complexity of the fault section estimation problem, a number
of computational intelligence-based methods and solutions
techniques have been developed in the past years. Expert Systems
(ES) [3–6] and Artificial Neural Networks (ANN) [7–10] are among
the most traditionally used methodologies. ES use rule-based
knowledge models to represent the operator expertise in dealing
with protective systems and an inference engine to estimate the
⇑ Corresponding author. Tel.: +55 (18) 3743 1150; fax: +55 (18) 3743 1163.
E-mail addresses: bertequini@dee.feis.unesp.br (F.B. Leão), ddigo@yahoo.com
(R.A.F. Pereira), mant@dee.feis.unesp.br (J.R.S. Mantovani).
sections under fault. Although ES are widely used, the acquisition,
validation and maintenance of knowledge bases from large power
systems impose serious difficulties and otherwise the inability of
generalization limits the ES to producing results based solely on
its knowledge base. Although an ANN overcomes the generaliza-
tion incapability found in ES, ANN requires a training process car-
ried out in a pre-processing manner for weighting and preparing
the neural network to perform the fault diagnosis. Although
advances have been made, some difficulties related to fault section
estimation problem for large electric power systems still persist
such as slow convergence in the training process, size of training
sample to ensure reliable results, determination of the network
parameters as hidden units, and the quantity of layers.
Fuzzy Logic (FL) is robust in the face of dealing with uncertain-
ties of protection systems and the majority of works use FL
combined with other techniques such as ES [11,12], ANN [13],
cause-effect network (CE-Net) [14–16], Petri Nets (PN) [17] and
Fuzzy Reasoning Spiking Neural P Systems (FRSN P systems) [18].
Ref. [14] presents the combined use of the cause-effect network
(CE-Net) and the FL in which the CE-Net is used to model the
operation logic of the protection system while FL performs the pro-
cess of classifying faults using currents and voltages. In same way,
[15,16] use FL to model the uncertainty relationship among
protective devices operated and fault sections. However, the major
drawback of FL is the modeling of membership functions generally
based on historical data, experience or trial-and-error.
Similar to CE-Nets, the PN and FRSN P System model the protec-
tion system operation through graphic representation [17–20].

http://dx.doi.org/10.1016/j.ijepes.2014.06.052
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
788 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805
Although CE-Nets, PN and FRSN P Systems present a clarifier idea
about the cause-effect relationship of protective devices, the
graphical mapping of protection system operation is a drawback
when applied to a large electric power system employing a
complex protection system.
Immune Algorithms (IA) [21,22], Genetic Algorithms (GA)
[22–27], and more recently Artificial Bee Colony (ABC) algorithm
[28], Biogeography-Based Optimization algorithm (BBO) [29] and
Enhanced Honey-Bee Mating Optimization (EHBMO) algorithm
[30] have proved to be suitable approaches for solving such
problem. Although IA and GA techniques present robustness,
easiness for coding and reduced processing time, however, their
major drawback is their high number of control parameters
necessary to be tuned. Because these parameters must be tuned
in an exhaustive trial-and-error basis, an expressive number of
parameters make this process laborious.
In this paper a novel mathematical model suitable for fast fault
section estimation in DCC is proposed. The model is intended for
advanced DMS (Distribution Management System) [1] and takes
into account a meshed subtransmission system and the several
protections and control layers of distribution substations.
The proposed model is divided into two parts: (1) a protection
system operations model, and (2) an Unconstrained Binary Pro-
gramming (UBP) model. The protection system operation model
is composed of a set of expected state equations of the protective
relay functions based on the operation logic of protection functions
such as overcurrent, differential, distance along with in the protec-
tion philosophy employed for the protection system. On the other
hand, the UBP model, which is an extension of [22], is established
through an objective function based on the parsimonious covering
theory [23,31] for associating alarms of protective relay functions
informed by SCADA system with the expected states of the protec-
tive relay functions formulated in part (1). The model of the protec-
tion system operations is based on the operator’s heuristic
knowledge of the performance of protection system, and the opti-
mization UBP model is used to formulate the solution hypothesis
based on abductive inference logic. Differently from the model
proposed in [22] the model herein developed is another model
much more complex and takes into account the number of
protection and control layers presented in modern distribution
substations. The AGA carries out the process of fault diagnosis by
minimizing the UBP model in each generation, having a dedicated
stopping criterion based on fault classification, only two control
parameters to tune and crossing over and mutation rates are
automatically tuned in each generation based on population
saturation. Comparing with AGA, solution techniques based on
optimization such as IA [22] and GA [22,25,27] presents a much
more number of parameters to be tuned.
In order to test and validate the overall methodology, a large
number of alarms from a real four-interconnected distribution
substation system are automatically generated using the proposed
Alarm Probabilistic Generator Algorithm (APGA). The aim of using
AGA is to verify and validate the mathematical model developed
and the efficiency of algorithm is compared with a dedicated GA
[22]. Obtained results confirm the efficiency and accuracy of the
proposed methodology in enhancing operator decision making
process in the fault diagnosis task in real time.
Mathematical model for the fault section estimation problem
The protection system operation model is a set of expected state
equations of the protective relay functions in each equation of this
set mathematically models the operation logic of the protective
functions located at the distribution substations. The set of
expected state equations are obtained in an off-line manner and
developed taking into account: (1) heuristic knowledge which
intuitively describe the operation logic of protective functions;
(2) protection philosophy used by experts in protection specifica-
tion, selectivity and coordination [32]; and (3) data from electric
power system.
Fig. 1(a) presents a diagram of a typical distribution substation
used to clarify the mathematical model development. The protec-
tion system is compound of multifunctional digital relays and the
function numbering is based on the ANSI/IEEE C37.2 standard
[33]. Fig. 1(b) illustrates the schematic diagram to model multi-
functional digital relay.
Heuristic knowledge of protection logic operation
In order to obtain the expected states of the protective functions
for a distribution substation is necessary to state in a comprehen-
sive mathematical manner the heuristic knowledge describing the
operation logic of every protective function when a fault occurs in
different parts of electric power system [32]. We translate this
knowledge for every protective device as follows.
Digital Relay: Based on the schematic diagram from Fig. 1(b), it
is considered that any protective function of any relay r does not
operate if there is any problem with DC/AC power supply, i.e., a
DC protection alarm appears (pvccr A = 1), or an AC protection
alarm appears (pvcar A = 1), or an auto-diagnosis signal appears
(audr A = 1). This rule defines the behavior of digital relay in the
protection system. If the digital relay presents any problem, it is
expected that the protective functions of the relay do not operate.
Circuit Breaker: A circuit breaker is considered defective only if
after a tripping signal, it does not open. In Fig. 2(a) is depicted the
relationship between protective functions and circuit breakers.
Breaker Failure Protection: 50BF will only operate if the
following events occur simultaneously: (1) there is a fault in a
given section that is protected by a relay whose functions include
function 50BF; (2) any of those functions operates due to the fault;
and (3) the tripped circuit breaker is defective. In Fig. 2(b) is
depicted the operation logic of 50BF protective function.
Auto reclosing: Function 79 will only operate if the following
events occur simultaneously (Fig. 2(c)): (1) there is a fault in a
given section that is protected by a relay whose functions include
function 79; (2) any of those functions operates due to the fault;
and (3) either function 50BF of the relay does not operate or func-
tion 50BF of any other relay that trips the same circuit breaker
tripped by function 79 does not operate.
Overcurrent and distance protection: Overcurrent functions
50/51, (50/51)N, 67/67N, 46, (50/51) NS (sensitive ground-fault
detection), function 61 and distance functions 21/21N are formu-
lated considering the following: (A) The protective function will
only operates if there are faults in its protection zone provided
the functions have no Logic Selectivity (LS) input. Even with a fault,
the function will not operate if it is blocked by LS. This logic input/
output is considered in order to coordinate relays too near or, in
other words, where a short circuit level is very similar. For
example, in Fig. 1(a), functions 50/50N of r07 and r09 are blocked
whenever the functions 50/50N of r10, r11, r12 and r13 operate.
This blocking ensures that relays r07 and r09 do not operate and
de-energize busbars M-MV and T-MV incorrectly even if a short
circuit occurs in feeders or capacitor banks; (B) The protective
function will operate due to short circuit current if at least one cir-
cuit path between a reference node and protected section has all
switchgears and circuit breakers closed (except for circuit breakers
tripped by the protective function considered). A reference node is
considered as: (1) an installation point of the Current Transformer
(CT) that supplies current to the relay with the protective function
considered or (2) a reference busbar. A reference node takes into
account the radial feature of distribution substations as well as
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 789

Fig. 1. (a) Typical distribution substation with protection and control apparatus and (b) schematic diagram to model digital relay.

the protective function under consideration and the protected sec-
tion. For example, in Fig. 1(a), node 02 is the reference node for the
protective functions 50/51, (50/51)N and 67/67N of r01 consider-
ing section IL01 protected. In this case there are two circuit paths
between nodes 02 (reference) and 01 (section) considering: (1)
the trip transfer function 43–1 not transferred, i.e., switchgears
c02 and c01, and (2) function 43–1 transferred, i.e., switchgears
c03, c04 and c23. Also, node 01 is the reference node for functions
50/51, (50/51)N for same relay considering busbar M-HV pro-
tected. In this case, there are two circuit paths between nodes 01
(reference) and 02 (section); (C) Protective functions 21/21N are
MHO directional distance functions and have three zones: zone
‘‘p’’ (main protection) that reaches 80% of the transmission line
length; zone ‘‘s’’ (second zone or primary backup protection) that
reaches 150% of the line length and zone ‘‘t’’ (third zone or second-
ary backup protection) that reaches 100% of the protected line
length plus 100% of the longest adjacent protected line. Based on
expert knowledge [32] can be assumed that when a fault occurs
in the first zone all protection zones are sensitized. In this situation
distance protection of zone ‘‘p’’ is the first to operate without delay.
If protection of zone ‘‘p’’ fails, the second zone should operate with
an intentional delay. Finally, if the first and second zones fail, third
zone should operate. The same reasoning can be used to faults in
zone ‘‘s’’. In this case zone ‘‘p’’ is not sensitized. Fig. 3 depicts the
790 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

(a)

(b)

(c)
Fig. 2. Operation logic diagrams. (a) Circuit breakers; (b) protective function 50BF; and (c) protective function 79.

operation logic of distance functions considering faults in zones
‘‘p’’ and ‘‘s’’, respectively. Fault in zone ‘‘t’’ is sensed only by 21/
21N-t. In Fig. 3, circuit breaker (1) is the circuit breaker that the
functions 21/21N trip. Circuit breaker(s) (2) is (are) the circuit
breaker(s) that function 50BF trips and pertaining to the same
relay that the functions 21/21N.
Under/overvoltage protection: functions 27/59 respectively
operate only if there is an under/over voltage in the protected sec-
tion. Thus, it must be considered: (1) if the Potential Transformer
(PT), which supplies the relay, receives a voltage signal from a
protected section and (2) if the relay is connected in the system.
The protective function will operate due to under/over voltage if
at least one circuit path between a reference node and the
protected section has all switchgears closed. For these functions,
a reference node is considered as the CT that supplies the relay.
For instance, node 13 in Fig. 1(a) is a reference node for relay r10
considering sections M-MV and T-MV. Considering section M-MV,
the circuit path is switchgear c17. When the section T-MV is taken
into account, the circuit path is switchgear c17, c31, c13 and c14.
Transformer protection: Protective functions 51G, 87, 63, 63A,
80, 26, 49 and 71 operate only if there is a fault in the protected
transformers. No circuit paths are considered when modeling these
functions since they directly monitor the transformers.
Equations of the expected states of protective relay functions
The equations of protective relay functions are modeled
based on the stated knowledge aforementioned as well as the

(a)

(b)
Fig. 3. Operation logic diagrams – protective functions 21/21N. (a) fault in zone ‘‘p’’ and (b) fault in zone ‘‘s’’.
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 791

relationships between control and protective devices are taking formulate the expected state equations for relays whose CT are
into account in a comprehensive mathematical manner. Consider installed at bays IL, OL or HV-TF, considering in this case that all
equation of 50BF function given by Eq. (1): sections are protected. Based on Fig. 1(a), the expected state
 X  equations for the protective functions of the relay 01 are
rE rA rE
f50BF ¼ MAX fp2XRr fp  ORC j2J rfp
fpj formulated using equations such as type 1, i.e., CT01 supplying
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
1 relay 01 is installed in IL bay.
 Y  X  X 
 1
A
ð1  fti Þ  ORC
A A A The equations for type 1 overcurrent functions are given by:
i;j d þ fti  ORC i;j d  8 2 0 0 0 11 0 0 1113
i2Xffpr k2XdN k k2XdT k
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} >
<

rEð1Þ 6XB A B B X A CC A B B X A CCC7

2 ffp ¼ MAX sj  4 @ 1  ft i  @ 1  ORC @ c k A A þ ft i  @1  ORC @ c k AAA5
h
i r >
j2Jfp : r
i2Xffp i;j
k2XcN
i;j
k2XcT
rA
 1  ORC pv ccr A þ pv car A þ aud ð1Þ 2 0 0 0 11 0 0 1113
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
3 6 B B B CC B B CC C7
6 B B B X C
CC B B X CC C7
6 Y B
B B B CC7
6 B A B A CC A B B A CCC7
r E 6 B 1  fti  B1  ORC BB cAk1 þ dk2 C
CC þ fti  B1  ORC B
B cAk1 þ dk2 C
CC C7
where f50BF is the expected state of protective function 50BF of the 6 B
6i2Xcafpj;r B
B
B B
@ k1 2 Xci;j
C
CC
AA
B
B B
@ k1 2 Xci;j
CCC
C
AAC
7
7
4 @ @ N @ T A5
relay r; MAX{} is the maximum value of the elements considered; k2 2 XdN
i;j
k2 2 XdT
i;j

h
i h
i
ORC (x) = 1 if x P 1, otherwise ORC (x) = 0; XRr is set of protective  1  f67 rA
LS
g  1  ORC pv ccr A þ pv car A þ aud
rA

functions of the relay r; fpr A represents the alarm of the protective ð3Þ
function fp of the relay r; (1 for operated, 0 for non-operated); J rfp
is the index set of sections protected by protective function fp of where ffprE is expected state of overcurrent function; sj is the estate of
rE
the relay r; fpj is term related to section j from the equation of section j (1-fault, 0-normal); cAk is the alarm of switchgear k (1 for
expected estate of the protective function fp of the relay r; Xffpr ‘‘open’’, 0 for ‘‘closed’’); Xci;j i;j
N ; Xc T are the index set of switchgear(s)
is index set of trip transfer functions that transfer trip from fp of connected to the circuit path from a reference node to a protected
the relay r, where if Xffpr ¼ ; then the product P is equal to the cir- section j considering the trip transfer function i did not transfer
A
cuit breaker tripped by fp; ft i is the alarm of trip transfer function (N) and transferred (T) respectively; Xcaj;r fp is the index set of trip
A transfer functions that transfer trip to circuit breakers connected
i (1 for ‘‘transferred’’ (T) and 0 for ‘‘not transferred’’ (N)); dk is
alarm of circuit breaker k (1 for ‘‘open’’, 0 for ‘‘closed’’); and to every circuit path between a reference node (considering the
Xdi;j i;j
N ; XdT are index sets of circuit breaker(s) connected to the cir-
protective function fp) and the protected section j, except the index
cuit path from a reference node toward the protected section j, of the set Xffpr ; and ffpr LSA is the alarm of protective function fp of the
considering the trip transfer function i is not transferred (N) and relay r that has LS with the modeled function. (ffpr LSA ¼ 0 if the relay
transferred (T), respectively. does not have logic selectivity with any other relay).
Eq. (1) can be separated into three major products as seen in the In Eq. (3), the first and the second terms between square
equation. The first one considers the dependence on protective brackets multiplied by sj represent mathematically the circuit path
function alarm given by events (1) and (2) from heuristic knowl- between a reference node and the protected section sj. The third
edge of breaker failure protection (Section ‘Heuristic knowledge term represents the LS interlock. In addition the number of terms
of protection logic operation’). The second product term takes into between curly brackets depends on the number of sections
account event (3) from breaker failure protection since operation protected by protection function formulated. Equations of type 2
of circuit breaker depends on trip transfer function state. overcurrent functions are given by:
8 2 0 0 0 11
Moreover, third term considers the tripping function 50BF >
< Y

dependence on normal operation of digital relay. r Eð2Þ 6 B A B B X A CC A

ffp ¼ MAX s j  4 @ 1  ft i  @1  ORC @ c k AA þ ft i
j2J rfp >
: k0 ;r
Based on knowledge of auto reclosing protection, the equation j;r
i2Xcafp Xfdfp k2Xci;j
N
0 0 1113 2 0
for the protective function 79 is written as:

B B X A CCC7 6 B X A
82 0 139 @1  ORC @ ck AAA5  4ORC @ 1  fti
< X E =   k2Xci;j
T
0
k ;r
i2Xfdfp
r A
r E
f79 ¼ MAXr 4fp  ORC @ fpj A5  1  f50BF
r A 0 0 11113 9
>
fp2XR : r ; B C C C7
>
>
>
j2J fp B B CCCC7 >
>
B B CCCC7 h
i >
=
h
i B B X A CCCC7
r A B1  ORC B ck1 þ dk2 C
A r A
CCC7  1  f50LS
 1  ORC pv ccr A
þ pv car A
þ aud ð2Þ B
B
B
B
C
C C C C7 >
>
>
@ @ k1 2 XcNi;j AC CC7
AAA5 >
>
>
>
i;j 0 ;
k2 2 XdN  fk g
Analogous reasoning from Eq. (1) can be used to interpret Eq. h
i
r A
(2) but in this case function 79 depends on breaker failure  1  ORC pv cc þ pv ca þ aud
r A r A
ð4Þ
protection alarm. k0 ;r
where X fdfp
is the index set of trip transfer functions that transfer
Due to topological features of the distribution substation we
trip to the circuit breaker k0 tripped by the protective function fp of
proposed that overcurrent functions 50/51, (50/51)N, 67/67N, 46,
the relay r.
(50/51) NS are classified considering the substation bays as shown
In Eq. (4) the first term in square brackets represents the circuit
in Table 1. For instance, equations classified as type 1 are used to
path between a reference node and the function formulated. For
example, in order to formulate function 50 of r07 in Fig. 1(a) con-
Table 1 sidering TF01 (s5) protected, the reference node is node 02. Because
Classification of generic equations for overcurrent functions considering installation if a short circuit has occurred in TF01 the short circuit current flow
bay of the relay and protected sections. by only busbar M-HV or both M-HV and T-HV (depends on 43–6
Type Installation bay of the relay CT Protected sections state) plus TF02 and only busbar M-MV or both M-MV and T-MV
1 IL, OL, HV-TFa IL, OL, TF, Busbars, CB, FD (depends on 43–5 state). The second term between square brackets
2 MV-TFa TF, Busbars, CB, FD is added to continue the circuit path from first term (node 10) to
3 Transfera TF, Busbars, CB, FD protected section. Notice that, r07 trips directly circuit breaker
4 CB, FD CB, FD d06 and this relay will be in operation only if 43–4 state is not
a
HV-TF is related to the relay CT installed on the HV side of transformers, e.g., transferred.
CT03-r03 and CT05-r04 in Fig. 1(a). MV-TF is related to the relay CT installed on MV In order to model the equations of type 3, we consider two
side of the transformers, e.g., CT08-r07 and CT10-r09. Transfer is related to the relay groups of protective functions: group (1) consists of protective
installed in the transfer bay, e.g., relay r08.
functions for substituting the relays installed on the MV side of
792 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

transformers, indicated by (⁄) (see Fig. 1(a)), and group (2) for Distance functions, 21/21N are considered as having three
substituting feeder relays. Thus the equations of type 3 considering zones. The equations for zone ‘‘p’’ are given by:
group (⁄) are given by: 8 2 0 0 0 11
8 2 0 0 0 11 >
<

> rE 6 XB A B B X A CC
<
f21p ¼ MAX sj  4 @ 1  ft i  @1  ORC @ c k AA
rEð3Þð1Þ 6 Y B A B B X CC j2Jr21p >
: r
cAk AA i2Xf21p i;j
ffp ¼ MAX sj  4 @ 1  fti  @1  ORC @ k2XcN
j2J rfp >
: 0 0 11139
j;r k0 ;r k2 X c i;j
>
i2Xca Xfd =h
i
B X A CCC7
fp fp N
A B rA
0 0 1113 þ ft i  @1  ORC @ ck AAA5 1  ORC pv ccrA þ pv carA þ aud
i;j
>
;
B X A CCC7
k2Xc
A B T
þfti  @1  ORC @ ck AAA5 ð8Þ
k2Xci;j
T
2 0 0 0 0 11113 Since zone ‘‘p’’ is sensitized by faults within this zone, Eq. (8)
takes into account only sections protected by zone ‘‘p’’. While for
6 B X B B B X CCCC7
6 B B A B B A C CCC7 zone ‘‘s’’, the equations are given by:
6
6 ORC B
B
Bft  B1  ORC B
B i B B c A
k1 þ dk2
CCCC7
CCCC7 n o
4 @i2Xfdk0 ;r @ @ @ k12Xci;j AAAA5 rE
f21s
rEð1Þ rEð2Þ rEð3Þ
¼ MAX f21s ; f21s ; f21s
fp N
i;j
k22Xd fk0 g
N h
i
r A
h
io h
r A

i  1  ORC pv ccr A þ pv car A
þ aud ð9Þ
rA
 1  f50 LS
 1  ORC pv ccr A þ pv car A þ aud
The three terms between curly brackets of (9) are established as
ð5Þ follows:
Eq. (5) is very similar to Eq. (4) except the second term between n h io
square brackets. This is because when this type of relay is operat- rEð1Þ
f21s ¼ MAX r A
sj  1  f21p ð10Þ
r
ing, the trip transfer function of MV-TF bay must be transferred. j2J 21p

For example, when r8⁄ substitute r07 in Fig. 1(a), the state of 43– 8 2 0 0 0 11
4 must be transferred and c11 and c12 are open. While group (2) >
>
>
> 6 X B
of type 3 take the form: < 6 B
B
B
BX
B
CC
rEð2Þ 6 B 1ftA  B1ORC B A A C C
CC
8 2 3 f21s ¼ MAX s j  6 B i B B c k1 þdk2 CC
j2J r21p >
> 4i2Xf21s
r @ @ @k12Xci;j AA
>
< >
>

6 Y
: N
rEð3Þð2Þ A 7 i;j
k22Xd
ffp ¼ MAX sj  4 1  fti 5 0 0 1113
N

j2J rfp >

: 0
i2Xfdkfp ;r
2 0 0 0 11139 B BX CCC7
A B B A CC C7
>
> þft i  B B
B1ORC B cAk1 þdk2 C CC7
CCC7
6 B X B B X CCC7> >
= @ @k12Xci;j AAA5
6 B B B A C CC 7 T
6
6 ORC B
B
B1  ORC B
B B c A
k1 þ d CCC7
k2 CCC7
k22Xd
i;j
T
4 @i2Xfdk0 ;r @ @ k12Xci;j AAA5> >
> 2 0 0 0 0 11
fp N >
;
i;j
k22Xd fk0 g 6 B X B
B BX CC
h
N

i 6 B B B B A CC
6 B B 1ft A  B1ORC B cAk1 þdk2 C C
r A 6ORC B B i B B CC
 1  ORC pv ccr A þ pv car A þ aud ð6Þ 4 @i2Xf50BF
r @ @ @k12Xci;j AA
N
i;j
k22Xd
N
Eq. (6) states overcurrent protection functions when the relay 0 0 13 9
111
>
>
from transfer bay, i.e., r08 in Fig. 1(a), is substituting relays from B BX CCC C7 >
>
feeder bay, i.e. r12 and r13. In this situation, both relays r07 and A B B A CCC  C7=
þft i  B
B1ORC B
B c A
k1 þd CCC þ 1f
k2 CCC
r A C7
50BF C7 ð11Þ
r09 from MV-TF bay must be operating, i.e., 43–4 and 43–5 are @ @k12Xci;j AAA A5 >
>
>
T >
;
not transferred in Fig. 1(a). This assumption is guaranteed by the i;j
k22Xd
T

substation automation system where it permits that r08 can

8 2 0 0 0 11
substitute r12 or r13. In other words, r08 can substitute only >
< X
X
relay from MV-TF bay or feeder bay, not simultaneously. This rEð3Þ 6 B A B B CC
f21s ¼ MAX sj  4 @ 1  fti  @1  ORC @ cAk AA
assumption is modeled by first term between square brackets in j2J r21s Jr21p >
: r i;j
i2Xf21s k2Xc N
Eq. (6). Equations of type 4 are written as: 0 0 1113
8 2 0 139
>
< X >
= A B B X A CCC7
rEð4Þ 6 B A C7
þfti  @1  ORC @ ck AAA5
ffp ¼ MAX r >
s j  41  ORC @ c k A 5 k2Xci;j
j2J fp : j;r
>
; T
k2Xfcfp 2 0 0 0 11
h
i
r A
 1  ORC pv ccr A þ pv car A þ aud ð7Þ 6 Y
6
B
B
B
B
BX
B
CC
B 1  ftA  B1  ORC B A C C
6
6 B i B B c A
k1 þ d CC
k2 CC
j;r
where Xfcfp is the index set of switchgear (s) connected to the cir- 4i2Xcaj;r @ @ @k12Xci;j AA
21s N
cuit path containing the circuit breaker tripped by the protective k22Xd
i;j
N
function fp of the relay r, between a reference node and protected 0 0 11139
>
>
section j. B BX CCC7> >
B B C C C7 =
Eq. (7) is simpler, because the circuit path from capacitor bank A A
þfti  B
B1  ORC B
B c A
k1 þ d CCC7
k2 CCC7 ð12Þ
and feeder bays depends on switchgears states. For example, if @ @k12Xci;j AAA5> >
>
switchgears c18 and/or c19 are open in Fig. 1(a), is expected that T >
;
i;j
k22Xd
T
overcurrent functions from relay r12 does not trip circuit
breaker d11.
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 793

rEð3Þ  r A

Eqs. (10) and (11) represent the logical operation of zone ‘‘s’’ f21t ¼ MAX sj  1  f21s ð16Þ
r r
j2J 21s J 21p
when a fault occurs in zone ‘‘p’’ (see Fig. 3(a)) while Eq. (12)
corresponds to a fault in zone ‘‘s’’ (see Fig. 3(b)). In Eq. (11) 8 2 0 0 0 11
the first term between square brackets represents the circuit >
>
>
> 6 X B
< 6 B
B
B
BX
B
CC
path from a reference node and the protected section. Note that rEð4Þ B 1  ft A  B1  ORC B A CC
f21t ¼ MAX r >
sj  6
6 B i B B cAk1 þ dk2 C
CC
C
the circuit breaker alarm is considered in this term because zone r
j2J 21s J 21p >
> 4i2Xf21t
r @ @ @k12Xci;j AA
>
: N
‘‘s’’ depends on operation of zone ‘‘p’’ and circuit breaker is i;j
k22Xd
N
tripped by this zone (circuit breaker (1) in Fig. 3(a)). The second 0 0 1113
term in Eq. (11) takes into account the operation of 50BF func-
B BX CCC7
tion as well as the circuit breaker tripped by this function (cir- A B B A CC C7
þfti  B B
B1  ORC B cAk1 þ dk2 CCC7
CCC7
cuit breaker (2) in Fig. 3(a)). While Eq. (12) takes into account @ @k12Xci;j AAA5
T
the circuit path between a reference node and protected sections k22Xd
i;j
T
only by zone ‘‘s’’. Furthermore, the equations for zone ‘‘t’’ are 2 0 0 0 0 11
written as:
6 B X B
B BX CC
6 B B B B A CC
n o 6 B B 1  ft A  B1  ORC B cAk1 þ dk2 C C
rE rEð1Þ rEð2Þ rEð3Þ rEð4Þ rEð5Þ 6ORC B B i B B CC
f21t ¼ MAX f21t ; f21t ; f21t ; f21t ; f21t 4 @i2Xf50BF
r @ @ @k12Xci;j AA
N
h
i k22Xd
i;j
N
r A
 1  ORC pv ccr A þ pv car A þ aud ð13Þ 0 0 111 13
B BX CCC C7
where the terms in (13) are given by: A B B A C CC   C7
þfti  B
B 1  ORC B
B c A
k1 þ dk2
CCC þ 1  f r A C7
CCC 50BF C7
n h i @ @k12Xci;j AAA A5
rEð1Þ r A r A
o T
f21t ¼ MAX
r
sj  1  f21p  1  f21s ð14Þ k22Xd
i;j
T
j2J 21p
2 0 0 0 11
8 2 0 0 0 11 6 Y B
6 B
B
B
BX
B
CC
>
> B 1  ft A  B1  ORC B A C C
>
>
< 6 X B
B BX CC 6
6 B i B B c A
k1 þ d CC
k2 CC
6 B B B A C C 4i2Xcaj;r @ @ @k12Xci;j AA
rEð2Þ
f21t ¼ MAX s  6 B 1  ft A  B1  ORC B c A
þ d CCþ
j2Jr21p >
j 6 B i B B k1 k2 CC 21t N
>
> 4i2Xf21t @
r @ @k12Xci;j AA k22Xd
i;j
>
: N
N
i;j
k22Xd
N
0 0 11139
0 0 1113 >
>
B BX CCC7>>
A B B A CCC7=
B
B
BX
B
CCC7 þfti  B B
B1  ORC B ck1 þ dk2 C
A CC7
CCC7> ð17Þ
A A CC C7
@ @k12Xci;j AAA5>
ft i  B B
B1  ORC B cAk1 þ dk2 C CC7
CCC7 >
>
@ @k12Xci;j AAA5
T
i;j
;
k22Xd
T T
i;j
k22Xd
T
2 0 0 0 0 11 8 2 0 0 0 11
>
<

6 B X B
B BX CC rEð5Þ 6XB A B B X A CC
6 B B B B A C C f21t ¼ MAX sj  4 @ 1  fti  @1  ORC @ c k AA
6 ORC B B 1  ft A  B1  ORC B c A
þ d CC j2J r21t J r21s >
6 B B i B B k1 k2 CCþ : r
i2Xf21t k2Xci;j
4 @i2Xf50BF
r @ @ @k12Xci;j AA N
N
k22Xd
i;j
N
0 0 1113
0 0 111 13 9
> A B B X A CCC7
> þfti  @1  ORC @ ck AAA5
B BX CCC C7 >>
=
A B B C
A CCCC C  C 7 k2XcT
i;j
B B
ft i  B1  ORC B A r A C7
ck1 þ dk2 CCC þ 1  f50BF C7
@ @k12Xci;j AAA A5 >>
> 2 0 0 0 11
T >
;
i;j
k22Xd
T
6 Y B
B BX CC
6 B B B A C C
ð15Þ 6 B 1  ftA  B1  ORC B c A
þ d CC
6 B i B B k1 k2 CC
4i2Xcaj;r @ @ @k12Xci;j AA
21t N
i;j
k22Xd
N

0 0 11139
>
>
B BX CCC7> >
=
A B B C C
A CCC7 C7
Table 2 B B
þfti  B1  ORC B A
ck1 þ dk2 CCC7 ð18Þ
Rule for obtaining Coi and Ini. @ @k12Xci;j AAA5> >
>
T >
;
i;j
ffpA ffpE Coi Ini k22Xd
T
i i

0 0 0 0 Eqs. (14) and (15) represent the logical operation of zone

0 1 0 1
‘‘t’’ when a fault occurs in zone ‘‘p’’ (see Fig. 3(a)). The reason-
1 0 1 1
1 1 0 0 ing here is the same from Eqs. (10) and (11) but in this case
zone ‘‘t’’ depends on alarm from zone ‘‘p’’ and zone ‘‘s’’. While
Eqs. (16) and (17) take into account the logic of zone ‘‘t’’ when
a fault occurs in zone ‘‘s’’ (Fig. 3(b)). Note that in these equa-
Lines Substation 1 Substation 2 Substation n tions only the sections protected by zone ‘‘s’’ are considered.
Moreover, Eq. (18) takes into account faults occurring only in
0 1 1 ... ... ... ... ... ... 1 0 1 zone ‘‘t’’.
s1 s2 s3 ... ... ... ... ... ... sns-2 sns-1 sns
The equations of protective functions 27 and 59 are classified in
Fig. 4. Codification proposed for centralized fault diagnosis. A chromosome with ns two types. The first type is considering relay CT installed in IL or OL
genes. bays and they are given as:
794 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

8 2 0 0 0 11
>
< X
X
r07 and r09 are primary protection of busbars M-MV, T-MV and
rEð1Þ 6 B A B B CC TF01, TF02 due to the possibility of parallel operation of transform-
f27 ¼ MAX sj  4 @ 1  fti  @1  ORC @ cAk AA
j2J r27 >
: r i;j ers (switchgears c31 and c32 closed). These functions have LS input
i2Xf27 k2XcN
0 0 11139 activated in order to be coordinated with the capacitor and feeder
>
= relays. Also, functions 51/51N are backup protection of the same
A B B X A CCC7
þfti  @1  ORC @ ck AAA5 sections plus CB01, CB02, FD01 and FD02. Relay r08, on the other
i;j
>
;
k2XcT hand, has two groups of protective functions. The group (⁄) is acti-
h
i vated when r08 substitutes relays r07 or r09 while the second
r A
 1  ORC pv ccr A þ pv car A þ aud ð19Þ
group is activated when r08 substitutes relays r12 or r13.
Based on the protection philosophy adopted and basic substation
rEð1Þ rEð1Þ
f59 ¼ f27 ð20Þ data, every set presented in Section ‘Equations of the expected
states of protective relay functions’ can now be obtained. In the
Since relays from IL or OL bays trip circuit breakers through trip example, the expected state equations of the protective functions
transfer function, e.g. relays r01 and r02 from Fig. 1(a), Eq. (19) of r07 are considered. Since the functions of r07 are classified as type
takes into account trip transfer function alarms. While the second 2, the equations used are (1) and (4). Hence, the following sets are
type is considering relay CT installed in the CB or FD bays. These
defined: XR7 = {50, 50N, 51, 51N}, J 750 ¼ J 750N ¼ J 751 ¼ J 751N ¼ f5; 6; 7;
types of equations take the form: 5;7
7 7 7 7
8; 9; 10; 11; 12g, Xf50 ¼ Xf50N ¼ Xf51 ¼ Xf51N ¼ ;; Xca50 ¼ f4; 5; 6g,
8 2 0 139
> > Xca6;7
50 ¼ f3; 4; 5g, ca50
X7;7
¼ ca8;7 X 50 ¼ ca9;7 X50 ¼ ca10;7
X
50 ¼ ca11;7
50 X
< =
rEð2Þ 6 B X A C7 12;7 j;7 j;7 j;7
¼ ca50 ¼ f3; 4g, ca50 ¼ ca51 ¼ ca50N ¼ ca51N j 2 J 50 ; fd50 ¼
X X X X Xj;7
8 7 6;7
X
f27 ¼ MAX s j  41  ORC @ c k A5
j2J 27 >
r
: >
; 6;7 6;7 6;7
j;r
k2Xfc27 X fd51 ¼ fd50N ¼ fd51N ¼ f4g. From theses sets, it is defined:
X X
h
i 5;7 6;7 6;7 6;7
r A X ca50  fd50 ¼ f5; 6g,
X ca6;7
X 50  fd50 ¼ f3; 5g,
X ca7;7
50  fd50
X X
 1  ORC pv ccr A þ pv car A þ aud ð21Þ 6;7 6;7 6;7 6;7
¼ ca8;7
X X X 9;7
X X 10;7
50  fd50 ¼ ca50  fd50 ¼ ca50  fd50 ¼ ca50  fd50 X 11;7
X X
rEð2Þ rEð2Þ
f59 ¼ f27 ð22Þ 12;7
¼ ca50  fd50 ¼ f3g and
X X 6;7
X cN ¼ cN ¼ cN ¼ cN ¼ cN3;10 ¼
3;6 3;7
X X3;8 3;9
X X
Eq. (21) is simpler because there is not transfer of trip in CB or X cN ¼ cN ¼ f7;8g, cT ¼ c3;7
3;11
X 3;12 3;6
X X 3;8
T ¼ c T ¼ c T ¼ cT
X 3;9
X 3;10
X ¼ c3;11
T X
¼
FD bays. In addition, the equations for functions 87, 51G, 26, 49, 63, 3;12 5;5 5;6 5;5 5;6
X cT ¼ f3;4;25g, cN ¼ cN ¼ f15;16;32g, cT ¼ cT ¼ f13;14;
X X X X
63A, 80 and 71 consider only transformers protected and are writ- 28g, c6;5 c6;5 c4;5 4;6
X N ¼ f9;10g, X
T ¼ f3;4;26g, XN ¼ cN ¼ f11;12;31g,
X
ten as: 4;7 4;8 4;9
X cN ¼ f11;12g, cN ¼ f11;12;13;14;31g,
X cN ¼ f11;12;17g,
X X
  h
i
ffprE ¼ MAX sj  1  ORC pv ccr A
þ pv car A
þ aud
r A
ð23Þ c4;10
N ¼ f11;12;22;31;32g. For c X4;11
N and c
X 4;12
N two paths are possible
r
j2J fp
such as cN4;11 ð1Þ ¼ f11; 12; 18; 19g,
X c4;11
XN ð2Þ ¼ f11;12;13;14;29;
In order to apply the mathematical model to the monitored 31g and c4;12X N ð1Þ ¼ f11; 12; 20; 21; 31; 32g, c 4;12
N ð2Þ ¼ f11; 12; 13;
X
system, a protection philosophy must be established. The 14; 30; 31g. Each one of these two sets is considered separately in
protection philosophy stipulates: (1) which sections each the equations. The same reasoning is used in order to obtain
protective functions should protects and, therefore, the circuit 4;j 4;5
the sets XdN . Hence XdN ¼ XdN ¼ XdN ¼ f6g, XdN ¼ f6; 7g,
4;6 4;7 4;8

paths between a reference node and protected sections 4;9 4;10 4;11 4;11 4;8
X dN ¼ f6;9g, dN ¼ f6;10g, X XdN ð1Þ ¼ f6;11g, X dN ð2Þ ¼ Xd N ;
(final point of circuit paths), and (2) the coordination and selec- 4;12 4;12 4;8
tivity of protective functions (operation time, i.e., primary and X dN ð1Þ ¼ f6;12g, dN ð2Þ ¼ X X dN .
backup protection). Mathematically, the circuit paths are
established by means of the forming of sets Xffpr ; Xci;j i;j
N ; Xc T ; Unconstrained binary programming model
i;j i;j j;r k0 ;r j;r
XdN ; XdT ; Xcafp ; Xfdfp ; Xfcfp using data from the distribution sub-
station. The protected sections are taking into account through The process of fault diagnosis carried out by an operator at a
the forming of set J rfp . The latter set defines the quantity of pro- DCC is based on abductive logical inference. In this process, mani-
tected sections or, in other words, the quantity of terms between festations (alarms) are observed and by using abductive reasoning,
the curly brackets in Eqs. (1)–(23). a set of concurrent hypotheses or solutions (fault sections) that can
correctly explain every alarm occurrence observed is formulated
Example of forming of sets applied to model system by the operator. In this work, an extension of the UBP model pro-
In Fig. 1(a) main protection zones are indicated by two parallel posed in [22,25] is used to generate feasible solution hypotheses
dashed arrows. Thus the protective functions 50/50N/27/59 of the for fault section estimation. The UBP model, based on the parsimo-
relay r01, and 50/50N of r02 are primary protections of busbars nious covering theory using the minimality criterion [23,31], pre-
M-HV, T-HV, while functions 51/51N of both relays are backup pro- sents a plausible hypothesis or explanation for a set of informed
tection of all substation sections. Due to absence of coordination alarms when:
between functions 50/50N of r01 and r02 (short circuit at points
next to IL01 and OL01), functions 67/67N with LS are used. Hence, (1) The solution is a cover with the informed alarms, i.e., the
the protective functions 67/67N of r01 and r02 are primary informed alarms can be accounted for by the fault sections
protections of IL01 and OL01, respectively. In addition, the protec- in the solution.
tive functions 50/50N of r01 are blocked by LS through 67/67N of (2) The informed alarms are as consistent as possible with the
r02 while functions 50/50N of r02 are blocked through 67/67N of expected states of the protective functions of the relays that
r01. The arrows beside the functions indicate the directional were calculated for the solution.
feature. (3) The solution with a minimum number of fault sections is
The protective functions 50/50N of r03 and r04 are primary preferred to explain the fault occurrence.
protection of transformers TF01 and TF02 as well all functions of
r05 and r06. Moreover functions 51/51N of r03 and r04 are backup The three conditions established by the parsimonious covering
protection of TF01, TF02, busbars M-MV and T-MV, CB01, CB02, theory can be translated mathematically by minimizing the
and FD01 and FD02. Besides, the protective functions 50/50N of following objective function:
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 795

classflag fp j = 0

phaseclass fp j , 3φ = 0

s j incumbent = 1

s j incumbent = 1

classflag fp j = 1
fp j E = fp A

phaseclass fp j , 3φ = fp A, 3 φ

classflag fp j = fp A

(a) (b) (c)

Fig. 5. (a) Flow chart of the proposed AGA. (b) Stopping criterion and (c) fault classification algorithm.

! ! !
X
nf X
nf X
ns The first term on the right hand side of (24) defines whether or
Min E ¼ k1  Coi þ k2  Ini þ k3  si ð24Þ not a solution covers the informed alarms for a given fault scenario.
i¼1 i¼1 i¼1
This term is zero if the solution totally covers the informed alarms,
where nf is the total number of protective functions; ns is the total or otherwise, it provides the proximity that a solution has in order
number of sections. to cover the informed alarm.

s j incumbent = 1

classflag fp j = 1

s j incumbent , fp A, 3φ

(a)

s j incumbent

(b)
Fig. 6. Flow chart of overall methodology. (a) Main algorithm and (b) detailed diagnosis algorithm. Variables in figure: p: counter of diagnosis/solution; ndiag: maximum
number of diagnoses.
796 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805
Fig. 7. Real four-interconnected distribution substation system.
The second term on the right hand side of (24) defines the
inconsistency between the informed alarms and the expected
states of the protective functions. Therefore, Ini is calculated by:
 
 
Ini ¼ ffpAi  ffpE i  i ¼ 1; . . . ; nf
An alarm of a protection function is totally consistent with the
expected state when Ini = 0. When all alarms of a given fault situ-
ation are consistent with the expected state, then the second term
in (24) is zero. The closer this term is to zero, the more consistent
the alarms are with the expected states. For each one of the nf
functions, Coi and Ini are obtained using the rules presented in
Table 2.
The third term on the right hand side of (24) is included in the
objective function to comply with the third condition established
by the parsimonious covering theory.
Constants k1, k2 and k3 are integers, and the corresponding val-
ues must satisfy the condition: k1 > k2 > k3. Since this is a minimi-
zation problem is chosen k1, k2 and k3 to discard first solution that
does not cover any alarms reported, and secondly any solution that
is inconsistent with the alarms reported [23].
Adaptive genetic algorithm
The codification used in the proposed AGA is depicted in Fig. 4.
In this codification, all sections are grouped in a chromosome or
configuration. In each generation of AGA (Fig. 5(a)), the mathemat-
ical model given in Section ‘Equations of the expected states of
protective relay functions’ is used. First, the set of expected state
equations for the monitored electric power system is calculated
by inserting the sections states (generated first time randomly)
along with the alarms informed by SCADA system into the
expected state equations. Subsequently, the objective function
(24) is evaluated for each configuration.
In the pursuit of a global optimum, the classical genetic algo-
rithm can be trapped to a local optimum as soon as it is reached.
Since evidence of population saturation can be used to identify
ð25Þ
the occurrence of premature convergence, an efficient strategy
based on population diversity is proposed. The mechanism is based
on the use of the population diversification rate [27] to
dynamically control the evolutionary process. Population
diversification rate is given by:
Div ¼ 100  ðCeq=npÞ  100 ð26Þ
where Div is the percentage diversification rate and Ceq is the
maximum number of equal configurations calculated; and np is
the population size.
The strategy consists of determining the population diversifica-
tion rate in each generation of the AGA, using Eq. (26). The rates of
crossing over and mutation are dynamically and automatically
tuned based on diversity of the current population. Thus, the
crossing over and mutation rates are written as:
tcðqÞ ¼ ðDiv =100Þ  exp ððDiv =100Þ  1Þ ð27Þ
tv ðqÞ ¼ ð1  Div =100Þ  exp ðDiv =100Þ ð28Þ
where tc(q) is the crossing over rate and tv(q) is the mutation rate.
This strategy causes an immediate response to the population
saturation. Thus, adequate population diversity is maintained and
premature convergence is avoided. A single crossing over process,
i.e., the genetic material is permuted between two configurations
from a point randomly chosen generating two offspring, is adopted
along with one point mutation process, i.e., changing one randomly
chosen gene of 1–0 or vice versa.
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805
Fig. 8. Unifilar diagram of substation 01.
The central idea in the proposed stopping criterion (Fig. 5(b)) is
the usage of the fault classification algorithm (Fig. 5(c)). As a first
step toward solution convergence, the expected states of the relay
protective functions are determined for the best chromosome in
the current generation. These states are then used to classify the
fault, i.e., one-phase-to-ground, two-phase-to-ground, phase-to-
phase, three-phase or three-phase-to-ground short circuit.
In the fault classification algorithm, the term of the expected
state equation of protective function related to a given section j is
compared with its alarm (second conditional block of Fig. 5(c)).
The purpose here is to associate the operated alarm of the protec-
tive function with the term of expected state equation of the same
protective function. When the term, referring to section j of the
expected state equation of the protective function fp is equal to
the alarm of the same function and both are equal to 1, then it
can be stated that the fault in section j is associated to the operation
of the function fp. In other words, the alarm of fp is caused by a fault
j;3u
in section j. The variable phaseclassfp is updated with the three-
phase alarms of the protective function which are subsequently
used to classify the fault. This classification is done by comparing
the phases of the operated protective function so as to identify
the type of fault associated with the section j. For example,
j; a j; N j; b j; c
phaseclass50 ¼ phaseclass50 ¼ 1 and phaseclass50 ¼ phaseclass50 ¼ 0
implies a one-phase-to-ground short circuit in phase a, at section j.
On the other hand, is used in the stopping criterion to verify if the
alarm of function fp is associated with some section j. When all
alarms are associated with incumbent sections, i.e., all alarms can
be explained and related to all sections, then the flag variable
(Fig. 5(b)) is 1 for all faulted sections. Therefore the third condi-
tional block of Fig. 5(b) will always be false and the variable sc is
null, thus, provoking the stopping of AGA. A safety variable is used
to stop AGA if the convergence criterion is not satisfied after ngmax
generations.
797
Integration of AGA to the methodology
The overall methodology is presented in Fig. 6(a). The acquisi-
tion of alarms is made during a time window considering all
alarms reported immediately after the occurrence of a fault
(post-fault situation). Before acquisition, prefiltering and a
preprocessing should be done in order to extract non-redundant
information along with the alarms necessary for the execution of
the methodology.
Since the mathematical model of protection system operation
uses single-phase alarms, the filtering of three-phase alarms is
necessary. Hence, the ‘‘OR’’ Boolean operator is applied to each
set of three-phase alarms and where at least one phase is operated
then the protective function is considered operated. Besides, three-
phase alarms are also stored for use in the fault classification
algorithm.
Due to the fact that various protective functions protect the
same sections, it is possible that some terms of expected state
equations are equal for different protection functions. In order to
avoid calculation of these terms several times, an algorithm to
pre-compute these terms for all equations (pre-calculations block
in Fig. 6(a)) was developed.
The detailed diagnosis algorithm (Fig. 6(b)) uses the variable
j
classflag fp to verify if each incumbent section is a valid solution.
For each correct solution, the detailed diagnosis algorithm displays:
(1) the fault section; (2) the alarm relay associated with the section;
and (3) the fault type or the phase involved in the alarm. If the solu-
j
tion is not valid, i.e., classflag fp ¼ 0 for all protective relay functions,
then the variable flag is 0 and the algorithm displays the sections
incorrectly diagnosed. The global diagnosis is performed consider-
ing all diagnoses produced by AGA. The diagnoses are coherently
grouped and displayed. This type of diagnosis is simpler than the
detailed and presents all sections under fault.
798 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805
Fig. 9. Unifilar diagram of substation 02.
Results and discussion
The real electric power system shown in Fig. 7 is used to
evaluate the performance of the methodology. The distribution
substations unifilar diagrams are illustrated in Figs. 8–11. The
protection philosophy addressed in the forming of sets example
applied to model system presented in Section ‘Example of forming
of sets applied to model system’ of this paper will be used. More-
over, the reach of distance protection zones are illustrated in Fig. 7.
Several fault scenarios with a number of alarm sample
combinations are necessary in order to cover a great number of
fault situations capable of testing and validate the mathematical
model in a reliable and realistic manner. Due to the complexity
of the protection philosophy, the large number of alarms that can
be generated from the test system used, and the lack of real alarm
data from utilities the fault scenarios are very difficult and
laborious to be created. Therefore an intelligent technique should
be specifically developed to generate alarms in an efficient, flexible
and easy manner.
Alarm probabilistic generator algorithm
At this point, APGA (Fig. 12), is proposed to automatically and
consistently generate relays and circuit breaker alarms operation
covering a great number of real fault situations. This algorithm
probabilistically generates alarms using as input data: (1) probabil-
ity of relays and circuit breaker operate; (2) switchgear and trip
transfer function status and (3) state of each section phase. Since
the topology of substation (defined by (2) data) named herein fault
scenario (nfault) can be the same for a number of alarm sets, APGA
can generate a great number of relays and circuit breaker alarms
(nalarms) from a unique fault scenario (see Fig. 12(a)). This feature
is very interesting since the human effort is greatly reduced. The
relay alarms are randomly generated and matched with the
adopted protection philosophy and operation logic of the protec-
tive functions by using the protection system operations mathe-
matical model. On the contrary, the probabilistic generation of
circuit breaker operations is performed using the single phase relay
alarms based on the rule that a circuit breaker is open only if the
relay that trips the circuit breaker is operated and a random num-
ber generated is less than the probability of the circuit breaker to
operate. Hence, in order to generate circuit breaker alarms, the
probabilistic generation block of circuit breaker operations
(Fig. 12) is performed twice. First one, to estimate the expected
states of protective relay functions, and second one to use the final
relay alarms.
Simulations results
The proposed methodology was implemented in C++ program-
ming language and the simulations performed using a PC with
Intel(R) Core(TM) 2 Duo processor, 2.0 GHz and 4.0 Gb RAM. The sim-
ulations were carried out using six fault scenarios presented in
Table 3 which, in addition, also presents one set of alarms gener-
ated by APGA (output alarms column). The accuracy and efficiency
of AGA were compared with a dedicated GA [22].
The global diagnoses for every set of alarm generated from
Table 3 after 100 executions (ndiag = 100) of the AGA are presented
in Table 4 with all diagnoses results manually checked and consid-
ered correct. All different solutions obtained from the AGA, as well
as the percentage occurrences and the minimum and maximum
convergence time are presented in increasing order relative to
F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 799

Fig. 10. Unifilar diagram of substation 03.

Fig. 11. Unifilar diagram of substation 04.

800 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

Table 3
Six fault scenarios and one set of alarms generated by APGA (nfault = 6,nalarms = 1).

Fault Input data Alarms generated t

Scenario (s)  103
Fault Switchgears opened(substation) Trip transfer function Relay number (operated Circuit breaker
sectionsj state(substation) functions)(substation) opened(substation)
(phase)
0 1(aN); 7(1); 8(1); 13(1); 14(1); 17(1); 18(1); 19(1); 1N(1)–8N(1); 1N(2); r01(51ac/67a/67N)(1); r02(50a/50N)(1); 1(1); 2(1); 4(1); 8(1); 16
2(aN); 20(1); 45(1)–63(1); 7(2); 8(2); 29(2); 30(2); 2N(2); 3T(2); 4N(2)– r09(50ac)(1); r14(50ac/46ac/50BF)(1); 10(1); 1(2); 3(2); 4(2);
12(ac); 39(2); 40(2); 42(2)-49(2); 51(2)–54(2); 7N(2); 1N(3)–5N(3); r01(50a/50N)(2); r03(50a/50N/79)(2); 8(2); 16(2); 1(3)-4(3);
37(aN); 56(2); 5(3); 6(3); 31(3)–43(3); 1(4); 2(4); 1T(4); 2T(4) r09(51a/51N/51NS/79)(2); r01(50a/ 1(4)
44(aN); 17(4)–21(4) 50N/27a)(3); r02(50a/50N/27a)(3);
50(aN) r03(51a/51N/79)(3); r05(51a/51N)(3);
r08(51a/51N/50BF)(3); r09(51a/51N/
50NS/50BF)(3)
1 11(aN); 7(1); 8(1); 17(1); 18(1); 45(1)–65(1); 5(2); 1N(1)–8N(1); 1N(2)– r01(51a/51N/50BF)(1); r02(51a/51N)(1); 2(1); 4(1); 1(2)-4(2); 31
27(cN); 6(2); 15(2); 16(2); 39(2)–56(2); 5(3); 6(3); 7N(2); 1N(3)–5N(3); r03(51a/51N/50BF)(1); r06(51a/51N/ 1(3); 3(3)-11(3); 2(4)–
48(cN); 12(3); 13(3); 31(3)–35(3); 37(3)–43(3); 1N(4); 2N(4) 50BF)(1); r11(51a/51N/50BF)(1); 3(4)
67(cN) 3(4); 4(4); 6(4); 7(4); 16(4); 17(4); 19(4)– r01(51c/51N)(2); r04(51c/51N/50BF)(2);
21(4) r01(51c/51N)(3); r09(27c)(3);
r10(27c)(3); r11(27c)(3); r12(27c)(3);
r13(27c)(3); r14(27c)(3); r15(27c/
50BF)(3); r07(50NS/46c/50BF)(4)
2 1(bN); 3(1); 4(1); 15(1); 16(1); 45(1); 47(1)–50(1); 1N(1); 2T(1); 3N(1)- r01(50b/50N)(1); r02(50b/50N/59b)(1); 1(1)–4(1); 8(1); 9(1); 31
5(bN); 52(1)–63(1); 65(1); 7(2); 8(2); 15(2); 16(2); 5N(1); 6T(1); 7N(1); r03(50b/50N)(1); r12(27ab/50BF)(1); 14(1); 1(2); 2(2); 4(2);
9(ab); 39(2); 40(2); 42(2)–56(2); 5(3); 6(3); 23(3); 8N(1); 1N(2); 2N(2); r14(27ab/50BF)(1); r15(27ab)(1); 3(3); 11(3); 2(4)
16(bN) 24(3); 31(3)–39(3); 41(3)–43(3); 3(4); 4(4); 3T(2); 4N(2)–7N(2); r16(27ab/50BF)(1); r18(50b/50N/27ab/
16(4)–21(4) 1N(3)–5N(3); 1N(4); 50BF)(1); r01(51b/51N)(2); r03(50b/
2N(4) 50N/67b/67N/50BF)(2)
3 45(A⁄); 1(1); 2(1); 17(1); 18(1); 41(1)–44(1); 46(1)– 1T(1); 2N(1)–8N(1); r01(50ac)(3); r02(50ac/50BF)(3); 1(1); 22(1); 23(1); 47
46(A⁄); 65(1); 1(2); 2(2); 5(2); 6(2); 11(2); 12(2); 1N(2)–7N(2); 1N(3)– r03(50ac)(3); r04(51ac)(3); r05(50ac/ 3(2); 11(2); 17(2)–
49(ac); 15(2)–18(2); 20(2); 31(2)–56(2); 9(3); 3N(3); 4T(3); 5N(3); 50BF)(3); r06(63A)(3); r09(59ac/ 20(2); 1(3); 3(3)–5(3);
66(abc) 31(3)–33(3); 35(3)–43(3); 3(4)–4(4); 16(4)– 1N(4); 2N(4) 50BF)(3); r06(50abc/79)(4) 2(4); 5(4)
21(4)
4 7(A⁄); 1(1); 2(1); 7(1)–12(1); 15(1)–18(1); 41(1)– 1N(1)–8N(1); 1N(2)– r08(63/63A/80/26/50BF)(1); 1(1)-4(1); 22(1); 31
29(bN); 63(1); 9(2); 10(2); 33(2); 34(2); 39(2)– 3N(2); 4T(2); 5N(2)– r11(pvca)(1); r05(51b/51N/50BF)(2); 23(1); 1(2)-5(2); 8(2);
32(bN); 41(2); 43(2)–51(2); 53(2)–55(2); 3(3); 4(3); 7N(2); 1N(3); 2T(3); r08(51b/51N/50BF)(2); r09(51NS/46b/ 11(2); 13(2)–15(2);
39(bN); 7(3); 8(3); 31(3); 33(3)-43(3); 1(4)–21(4) 3N(3); 4T(3); 5T(3); 79)(2); r10(50b/50N/50BF)(2); 18(2)–20(2); 1(3);
46(abA⁄) 1N(4); 2N(4) r11(27b)(2); r12(50NS/27b)(2); r13(27b/ 2(3); 2(4); 4(4)–6(4)
50BF)(2); r14(27b)(2); r15(27b)(2);
r16(27b)(2); r17(27b/50BF)(2); r18(27b/
50BF)(2); r20(27b)(2); r21(27b)(2);
r01(51ab)(3); r04(50ab)(3); r06(87ab/
80/26/49/71/50BF)(3)
5 7(abA⁄); 1(1)–4(1); 7(1)–12(1); 15(1)–18(1); 21(1); 1N(1)–8N(1); 1N(2)– r06(50ab/50BF)(1); r08(51G/63/63A/ 1(1)–4(1); 11(1); 31
41(bN); 23(1)–32(1); 45(1)–65(1); 7(2); 8(2); 11(2); 6N(2); 7T(2); 1N(3)– 26/49/71/50BF)(1); r21(51NS/79)(2); 13(1)–17(1); 6(2);
44(bc) 12(2); 25(2); 26(2); 39(2)–42(2); 44(2)– 5N(3); 1N(4); 2N(4) r01(50bc/27bc/50BF)(3); r02(51bc/ 14(2); 20(2); 3(3);
47(2); 49(2)–51(2); 53(2); 54(2); 56(2); 27bc/50BF)(3) 4(3); 2(4); 4(4)–6(4)
5(3)–8(3); 31(3)–43(3); 1(4)–21(4)
⁄
Alarms not involving any one of the phases a, b, c or N, i.e., alarms of the type 50BF, 79, 63, 63A, 26.

the corresponding value of the objective function. The goal is to
exhaustively survey the mathematical model and overall method-
ology performance. The two parameters used in AGA are: np = 100
and ngmax = 1000 and the eight used in GA [22] are: np = 100,
ngmax = 1000, kc = 0.30, tc (0) = 0.75, kv = 0.03, tv (0) = 0.001,
tp = 0.5 and Divx = 25. The parameters are tuned in an exhaustive
trial-and-error basis. Constants k1, k2 and k3 are respectively
1000, 10 and 1. The maximum average time is 1.357(s) (fault sce-
nario 4) for AGA and 1.835(s) (fault scenario 3) for GA and the max-
imum time for convergence is 3.495(s) (fault scenario 0) for AGA
and 11.172(s) (fault scenario 1) for GA. In Table 5 are shown the
detailed diagnoses of the best solutions for AGA and GA [22]. The
results from Table 5 (two last columns) show that AGA is faster than
GA in all diagnoses, except for fault scenario 3, where the time dif-
ference is 0.416 s. In addition AGA has found more suboptimal solu-
tions compared with GA, being 79 in total from AGA versus 65 from
GA (GA did not find solution p = 86 for fault scenario 4 from Table 5).
Except for fault scenario 1, the best solutions presented in
Table 4 are obtained repeatedly several times. In fact, in fault sce-
nario 1, the best suboptimal solution at p = 1 (first occurrence) is
responsible for 46% of the resulting solutions while the first best
solution obtained by AGA at p = 42 is attributed to only 5% of the
obtained solutions. Other suboptimal solutions such as p = 15
and p = 8 were found to occur more frequently than the minimum.
In fact, on analyzing the protective functions associated with the
faulted sections for the solutions p = 42 and p = 1 for fault scenario
1 in Table 5, it is easy to observe that the only sections common to
both solutions are the sections where there are associated primary
protections, i.e., protective function 27 and 50NS. The remaining
sections are not common to both diagnoses and present only
associated backup protections. Commonly, backup protection
protects a number of sections, and their operation along with the
failure of main protections as well as circuit breakers increase
the complexity of the alarms. This is interpreted mathematically
as increasing the search space and the number of solutions
obtained, and as such, the greater difficulty of convergence to the
optimal solution. A detailed discussion on the complexity and
uncertainties of alarms and their relation to the multimodal
characteristic of objective function E is addressed in [22].
For each fault section, the approach herein proposed is capable
of associating the relay alarm operated as presented in Table 5,
thus indicating to the operator the involved phase. Additional
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 801

Table 4
Global diagnoses for all solutions (ndiag = 100) for AGA – alarms generated by APGA (Table 3).

Fault p Emin Times (%) t (s) p t (s) Emin Fault section(substation) Relay Circuit breaker
scenario average Min/ conditions conditions(substation)
Max
0 1; 3; 77; 4; 38; 1557; 1576; 1578; 19; 13; 1; 1.147 14 0.430 2094 LT1–4; LT2–1; LT3–1; normal out of service: 4(1);
22; 19; 6; 54; 35; 1597; 1617; 1618; 11; 6; 4; 1; IL03(1); M-HV(1); operation 7(1); 9(1); 10(1); 4(2);
36; 2; 26; 53; 42; 1628; 1637; 1647; 8; 1; 1; 2; 3; CB01(1); CB02(1); 16(2); 3(3); 1(4) check:
58; 92; 99; 70; 1649; 1657; 1658; 2; 1; 1; 1; 1; FD01(1); FD02(1); 13(1); 5(3); 6(3)
17; 80; 95; 64; 1659; 1668; 1678; 1; 1; 2; 1; 1; FD04(1); FD05(1);
44; 86; 97; 7; 84; 1680; 1688; 1689; 2; 2; 1; 2; 1; FD06(1); FD08(1);
47; 39; 32; 49; 1699; 1708; 1709; 1; 2; 2; 2; 1; FD09(1); FD10(1);
13 1719; 1728; 1730; 2 IL01(2); IL02(2); M-
1749; 1770; 1791; HV(2); T-HV(2); T-
1810; 1850; 1852; MV(2); FD05(2); IL01(3);
1912; 1994; 2094 IL02(3); M-HV(3);
TF02(3); M-MV(3);
CB01(3); CB02(3);
FD02(3); FD03(3);
FD05(3); FD06(3);
FD07(3)
31 3.495 1597
1 42; 1; 15; 17; 8; 484; 494; 504; 544; 5; 46; 10; 1; 1.031 32 0.601 656 CB02(1); FD04(1); normal out of service: 4(1);
7; 9; 40; 67; 99; 554; 564; 575; 625; 21; 4; 3; 1; FD06(1); FD07(1); operation 9(1); 3(2); 8(2); 3(3);
4; 18; 31; 61; 626; 635; 645; 646; 1; 1; 1; 2; 2; FD08(1); FD09(1); 2(4) check: 1(1); 3(1);
100 656; 685; 756 1; 1 FD10(1); FD11(1); 7(1); 10(1); 5(2); 12(3);
IL01(2); TF01(2); M- 6(4)
MV(2); CB01(2);
FD01(2); FD02(2);
FD03(2); FD04(2);
FD05(2); FD06(2);
FD09(2); IL01(3); M-
HV(3); M-MV(3); A-
MV(3); FD06(3); FD02(4)
30 1.582 564
2 3; 5; 48; 9; 6; 31; 344; 364; 365; 405; 48; 18; 3; 5; 1.068 29 0.639 586 LT2–1; IL03(1); M- normal out of service: 2(1);
36; 38; 80; 4; 28; 425; 426; 446; 447; 1; 1; 2; 1; 1; HV(1); T-HV(1); M- operation 8(1); 4(2); 8(2); 3(3);
30 1; 33; 15; 2; 467; 484; 495; 496; 5; 1; 1; 1; 2; MV(1); T-MV(1); 11(3); 2(4) check:
7; 42; 29; 37; 67; 504; 545; 557; 565; 3; 1; 1; 1; 1; FD05(1); IL01(2); 11(1); 13(1); 15(1);
12 566; 576; 586; 658; 1; 1; 1 IL02(2); M-HV(2); T- 17(1); 3(2)
696; 717 HV(2); TF01(2); FD01(2);
FD02(2); FD03(2);
FD04(2); FD05(2);
FD06(2); FD07(2);
FD08(2)
27 1.546 344
3 1; 87; 45; 81; 52 383; 404; 424; 425; 504 94; 1; 2; 1; 1.253 43 0.765 383 IL02(3); M-HV(3); T- normal out of service: 1(1);
2 HV(3); TF01(3); TF02(3); operation 9(1); 22(1); 23(1); 1(2);
M-MV(3); A-MV(3); 3(2); 6(2); 8(2); 9(2);
FD01(4) 11(2); 17(2); 18(2);
19(2); 20(2); 2(4)
check: 2(3); 6(3)
1 1.881 383
4 1; 3; 86 545; 546; 586 80; 19; 1 1.357 26 0.834 545 TF02(1); M-MV(2); T- check: out of service: 1(1);
MV(2); CB02(2); r11(pvca)(1) 4(1); 5(1); 6(1); 8(1);
FD07(2); IL01(3); 9(1); 22(1); 23(1); 5(2);
TF01(3) 18(2); 2(3); 4(3); 1(4)-
6(4) check: 7(1); 10(1);
6(2); 7(2); 9(2); 10(2);
12(2); 16(2); 17(2);
3(3); 5(3)
57 2.079 545
5 1 243 100 1.313 85 0.870 243 TF02(1); FD09(2); M- normal out of service: 1(1);
HV(3) operation 2(1); 4(1); 5(1); 6(1);
8(1); 9(1); 11(1); 13(1)-
17(1); 4(2); 6(2); 14(2);
3(3); 4(3); 1(4)-6(4)
check: 7(1); 10(1);
1(3); 2(3)
95 1.838 243

information about the function of the protective relaying operation (three last columns of Table 4) to obtain an idea of which sections
can be included in the diagnosis, e.g., 50a-instantaneous overcur- are under fault, the detailed diagnosis (Table 5) may be checked in
rent-phase ‘‘a’’ operated, 67b-directional phase overcurrent-phase order to determine the involved phases (short circuit type) and
‘‘b’’ operated, etc. Once the operator has used the global diagnosis relay alarm associated with the faulted sections. The detailed
802 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

Table 5
Detailed diagnoses of the best solutions (ndiag = 100) for AGA and GA [22] – alarms generated by APGA (Table 3).

Fault p Fault Alarm of relay operated and phase associated(substation) Emin t (s)
Scenario section(substation)
AGA GA
0 1 LT3–1 r01(51ac/67a/67N)(1); r01(50a/50N)(3); r02(50a/50N)(3); r03(51a/51N/79)(3) 1557 1.487 3.928
IL03(1) r01(51ac)(1); r02(50a/50N)(1)
FD01(1) r09(50ac)(1); r14(50ac/46ac/50BF)(1)
IL01(2) r01(50a/50N)(2); r03(50a/50N)(2)
FD05(2) r09(51a/51N/51NS/79)(2)
M-HV(3) r01(50a/50N/27a)(3); r02(50a/50N/27a)(3); r03(51a/51N)(3)
CB01(3) r03(51a/51N)(3); r05(51a/51N)(3); r08(51a/51N/50BF)(3); r09(51a/51N/50NS/50BF)(3)
1 42 FD10(1) r01(51a/51N/50BF)(1); r02(51a/51N)(1); r03(51a/51N/50BF)(1); r06(51a/51N/50BF)(1); r11(51a/51N/ 484 0.890 1.593
50BF)(1)
(2)
TF01 r01(51c/51N)(2); r04(51c/51N/50BF)(2)
M-MV(3) r01(51c/51N)(3); r09(27c)(3); r10(27c)(3); r11(27c)(3); r12(27c)(3); r13(27c)(3); r14(27c)(3); r15(27c/50BF)(3)
FD02(4) r07(50NS/46c/50BF)(4)
1 FD07(1) r01(51a/51N/50BF)(1); r02(51a/51N)(1); r03(51a/51N/50BF)(1); r06(51a/51N/50BF)(1); r11(51a/51N/ 494 0.938 1.903
50BF)(1)
FD03(2) r01(51c/51N)(2); r04(51c/51N/50BF)(2)
M-MV(3) r01(51c/51N)(3); r09(27c)(3); r10(27c)(3); r11(27c)(3); r12(27c)(3); r13(27c)(3); r14(27c)(3); r15(27c/50BF)(3)
FD02(4) r07(50NS/46c/50BF)(4)

2 3 LT2–1 r01(50b/50N)(1); r02(50b/50N)(1); r03(50b/50N)(1); r01(51b/51N)(2); r03(50b/50N/67b/67N/50BF)(2) 344 1.210 2.497

T-HV(1) r01(50b/50N)(1); r02(50b/50N/59b)(1); r03(50b/50N)(1)
T-MV(1) r12(27ab/50BF)(1); r14(27ab/50BF)(1); r15(27ab)(1); r16(27ab/50BF)(1); r18(27ab/50BF)(1)
FD05(1) r18(50b/50N/50BF)(1)

3 1 TF01(3) r01(50ac)(3); r02(50ac/50BF)(3); r03(50ac)(3); r04(51ac)(3); r06(63A)(3) 383 1.881 1.465

A-MV(3) r01(50ac)(3); r02(50ac/50BF)(3); r03(50ac)(3); r04(51ac)(3); r05(50ac/50BF)(3); r09(59ac/50BF)(3)
FD01(4) r06(50abc/79)(4)

4 1 TF02(1) r08(63/63A/80/26/50BF)(1) 545 1.551 2.22

M-MV(2) r05(51b/51N/50BF)(2); r08(51b/51N/50BF)(2); r10(50b/50N/50BF)(2); r11(27b)(2); r12(27b)(2); r13(27b/
50BF)(2); r14(27b)(2); r15(27b)(2); r16(27b)(2); r17(27b/50BF)(2); r18(27b/50BF)(2); r20(27b)(2); r21(27b)(2)
CB02(2) r12(50NS)(2)
FD07(2) r09(51NS/46b/79)(2)
TF01(3) r01(51ab)(3); r04(50ab)(3); r06(87ab/80/26/49/71/50BF)(3)
3 TF02(1) r08(63/63A/80/26/50BF)(1) 546 1.234 1.655
M-MV(2) r05(51b/51N/50BF)(2); r08(51b/51N/50BF)(2); r10(50b/50N/50BF)(2); r11(27b)(2); r12(27b)(2); r13(27b/
50BF)(2); r14(27b)(2); r15(27b)(2); r16(27b)(2); r17(27b/50BF)(2); r18(27b/50BF)(2); r20(27b)(2); r21(27b)(2)
T-MV(2) r09(51NS/46b)(2); r12(27b)(2); r18(27b/50BF)(2); r20(27b)(2); r21(27b)(2)
CB02(2) r12(50NS)(2)
FD07(2) r09(51NS/46b/79)(2)
TF01(3) r01(51ab)(3); r04(50ab)(3); r06(87ab/80/26/49/71/50BF)(3)
86 TF02(1) r08(63/63A/80/26/50BF)(1) 586 0.895 –
M-MV(2) r05(51b/51N/50BF)(2); r08(51b/51N/50BF)(2); r10(50b/50N/50BF)(2); r11(27b)(2); r12(27b)(2); r13(27b/
50BF)(2); r14(27b)(2); r15(27b)(2); r16(27b)(2); r17(27b/50BF)(2); r18(27b/50BF)(2); r20(27b)(2); r21(27b)(2)
CB02(2) r12(50NS)(2)
FD07(2) r09(51NS/46b/79)(2)
IL01(3) r01(51ab)(3)
TF01(3) r01(51ab)(3); r04(50ab)(3); r06(87ab/80/26/49/71/50BF)(3)

5 1 TF02(1) r06(50ab/50BF)(1); r08(51G/63/63A/26/49/71/50BF)(1) 243 1.252 2.163

FD09(2) r21(51NS/79)(2)
M-HV(3) r01(50bc/27bc/50BF)(3); r02(51bc/27bc/50BF)(3)

Table 6
Accuracy of the AGA (ndiag = 100) – 60 Set of alarms Generated by APGA (nfault = 6, nalarms = 10).

Fault Error per fault Errora Accuracya Set of p t (s) Emin Error per set of Errorb Accuracyb Overall Errorc Accuracyc
scenario scenario (%) (%) alarm alarm (%) (%) error (%) (%)
1 1 0.1 99.9 0 48 15.90 387 1 1 99 2 0.033 99.967
3 1 4 33 16.87 415 1
a
Percentage calculated for 1 fault scenario for 10 sets of alarms generating 100 diagnoses per set, thus 1000 fault diagnoses.
b
Percentage calculated for 1 set of alarms generating 100 diagnoses per set, thus 100 fault diagnoses.
c
Percentage calculated for 60 sets of alarms generating 100 diagnoses per set, thus 6000 fault diagnoses.

diagnosis can also be used by the operators to study the protection
behavior for a number of faults in an off-line manner.
Ideally, it is interesting the proposed methodology can be
executed more than once or for ndiag > 1. In this way, the method-
ology can properly explore the search space, thus obtaining
suboptimal solutions. Sections pertaining to the suboptimal solu-
tions have less probability to be under fault, although they should
be checked. On the other hand, a single diagnosis (ndiag = 1) can be
very useful in real-time decision making. In the case of alarms
represented in Table 3, a maximum number of three diagnoses
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805 803

Table 7
Accuracy of the GA [22] (ndiag = 100) – 60 Set of alarms generated by APGA (nfault = 6, nalarms = 10).

Fault Error per Error(a) Accuracy(a) Set of p t (s) Emin Error per Error(b) Accuracy(b) Overall Error(c) Accuracy(c)
scenario fault (%) (%) alarm set of alarm (%) (%) error (%) (%)
scenario
3 1 0.1 99.9 9 63 15.23 354 1 1 99 4 0.067 99.934
5 3 0.3 99.7 5 39; 16.04 399 3 3 97
59; 15.31 425
63; 16.16 378

Table 8
Best solution and incorrect solution from AGA (ndiag = 100) – 60 Sets of alarms generated by APGA.
(substation)
Fault Incorrect Emin Times Fault Associated relay alarm
scenario solution/best p (%) section(substation)
1 48 387 1 CB02(1) r13(50NS)(1)
T-HV(2) –
TF01(2) r04(50c/50N)(2); r06(87c)(2)
T-MV(2) –
M-MV(3) r09(27c/50BF)(3); r10(27c/50BF)(3); r11(27c/50BF)(3); r13(27c)(3); r14(27c/50BF)(3); r15(27c)(3);
r16(27c/50BF)(3); r17(27c/50BF)(3)
T-HV(4) –
FD02(4) r07(50NS/46c/79)(4)
1 384 75 CB02(1) r13(50NS)(1)
TF01(2) r04(50c/50N)(2); r06(87c)(2)
M-MV(3) r09(27c/50BF)(3); r10(27c/50BF)(3); r11(27c/50BF)(3); r13(27c)(3); r14(27c/50BF)(3); r15(27c)(3);
r16(27c/50BF)(3); r17(27c/50BF)(3)
FD02(4) r07(50NS/46c/79)(4)

Table 9
Diagnoses from AGA and GA [22] (ndiag = 1) – 600 Sets of alarms (nfault = 6, nalarms = 100) and 1200 Sets of alarms (nfault = 6, nalarms = 200).

Number of alarms Algorithms Total simulation time (min.) Fault scenario Set of alarm t (s) Min/Max Emin
600 AGA 12.63 1 73 0.45 1010
1 1 5.06 1535
GA 15.75 0 78 0.53 3266
3 13 6.75 303

1200 AGA 29.03 0 22 0.36 4437

5 107 6.16 364
GA 31.82 0 185 0.45 1387
1 103 7.36 354

(ndiag = 3) can be used to draw a realistic conclusion about the dis-
tribution substation conditions (the best solutions for fault scenar-
ios 0 and 2–5 are obtained and also the best suboptimal solution
for fault scenario 1 is obtained as presented in Table 4).
In order to analyze the accuracy and efficiency of the approach,
60 alarms generated by APGA based on the six fault scenarios given
in Table 3 were used. Results from AGA and GA are presented in
Tables 6 and 7 and the analysis of the accuracy is carried out con-
sidering three levels: fault scenario level, set of alarm level and
overall level. As may be seen in Tables 6 and 7, for AGA a maximum
error of 1% is found for the alarm level and for GA a maximum error
of 3% is found for the same level. It is noteworthy that GA has pre-
sented 4 overall errors (Table 7) while AGA only 2 errors (Table 6).
A comparative analysis between best solution and incorrect
solution for fault scenario 1 is presented in Table 8. Note that the
best solution found is 75% of 100 executions while the incorrect
one is only 1% which demonstrates the efficiency of the approach.
In the case of fault scenario 3, the best solution found is 74% while
the incorrect solution is 1%.
The important feature of the approach is the capability of iden-
tifying if a diagnosis is correct or not. The misdiagnosed sections
(sections no associated with any relay alarm) are given in the last
two columns of Table 8. For each section diagnosed incorrectly
(highlighted), there is no associated relay alarm in the right col-
umn. Verifying that these sections are misdiagnosed may be done
by comparing misdiagnosed sections presented in incorrect solu-
tion (p = 48) with the minimal solution (p = 1). Also, misdiagnosed
section solutions can be used to draw an accurate diagnosis if those
sections that are not associated to the relay alarm are simply dis-
carded. This reasoning can be extended to analysis of any solution.
In order to verify the robustness and computational efficiency
of AGA and GA during an extremely fast time response, 1800
alarms generated by APGA divided into 600 and 1200 were used
(Table 9). For all diagnoses was adopted ndiag = 1 and all sets of
alarms were correctly diagnosed. Furthermore, for all diagnoses
the minimum and maximum processing times of GA were greater
than AGA.
Case study: fault scenario 4
Specifically to fault scenario 4 (Tables 4 and 5), the following
out-of-service sections were considered for the system depicted
in Fig. 7: Lines LT_3–1 and LT_1–4, transformer TF01 and FD11 at
DS 01 and the whole DS 04 (repotencialization of substation
transformer condition). Considering the alarms presented in
804 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805

opened), the trip of protective functions 51NS, 46 phase b and 79

fp A = fp prob × fp E
fp A, 3φ = 0
fp A = 1
( fp prob )
fp j E = 1
fp A, 3φ = s j 3φ
( fp E )
(b)
(a)
Fig. 12. Flow chart of the APGA. (a) Main algorithm and (b) generation of relay
alarms algorithm. Variables in figure: nfault: maximum number of fault scenarios;
nalarms: maximum number of alarm sets generated for each fault scenario; fpprob:
probabilistic alarm generated (single phase) for protective functionfp; s3/
j : state of
section j considering phases a, b, c, and N. (1-phase under fault and 0-normal).
Table 3 for fault scenario 4, one best solution p = 1, and two subop-
timal solutions p = 3 and p = 86, were obtained (see Tables 4 and 5).
Based on the protection philosophy adopted and knowledge of the
operation logic of the protection system, it is plausible to ascertain
the correctness of solutions for fault scenario 4. Solution p = 1 is
found to be the best solution because it presents the least faulted
sections and it is able to account for informed alarms in a reliable
and safe manner. The analysis of fault diagnosis for the above fault
scenario is as follows: a tripping of functions 63, 63A, 80 and 26 of
relay r08 at DS 01 is due to a fault at transformer TF02(1). Due to
the circuit breakers d07(1) and d10(1) not opening, function 50BF
sends a trip to d02(1) and d03(1) thus de-energizing the whole sub-
station because line LT_3–1 is out of service. Since d07(1) and d10(1)
are not open, there is a failure in these equipments. With relay r09
substituting r19 at DS 02 (c15, c16 and c52 closed and c33, c34
of r09 is due to a fault in feeder FD07(2) and this faulted section
is isolated through d08(2). The protective function 50NS of the relay
r12 trips is due to a fault in capacitor bank CB02(2). This section is
isolated through circuit breaker d11(2). Simultaneously a short-cir-
cuit in busbar M-MV(2) results in tripping of protective functions
51 phase b, 51N and 50BF of r05, 51 phase b, 51N and 50BF of
r08, 27 phase b of r20 and r21, 50 phase b, 50N and 50BF of r10
and 27 phase b of r11–r18. In this case, relay r10 sends a trip signal
to d09(2) but this equipment does not open causing backup protec-
tion r05 operates. Since d06(2) does not open, function 50BF of r05
sends a trip to d01(2), d02(2) and d04(2) through trip transfer
switches 43–1(N), 43–2(N) and 43–3(N), respectively. Moreover,
due to failure of d07(2), protective function 50BF of r08 sends a trip
to d03(2) through 43–4(T). Since protective function 51NS of r09
operates and this function protects busbar T-MV(2) and feeder
FD07(2), it is hard point out exactly which of these sections are
faulted using just this knowledge. On the other hand, protective
function 27 phase b of relays r12, r18, r20 and r21 is operated
and these protective functions protect busbar M-MV(2) as well as
T-MV(2). From an operator’s point of view, this analysis results in
a probable fault in busbar T-MV(2) (suboptimal solution p = 3)
and this section must be checked too. An analogous reasoning
can be used to explain faulted sections diagnosed TF01(3) (all three
solutions) and IL01(3) (suboptimal solution p = 86). Fig. 13 illus-
trates the performance of AGA and GA for fault scenario 4 showing
that AGA converges to the best solution before GA.
Conclusions
In this paper a novel mathematical model for fast fault section
estimation in distribution control centers was presented. The
model is divided into two parts: (1) a mathematical model for pro-
tection system operations and (2) a UBP model. The major contri-
butions of this work are provided by first part of the model such as:
– The equations developed can be modified at any time, the
model also presents scalability, and thus, the equations can be
subtracted or added without laborious work. This feature is
ideal for the model applicability to several substations (DCC
environment).
– The model presents modularity and flexibility. The equations
can be formulated for any relay at the substation. This feature
is ideal for digital multifunction relays since the protective
functions can be activated or not depending on the protection
philosophy adopted. Furthermore, with the protection
philosophy defined, the model can be applied to the distribution
substation with several control and protection layers.

AGA-Solution 1 (545) AGA-Solution 3 (546) AGA-Solution 86 (586) GA-Solution 2 (545) GA-Solution 7 (546)
2200
2100 720
2000
1900 690
1800 660
1700
1600 630
1500
Emin

1400 600
1300
1200 570
1100
1000 540
900 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85
800
700
600
500
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113 117 121125129
Generation

Fig. 13. Performance of AGA and GA [22] – fault scenario 4 (Tables 4 and 5).
 F.B. Leão et al. / Electrical Power and Energy Systems 63 (2014) 787–805
– The equations are computationally lightweight and they can be
implemented computationally through summation, subtraction
and multiplication. Also, the model presents robustness which
implies that any change in the substation topology does not
impose changes in the equations. This feature is important in
distribution substations because this type of substation
suffers high frequency of topology changes in the maintenance
of circuit breakers, switchgears and parallel operation of
transformers.
In order to solve UBP model is proposed the AGA. This algorithm
has crossing over and mutation rates automatically tuned in each
generation based on population saturation. Also, the algorithm
has only two control parameters to tune manually. Using a dedi-
cated stopping criterion based on fault classification, the AGA is
capable of associating relay alarms to faulted sections and identify-
ing the correctness of diagnosis produced, thus presenting great
advances in the problem area.
In order to test and validate the methodology, a high number of
alarms from a real four-interconnected distribution substation sys-
tem were generated automatically using the APGA. This algorithm
is capable of efficiently, rapidly, flexibly generating an infinite
number of alarms.
Obtained results show that the approach is adequate in estimat-
ing faults in a complex system of distribution substations with
multiple solutions (suboptimal solutions) from alarm reports of
simple, complex (simultaneously operated alarms) or misopera-
tion situations of the protective devices. Based on an exhaustive
simulation, AGA has proved to be better than dedicated GA. The
robustness, rapidity and efficiency of the overall methodology pro-
posed also were proven. Furthermore the methodology proposed
can be a powerful tool in aiding operators in the task of real-time
fault diagnosis in distribution control centers environment.
Acknowledgment
The authors acknowledge the financial support from FAPESP
under Grant 06/02569-7 and CNPq under Grant 305371-2012-6.
References
[1] Fan J, Borlase S. The evolution of distribution. IEEE Power Energy Mag
2009:63–8. March/April.
[2] Sekine Y, Akimoto Y, Kunugi M, Fukui C, Fukui S. Fault diagnosis of power
systems. In: Proceedings of the IEEE. May 1992. p. 673–83.
[3] Fukui C, Kawakami J. An expert system for fault section estimation using
information from protective relays and circuit breakers. IEEE Trans Power
Deliv 1986;PWRD-1(4):83–90.
[4] Protopapas CA, Psaltiras KP, Machias AV. An expert system for substation fault
diagnosis and alarm processing. IEEE Trans Power Deliv 1991;6(2):648–55.
[5] Yang HT, Chang WY, Huang CL. On-line fault diagnosis of power substation
using connectionist expert system. IEEE Trans Power Syst 1995;10(1):323–31.
[6] Lee HJ, Ahn BS, Park YM. A fault diagnosis expert system for distribution
substations. IEEE Trans Power Deliv 2000;15(1):92–7.
[7] Yang HT, Chang WY, Huang CL. Power system distributed on-line fault section
estimation using decision tree based neural nets approach. IEEE Trans Power
Deliv 1995;10(1):540–6.
805
[8] Junior GC, Rolim JG, Zürn HH. Application of neural-network modules to
electric power system fault section estimation. IEEE Trans Power Deliv
2004;19(3):1034–41.
[9] Silva APA, Insfran AHF, Silveira PMD, Lambert-torres G. Neural networks for
fault location in substations. IEEE Trans Power Deliv 1996;11(1):234–9.
[10] Bi T, Yan Z, Wen F, Ni Y, Shen CM, Wu FF, et al. On-line fault section estimation
in power systems with radial basis function neural network. Electr Power
Energy Syst 2002;24(4):321–8.
[11] Chang CS, Chen JM, Srinivasan D, Wen FS, Liew AC. Fuzzy logic approach in
power system fault section identification. IEE Proc-Gen Trans Distrib
1997;144(5):406–14.
[12] Monsef H, Ranjbar AM, Jadid S. Fuzzy rule-based expert system for power
system fault diagnosis. IEE Proc-Gen Trans Distrib 1997;144(2):186–92.
[13] Souza JCS, Meza EM, Schilling MT, Filho MBC. Alarm processing in electrical
power systems through a neuro-fuzzy approach. IEEE Trans Power Deliv
2004;19(2):537–44.
[14] Chen WH, Liu CW, Tsai MS. On-line fault diagnosis of distribution substations
using hybrid cause-effect network and fuzzy rule-based method. IEEE Trans
Power Deliv 2000;15(2):710–7.
[15] Chin HC. Fault section diagnosis of power system using fuzzy logic. IEEE Trans
Power Syst 2003;18(1):245–50.
[16] Chen WH. Fault section estimation using fuzzy matrix-based reasoning
methods. IEEE Trans Power Deliv 2011;26(1):205–13.
[17] Luo X, Kezunovic M. Implementing fuzzy reasoning petri-nets for fault section
estimation. IEEE Trans Power Deliv 2008;23(2):676–85.
[18] Xiong G, Shi D, Zhu L, Duan X. A new approach to fault diagnosis of power
systems using fuzzy reasoning spiking neural P systems. Math Probl Eng
2013;2013:1–3 [Article ID 815352].
[19] Lo KL, Ng HS, Grant DM, Trecat J. Extended petri net models for fault diagnosis
for substation automation. IEE Proc-Gener Trans Distrib 1999:229–34. May.
[20] Huang YC, Huang CM, Liao CC, Chen JF, Yang HT. A new intelligent fast petri-
net model for fault section estimation of distribution systems. In: Proceedings
PowerCon international conference power system technology. December
2000. p. 217–22.
[21] Huang S-J. Application of immune-based optimization method for fault-
section estimation in a distribution system. IEEE Trans Power Deliv 2002;
17(3):779–84.
[22] Leão FB, Pereira RAF, Mantovani JRS. Fault section estimation in electric power
systems using an optimization immune algorithm. Electr Power Syst Res
2010;80(11):1341–52.
[23] Wen FS, Chang CS. A new approach to fault diagnosis in electrical distribution
networks using a genetic algorithm. Artif Intell Eng 1998;12(1–2):69–80
[January–April].
[24] Wen FS, Han Z. Fault section estimation in power systems using a genetic
algorithm. Electr Power Syst Res 1995;34:165–72.
[25] Leão FB, Pereira RAF, Mantovani JRS. Fault section estimation in automated
distribution substations. IEEE power & energy society general meeting. Calgary
(AB): IEEE; 2009. July 26–30.
[26] Wen FS, Chang CS. Probabilistic approach for fault-section estimation in power
systems based on a refined genetic algorithm. IEE Proc-Gen Trans Distrib
1997;144(2):160–8.
[27] Leão FB, Silva LGW, Mantovani JRS. Defects location in components of power
systems through a dedicated genetic algorithm. Transmission and distribution
IEEE/PES/T&D LATIN AMERICA. São Paulo: IEEE; 2004.
[28] Huang S-J, Liu X-Z. Application of artificial bee colony-based optimization for
fault section estimation in power systems. Int J Electr Power Energy Syst
2013;44:210–8.
[29] Huang S-J, Liu X-Z. Fault section estimation in distribution systems using
biogeography-based optimization approaches. Int Trans Electr Energy Syst
2013. http://dx.doi.org/10.1002/etep.1831.
[30] Huang S-J, Liu X-Z, Su W-F, Ou T-C. Application of enhanced honey-bee mating
optimization algorithm to fault section estimation in power systems. IEEE
Trans Power Deliv 2013;28(3):1944–51.
[31] Peng Y, Reggia JA. Abductive inference models for diagnostic problem –
solving. New York: Springer-Verlag; 1990. p. 98.
[32] Anderson PM. Power system protection. New York: IEEE Press Series on Power
Engineering, McGraw-Hill; 1999.
[33] IEEE. IEEE Std C37.2. In: IEEE standard for electrical power system device
function numbers, acronyms, and contact designations. 2008. 46p.