Electric Power Systems Research 110 (2014) 122132

Contents lists available at ScienceDirect

Electric Power Systems Research

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

On-line coordination of directional overcurrent relays: Performance

evaluation among optimization algorithms

Meng Yen Shih , Arturo Conde Enrquez 1 , Luis Martn Torres Trevino
Universidad Autnoma de Nuevo Len, UANL, FIME, Av. Universidad S/N Ciudad Universitaria, San Nicols de los Garza, N.L. C.P. 66451, Mexico

a r t i c l e

i n f o

Article history:
Received 1 November 2013
Received in revised form 7 January 2014
Accepted 16 January 2014
Available online 8 February 2014
Keywords:
Ant colony algorithm
Differential evolution algorithm
Dynamic network topologies
Genetic algorithm
On-line
Overcurrent relay coordination

a b s t r a c t
The coordination of directional overcurrent relays is most commonly studied based on a xed network
topology in a mesh electrical system. However, in real life, the network topology and the operation of
elements of sub-transmission and distribution systems are constantly changing. In addition, modern
electrical systems tend to operate near boundary conditions to satisfy the clients. As a result, the directional overcurrent relays (DOCRs) lose speed and sensitivity. Therefore, the principal objective of this
paper is to perform on-line coordination of the directional overcurrent relays in order to enhance speed
and sensitivity. The secondary objective is to formulate the differential evolution algorithm and compare
it with the genetic algorithm and ant colony algorithm. The objectives were accomplished by developing
an on-line algorithm that worked with the optimization algorithms.
2014 Elsevier B.V. All rights reserved.

1. Introduction
The fundamental objective of a protective relay is to detect
and isolate the faulted element as soon as possible, so that the
impact to the rest of the system is minimized, leaving intact
as many non-faulted elements as possible. As different protections are used in different voltage levels of the power system, the
directional overcurrent relays (DOCRs) are widely implemented in
sub-transmission and distribution systems due to their competing costs. The propose of coordinating the DOCRs is to encounter
settings that minimize the operation time for faults within the protective zone, while at the same time offering pre-specied timed
backup for relays that are in the adjacent zones. Two settings
(degrees of freedom) were considered: dial which is also known
as time-dial setting and k which is the security factor that multiplies with the load current in order to obtain the pickup current
setting.
As the number of clients increased, the size of sub-transmission
and distribution systems grew. More elements (fault limiters, distributed generators, batteries, etc.) are introduced into the network
which made the system become more reliable but at the same time
more complicated. Loads of modern electrical system tends to grow

Corresponding author. Tel.: +52 8115015626.

E-mail addresses: sonofafriend@gmail.com (M.Y. Shih), con de@yahoo.com

(A. Conde Enrquez), luis.torres.ciidit@gmail.com (L.M. Torres Trevino).

1
Tel.: +52 8112908037.
0378-7796/$ see front matter 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.epsr.2014.01.013

exponentially with no accurate predictable data. Therefore, the system is forced to operate near boundary conditions to satisfy the
new clients. The load currents increased signicantly, but did not
increase the fault currents at all. To avoid improper relay operations, the possible increase of load currents is accounted for the
coordination purpose using the security factor k. When these ideas
are applied, the accounted pickup currents are far bigger than they
really need to be. In addition, the high value of pickup currents
is held xed throughout the coordination study. This made the
obtained coordination settings become stiff and rigorous which
also lead to greater relay operation time and lower relay sensitivity.
Over the past decades, manual coordination of DOCRs has
been the most common practice performed by protection experts.
However, because of its complexity and nonlinearity, manual coordination has been formulated as an optimization problem. Several
optimization methods (deterministic, heuristic, hybrid) have been
proposed to solve this problem. Such as the use of linear programming (LP) [13], particle swarm optimization (PSO) [4], genetic
algorithm (GA) [58], hybrid GA and mixed PSO [9,10]. These algorithms have been used in the studies of coordinating the DOCRs off
line for xed network topologies and also for considering different
network topologies [9,11,12].
The DOCRs have constantly improved throughout the years from
electro-mechanic relays to digital relays. The digital relays emulate the operating principles of the electro-mechanic relays with
new features such as data registration, settings groups, the use of
no conventional curves and the ability to use continuous parameter settings. These new features allow re-setting of relays (possible

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

Fig. 1. On-line coordination ow diagram.

on-line coordination). This idea has been implemented in radial

distribution lines to solve specic problem but has not yet been
implemented in interconnected system due to its complexity. Manufacturers such as SEL and SIEMENS have dedicated their effort in
manufacturing these relays.
When DOCRs are coordinated off line according to the reported
literatures, there are disadvantages such as: high operation time of
relays, low sensitivity of relays and limited or no ability to withstand contingencies.
On the other hand, the on-line coordination proposed in this
paper is to re-coordinate all DOCRs for every change of network
topology and element operation. The advantages by doing so are
minimum relay operation time, increase of sensitivity, and the ability to withstand another unknown contingency. Moreover, the idea
is to coordinate DOCRs on-line, which as a result enhances in meeting the fundamental requirements [13].
First, the developed on-line algorithm updates data from the
latest changes of the system. Then it computes load ow and fault
analysis in order to obtain input data for the optimization algorithms. In this paper, ant colony optimization (ACO), differential
evolution (DE) and GA were selected to work with the on-line
algorithm. ACO has lately been used to study reactive power ow
planning [14], power ow economic dispatch [15] and power generation scheduling [16]. ACO is a powerful tool in solving complex
problems in different areas [17]. The DE has not yet been used in any
power analysis, but it is reported to be notably efcient in different
areas [18]. The advantage of ACO compared with GA is the role of
global memory played by pheromone matrix which improves solution convergence, whereas the advantage of DE compared with GA
is the fast and simple execution logic of the algorithm. Hence, the
idea to formulate the coordination problem using DE is original. GA,
which is widely known in the coordination area, is used as the comparison reference. In addition, GA is improved and selected because
of its simplicity, robustness and easy implementation.

system is assumed to have been manufactured; the hardware

requires only the installation of an appropriate on-line algorithm.
The GA, ACO and DE in Fig. 1 are for comparison purpose only.
The detailed description of the algorithm is as follows. First,
the systems data are updated according to element and network changes. Then, the Ybus is constructed or modied from
the obtained data using the Incident method and the inverse of
Inspection method. Next, both lists of Relay Names and Coordination Pairs are generated automatically. After that, the load
ow analysis is run using the Newton Raphson or another method.
Then, the Zbus is constructed or modied by the Block construction method and Partial Inversion Motto. Finally, fault analysis is
run using Thevenins method or Symmetrical Components [19].
When all of the above are done, the algorithm will have dened
the coordination pairs and computed the maximum load currents
and fault currents (3-ph primary, 3-ph backup, 2-ph backup, 1ph) Iload , coordinationpairs , Isc3 primary , Isc3 backup , Isc2 backup of each
relay for the optimization algorithms of the original network topology. Therefore, the on-line algorithm is actually an implementation
of existing methods.
However, to ensure that the relay settings obtained from the
posterior coordination algorithm are suitable for at least one element output without coordination loss, the maximum load and
fault currents must be computed according to the different n 1
contingency topologies. All elements are taken out one at a time
and the simulation is repeatedly performed for the different n 1
contingency topologies. Only the maximum load and fault currents
are stored as data for coordination use.
Finally, this algorithm performs a sensitivity ltration before
passing the data to the optimization algorithms which coordinate
the overcurrent relays. This step ensures that all coordination pairs
can be coordinated. The coordination pairs that do not satisfy the
requirement of sensitivity analysis will be omitted from the coordination process. Thus, the optimization algorithms will not spend
extra time to nd settings for these insensitive relay pairs, which
have no settings that will suit them.
2.2. Formulation of coordination problem
2.2.1. Objective function
The purpose of formulating the coordination of DOCRs as an
optimization problem is to minimize the principal and backup
relay operation time while maintaining selectivity. It is of great
importance to establish a good objective function that evaluates
the tness of the settings because this is the key to encounter optimum solutions using optimization algorithms. The tness is given
in the following equation:
Fitness =

2. Formulation of the on-line coordination

As presented in the previous section, the on-line coordination of
DOCRs enhances meeting the fundamental requirements of sensitivity, selectivity, reliability and speed [13]. Therefore, the on-line
coordination algorithm and the formulation of coordination problem are presented sequentially.
2.1. On-line coordination
The ow diagram of on-line coordination of DOCRs is presented
in Fig. 1.
The on-line algorithm is a very important segment of this paper,
as presented in Fig. 1. It consists of collecting data of the latest
element and network changes, from which input data for posterior relay coordination are computed. The online update hardware

123

 NV 
NCP



 NCP


NCP
t
a=1 primarya

NCP

 NCP

t
b=1 primaryb

NCP


ECTIL

(1)

L=1

where , and are factors that increase or decrease the inuence

of each sub-objective function and will do for any other system,
NV is the number of violation of coordination constraints, NCP is
the number of coordination pairs, tprimarya is the primary operation
time of relay a, tprimaryb is the backup operation time of relay b, and
ECTIL is the CTI error of Lth coordination pair.
2.2.2. Primary and backup relay constraints
To coordinate the relays, there must be a time difference
between the primary and backup relays. This time difference is
called coordination time interval (CTI). In this way, whenever the

124

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

Table 1
IEEE relay parameters.

The population is shown in the following equation:

Norm

Curve type

IEEE

Moderate inverse
Very inverse
Extremely inverse

0.0515
19.61
28.2

0.114
0.491
0.1267

0.02
2
2

primary relay fails to extinct a fault, the backup relay enters and
tries to extinct the fault after a pre-specied delay. It is normally
between 0.2 and 0.5 s, but the value of 0.3 s is used in this paper.
The coordination constraint for every coordination pair is given in
the following equation:
CTIL = tbackup tprimary

(2)

where CTIL is the CTI of the Lth coordination pair, tprimary is the
primary operation time, and tbackup is the backup operation time.
There is also a range for each relay setting, from which feasible
solutions are encountered. This is given in the following equations:
dialmin dial dialmax

(3)

Ipickupmin Ipickup Ipickupmax

(4)

where dial is the relay dial setting in the range between the maximum dialmax and minimum dialmin setting. And Ipickup is the relay
pickup current in the range between the maximum Ipickupmax and
minimum Ipickupmin setting.
2.2.3. Relay characteristic curve
The relays function according to the relay characteristic curve
(inverse time curve). This inverse time curve operates with less
time as fault magnitude raises and more time as fault magnitude
drops. The IEEE standard norm C37.112-1996 is used in this paper
as given in the following equation:

t=

(Isc3 max /Ipickup )n 1

+ B dial

(5)

where t is the relay operation time, Isc3 max is the maximum 3-ph
short-circuit current, Ipickup is the pickup current, dial is the relay
dial setting, and A, B and n are constants of the IEEE standard.
The IEEE constants of DOCRs are shown in Table 1. These
constants correspond to the conventional curves: moderate inverse
(MI), very inverse (VI) and extremely inverse (EI). The IEEE VI curve
is used in this paper, but other curves such as the IEC standard can
also be used.
3. Genetic algorithm
The GA is a simple algorithm that has gained popularity over the
past decades. It performs a heuristic search based on evolutionary
ideas of natural selection of genes. A population of search space
that contains sets of feasible solutions (chromosomes) is created.
The decision variables (dial, k) are encoded as genes into the chromosome strings. Then, genes are evaluated, penalized, ranked and
selected according to their tness value of the objective function.
Subsequently, genetic evolution principles (crossover, mutation)
are applied, and the new population is formed. The entire process
is repeated until the stopping criterion is met.
The population size indicates the number of chromosomes in the
population (in one generation). If there are too few chromosomes,
the algorithm will have few possibilities to perform crossover and
only a small part of the search space is explored. On the other hand,
if there are too many chromosomes, the algorithm will explore
more variety of feasible solutions, but the execution time is excessively increased.

P=

dial(1,1)

dial(1,NR)

k(1,NR+1)

k(1,NR2)

..
.

..

..
.

..
.

..

..
.

dial(NC,1)

dial(NC,NR)

k(NC,NR+1)

(6)

k(NC,NR2)

where NC is the number of chromosomes and NR number of relays.

The population size is the number of chromosomes multiplied by
the number of relays times two (NC, NR*2). Although chromosomes
can be encoded as binary strings, both dial and k parameters are
encoded as continuous integer strings in this paper.
Because GA is a well known algorithm in the coordination
area, penalization, ranking and genetic operators such as selection,
reproduction, mutation are not presented.
3.1. Steps of protection coordination using GA
Detailed steps of the GA performed in this paper are presented
as followed:
1. Randomly generate the initial population of n chromosomes, in
which each gene is a possible solution to the problem. The genes
must be found within the specied ranges introduced in (3) and
(4).
2. Compute the primary and backup time of each relay according
to each chromosome.
3. Evaluate the tness f(x) of each chromosome x in the population.
4. Create a new population in each iteration:
a. Selection: select parent chromosomes from the population
according to their tness by performing roulette wheel, rank and
elitism.
b. Reproduction: use a crossover probability to crossover the parents to form a new child or children. Perform non-uniform
crossover.
c. Mutation: use a mutation percentage to mutate the genes of the
chromosomes. Perform non-uniform mutation at the rst stages
of the algorithm, and intelligent mutation at the posterior stages.
d. New population: place the results of reproduction, mutation
(new children) and elitism in the new population.
5. Execute the algorithm again using the new population.
6. Terminate the algorithm if stopping criteria is met, otherwise,
repeat steps 25.
The enhancement of GA here is the addition of intelligent mutation at posterior stages of the algorithm. It consists of detecting
coordination pairs that are not coordinated, and mutating only
the settings of those relays while leaving the coordinated pairs
untouched. As intelligent mutation was incorporated into the
algorithm, numerous chromosomes and iterations are no longer
necessary. This was a notably successful improvement of this particular coordination problem because the objective function results
in all coordination pairs being coordinated most of the time. The
solution reported here may not be the global optimum, but it is
excusable because coordination is performed on-line.
4. Ant-colony optimization
The ACO algorithm is part of the swarm intelligence computing,
it is a meta-heuristic optimization aiming to search for an optimal path in a graph, based on the behavior of ants seeking a path
between their home colony and food source. Real ants can nd the
shortest path between their home colony and the food source using
only information of chemical deposits called pheromone trails. Ants
walk on pheromone trails that were previously deposited by other
ants, while at the same time depositing pheromone trails on the

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

ground for future ants to follow. The pheromone trails are volatile
over time. When an ant encounters more than one pheromone trail,
it selects the most intense trail to follow according to the transition rule. The intensity of a pheromone trail deposited on a route
depends on the number of ants traveled on it and the amount of
food found through it. This biological behavior inspired the ACO
algorithm, in which a set of articial ant agents adopt the behavior of real ants in solving a problem by exchanging information via
pheromones deposited on a graph.
4.1. Nomenclature
Ant agents are articial ants that build solutions to an optimization problem and exchange information on their quality through a
communication scheme that is reminiscent of the one adopted by
real ants [17].
The AS-graph (search space) is a matrix that contains discrete
settings (states) of the control variables (stages) [14]. In other
words, this graph or matrix contains the set of feasible solutions
to the optimization problem, which will be explored by the ant
agents. Another matrix, which is called the pheromone matrix, is
created to represent the attractiveness of each discrete setting.
The pheromone matrix contains information about the chemical
pheromone deposited by ants. The matrix shows the pheromone
intensity of each discrete setting and describes the attractiveness
of every possible route to the solution. The more intense a setting
is, the more probability it has to be chosen by an ant agent as part
of the solution.
The transition rule is the probabilistic and stochastic mechanism
that ant agents use to evaluate the pheromone intensity to decide
which point is the most attractive to visit next.
The pheromone update is the process where pheromone intensities are increased or decreased according to the evaluated results,
regardless of whether the settings lead to good or bad solutions.
This is achieved by decreasing the pheromone values through
pheromone evaporation and increasing the pheromone levels by
depositing more pheromone if it is a set of good solution.
The algorithm is started with many sets of solutions (states);
together, all states form the AS-graph search space. This AS-graph
is held constant throughout the whole searching process; therefore,
it does not change from iteration to iteration. On the other hand,
pheromone matrix is a representation of attractiveness of each discrete setting (edge) that changes in all iterations and in every ant
tour. The settings that t well will consequently lead the ant agents
to deposit more and more pheromone until all ants converge on
this route (set of settings). The optimal solution is found using this
method. The whole process is repeated until the stopping criterion
is met.
The AS-graph size indicates the number of states in the ASgraph. If there are too few states, the algorithm will have fewer
possibilities to obtain the optimal solution to the problem, and only
a small part of the search space is explored. On the other hand, if
there are too many states, the algorithm increases the possibility to
encounter the optimal solution, but it drastically slows down the
entire process.
4.2. Formulation of ant colony optimization
4.2.1. AS-graph
The AS-graph is a (m, n*NR) matrix, where m represents the
number of states, n the number of stages and NR the number of
relays. For example, if a system has 5 relays with 2 degrees of freedom (dial and k), then the size of the AS-graph for 20 states is (20,
10). There are a total of 200 discrete settings of relays. As more
degrees of freedom are added to the AS-graph, the size of AS-graph
increases.

125

The AS-graph is shown in the following equation:

dial(1,1)

dial(1,NR)

k(1,NR+1)

k(1,NR2)

..
.

..

..
.

..
.

..

..
.

dial(m,1)

dial(m,NR)

k(m,NR+1)

AS =

(7)

k(m,NR2)

To create the AS-graph matrix, necessary data such as upper

and lower limits and steps of the control variables are required. For
example, if a system has 2 relays with 2 degrees of freedom (dial, k)
which the range of dial is [0.5, 1.4] in steps of 0.1 and the range of
k is [1.4, 1.6] in steps of 0.05, then the AS-graph is a (10, 4) matrix.
Note that there are 10 values in the range of dial setting including
upper and lower limits, but there are only 5 values in the range of k
setting including upper and lower limits. Under this circumstance,
complete the remainder of the matrix of the k settings by repeating
the upper limit of k to have sequential order.
4.2.2. Pheromone matrix
The pheromone matrix (m, n) is constructed according to the
size of AS-graph, where m is the number of states and n is the number of stages. This matrix is initialized as shown in the following
equation:
(m, n) = 0 (m, n) = max

(8)

where  max is the maximum pheromone trail, which is given in the

following equation:
max =

1
fgbest

(9)

where fgbest is the global best solution over all the past iterations
and is an empirical value that best suits in the range [0.88, 0.99]
[11]. In the case of initializing the pheromone matrix, fgbest is an
initial estimation of the best solution.
In this paper, the pheromone matrix was rst constructed with
all equal edges as presented in (8). However, as in the construction of the AS-graph, the smallest settings of relays occupy the rst
rows of the AS-graph. These settings are the ideal settings to coordinate with, because they will minimize operation time. Hence, after
the pheromone matrix is constructed, the pheromones of the rst
rows of this matrix are increased to help the algorithm nding the
best time-operation settings in less time. Thus, one decides how
many rows to change and in what amount. One shall not change
too many rows and the changes should not be large because this
will signicantly affect the performance of the exploration of ants.
4.2.3. Transition rule
When ant-j is at the r-state of the (i 1)-stage, it will choose the
s-state of the (i)-stage as the next visit according to the transition
rule shown in the following equation:
p(r, s) =

(r, s)

(r, l)

s, l Nr

(10)

where Nr is a memory tabu list of ant-j that denes the set of points
to be visited when it is at point r. The pheromone of the next
possible visit of (i)-stage currently under evaluation is (r,s) and
l (r,
l) is the pheromone sum of the entire column of the (i)-stage under
evaluation.
4.2.4. Pheromone update
4.2.4.1. Local pheromone update. The pheromone trail on each edge
of the ant-j tour is updated immediately as the ant-j agent nishes
its tour. This is given in the following equation:
(r, s) = (r, s) +  k (r, s)

(11)

126

 k (r, s) =

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

1
Q f

(12)

where is the persistence of the pheromone trail in the range

0 < < 1, (1 ) represents the pheromone trail evaporation and
 k (r, s) is the amount of pheromone that ant-j puts on edge (r,s).
The desirability of the edge (r,s) is represented by  j , such as a
shorter distance, a better performance, and in this case a smaller
operation time. The objective function evaluation of the settings
of ant-j tour is represented by f and Q is a positive constant. It is
observed from (12) that as constant Q increases, the amount of
pheromone deposited by an ant decreases. Here, Q was chosen to
be 100.
4.2.4.2. Global pheromone update. After all ant agents have completed their tours in the iteration, the primary and backup operation
times are computed. The objective function is evaluated for each
ant tour and all pheromone edges (r,s) of the best ant tour (the ant
tour with best tness value) are updated according to the following
equation:
(r, s) = (r, s) +

R
fbest

r, s Jbest

(13)

where fbest is the best solution of this iteration, R is a positive conj

stant and Jbest is the location list of the best ant tour that records
the state of each stage when ant-j moves from one stage to another.
It is observed from (13) that as constant R increases, the amount of
pheromone deposited by an ant increases. Here, R was chosen to
be 5.
In this work, the global pheromone update was not performed
as described in the previous paragraph because updating only the
best ant tour leads to premature convergence. Therefore, the global
pheromone update was performed by updating a percentage of the
best ant tours. For example, 30% of the ant tours that ranked the
best were updated.
Empirical tests have shown that the ACO algorithm converges
faster when both Q and R are arbitrarily large numbers and
almost equal to one another. Studies in the literatures might use
Q = R = 1,000,000 for other problems, but such large constant is
not suitable to use here because the algorithm converges approximately after 30 iterations and leaves many coordination pairs
uncoordinated. Thus they were chosen empirically to be Q = 100
and R = 5 but can work for any network.
4.2.5. Intelligent exploration
Intelligent exploration of the AS-graph consists of exploring
the setting (dial) of specic relays after a pre-specied amount of
iterations. This process was programmed to help coordinating all
coordination pairs. The dial setting was chosen because it has more
inuence on the relay operation time.
First, the coordination pairs that are not coordinated are
detected. Then, a pair is selected to start with. Next, the setting
of this specic relay (primary) is obtained from the best ant tour
and used as the upper limit. Afterwards, (13) is applied again but
with modication as given in the following equation.
(r, s) = (ran1, s) +

R
fbest

r, s Jbest

The setting of the specic relay (backup) is obtained from the

best ant tour of the same coordination pair that was selected previously and is used as the lower limit. Then, (13) is applied again
but with the modication as given in the following equation.
(r, s) = (ran2, s) +

R
fbest

r, s Jbest

(15)

where ran2 is a random number selected from the interval [r:end of

state]. Then, pheromone is deposited on this edge. Note that r represents the lower limit (state) and s represents the specic relay
(stage). Depositing pheromone on this specic edge will lead the
ant agents to explore the corresponding setting from the AS-graph.
Because it has a lower limit, the newly explored edge will correspond to a bigger setting in the AS-graph, which increases the
backup operation time.
If all coordination pairs were coordinated, the coordination pairs
that have a greater CTI than the pre-specied value are detected and
reduced. A coordination pair is chosen to start the process and the
settings of both relays (primary and backup) are obtained from the
best ant tour and used as the upper limit. Then, (13) is applied again
for both the primary and backup relay. Thus, pheromone trails are
deposited on these specic edges (both primary and backup). This
leads ant agents to explore the corresponding settings from the ASgraph. Because they have an upper limit, the newly explored edges
will correspond to a smaller setting in the AS-graph, which reduces
both primary and backup operation time.
4.3. Steps of protection coordination using ACO
Detailed steps of the ACO algorithm performed in this paper are
presented as followed:
1. Generate the AS-graph (search space) that represents the discrete settings (states) of the control variables (stages). Each
discrete setting is a possible solution to the problem.
2. Generate the pheromone matrix (m,n) according to the size of
AS-graph, where n is the number of stages and m the number
of states.
3. Initialize the pheromone matrix (m,n) =  0 (m,n) =  max . Eqs.
(8) and (9), in this case fgbest is an initial estimation of best
solution.
4. Randomly place M ants on the states of the rst stage (i = 1).
5. Every ant must explore all stages sequentially; when ant-j is at
the r-state of the (i 1)-stage, it will choose the s-state of the
(i)-stage as the next visit according to the transition rule Eq.
(10).
6. Compute primary and backup time of each relay after ant-j
complete its tour (explores all stages sequentially).
7. Use Eq. (11) to update local pheromone value of (r,s)-trail made
by ant-j.
8. Evaluate the tness fk (x) of ant-j tour and save it to f(x).
9. Repeat steps 58 until all ants (M) have nished their tour.
10. Use Eq. (13) to update global pheromone value of (r,s)-trail
which belongs to the best ant tour (fbest ).
11. Terminate the algorithm if stopping criteria is met. Otherwise
continue steps 410.

(14)

where ran1 is a random number selected from the interval [1:r].

Then, pheromone is deposited on this edge. Note that r represents
the upper limit (state) and s represents the specic relay (stage).
Depositing pheromone on this specic edge will lead the ant agents
to explore the corresponding setting from the AS-graph. Because
it has an upper limit, the newly explored edge will correspond
to a smaller setting in the AS-graph, which reduces the primary
operation time.

5. Differential evolution
The differential evolution algorithm (DE) is a part of the evolutionary algorithms; it is a metaheuristic search algorithm based on
natural selection of genes.
The DE algorithm operates using similar computational steps of
a standard evolutionary algorithm (EA). However, unlike traditional
evolutionary algorithms, the differential evolution algorithm perturbs the current generation population members with the scaled

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

127

for 20 individuals is (20,10). The initial population is shown in the

following equation:

Fig. 2. Main stages of the DE algorithm.

differences of randomly selected and distinct population members.

Therefore, no separate probability distribution has to be used for
generating the offspring. This characteristic made the algorithm
has fewer mathematical operations and lower execution time than
the other algorithms.
Main stages of the DE algorithm are presented in Fig. 2.
5.1. Nomenclature
In the differential evolution community, the individual trial
solutions (which constitute the population) are called parameter
vectors or genomes. Each parameter vector contains a set of possible
solution information.
In DE-literature, a parent vector from the current generation is
called target vector, a mutant vector obtained through the trigonometric mutation operation is known as donor vector and nally an
offspring formed by recombining the donor with the target vector
is called trial vector.
Crossover is a recombination (or reproduction). Sons or new
individuals are formed by the recombining the genes of certain
parameter vectors called parents or target vectors. The newly
formed individuals (sons) are called trial vectors.
Mutation is a sudden change or perturbation with a random
element. A mutant or a donor vector is formed for each target vector.
Selection is the process that maintains the population size constant over the subsequent generations by determining whether the
target or trial vector survives to the next generation.
Search space is the space of all feasible solutions. Each point in
the search space represents a feasible solution and each feasible
solution can be marked by its tness value for the problem. This
feasible solution is then a maximum or a minimum.
The algorithm is started with many sets of solutions, together
all the parameter vectors form a population. Solutions from one
population are used to form a new population. And by employing
the selection process, it guarantees the population to either get
better (with respect to the minimization of the objective function)
or to remain the same in tness status, but never deteriorates.
The entire process is repeated until some condition (reached the
number of maximum iterations, improvement of the best solution)
is met. This is called stopping criteria.
The population size (NP) indicates the number of parameter
vectors in the population (in one generation). If there are too few
parameter vectors, the algorithm will have notably few possibilities to perform crossover and only a small part of the search space is
explored. On the other hand, if there are too many parameter vectors, the algorithm will explore more variety of feasible solutions
but the execution time is increased.
5.2. Formulation of differential evolution
5.2.1. Initial population
Create an initial population where all genes of all parameter
vectors or individuals are initialized somewhere in their feasible
numerical range of the corresponding setting of overcurrent relay.
Each row is an individual and each column is a gene/variable/setting
of a relay. The population size is (NP, D*NR) where NP represents
the number of parameter vectors, D the number of control variables
and NR the number of relays. For example, if a system has 5 relays
with 2 degrees of freedom (dial and k), then the population size

dial(1,1)

dial(1,NR)

k(1,NR+1)

k(1,NR2)

..
.

..

..
.

..
.

..

..
.

dial(NP,1)

dial(NP,NR)

k(NP,NR+1)

P=

(16)

k(NP,NR2)

where NP is the number of parameter vectors and NR the number

of relays.
5.2.2. Trigonometric mutation
To implement this scheme, for each target vector, three distinct
vectors are randomly selected from the DE population. Suppose

for the ith target vector X i,G , the selected population members are

X r1,G , X r2,G , X r3,G . The indices r1 , r2 and r3 are mutually exclusive

integers randomly chosen from the range [1,NP], which are also
different from the index i, these indices are generated only once for
each mutant vector. Now, three weighting coefcients are formed
according to Eqs. (18), (19) and (20).

 
 

p = f ( X r1 ) + f ( X r2 ) + f ( X r3 )
(17)

|f ( X r1 )|
p

|f ( X r2 )|
p2 =
/
p1 =

p3 =

(18)

(19)

|f ( X r3 )|
p

(20)

where f() is the function to be minimized. Let  be the

trigonometric mutation rate in the interval (0,1). Then, the trigonometric mutation scheme may be expressed as the following
equations.


X r1 + X r2 + X r3


V i,G+1 =
+ (p2 p1 ) X r1 X r2
3



+ (p3 p2 ) X r2 X r3 + (p1 p3 )

( X r3 X r1 ) if rand [0, 1] 

V i,G+1 = X r1 + F( X r2 X r3 )

else

(21)

(22)

where V i is the donor vector and F is a scalar number that typically

lies in the interval [0.4, 1]. Both parameters  and F are selected as
0.5 and 0.8, respectively, in this paper.
5.2.3. Binomial crossover/recombination
To enhance the potential diversity of the population, a crossover
operation occurs after generating the donor vector through mutation. The donor vector exchanges its components with the target

vector X i,G under this operation to form the trial vector U i,G =
[u1,i,G , u2,i,G , u3,i,G , . . ., uD,i,G ].
Binomial crossover is performed on each of the D variables
whenever a randomly generated number between 0 and 1 is less
than or equal to the Cr value. In this case, the number of parameters
inherited from the donor has a (nearly) binomial distribution. The
scheme may be outlined as the following equation

uj,i,G =

vj,i,G if(randi,j [0, 1] Cr or j = jrand )

xj,i,G

(23)

otherwise

where randi,j [0,1] is an uniformly distributed random number,

which is called anew for each jth component of the ith parameter vector. jrand [1,2, . . ., D] is a randomly chosen index, which

128

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

ensures that U i,G receives at least one component form V i,G . It is

instantiated once for each vector per generation. The Cr parameter
is selected to be 0.5 in this paper.
5.2.4. Selection
To maintain the population size constant over subsequent
generations, the next step of the algorithm calls for selection to
determine whether the target or the trial vector survives to the
next generation, for example at G = G + 1. The selection operation is
presented in the following equation.

X i,G+1 =

U i,G

if f ( U i,G ) f ( X i,G )

X i,G

if f ( U i,G ) > f ( X i,G )

(24)

where f ( X ) is the objective function to be minimized. Therefore, if

the new trial vector yields an equal or lower value of the objective function, it replaces the corresponding target vector in the
next generation; otherwise, the target is retained in the population.
Hence, the population either improves (with respect to the minimization of the objective function) or remains the same in tness
status, but never deteriorates.
5.3. Steps of protection coordination using DE
1. Randomly generate the initial population of n parameter vectors,
where each gene is a possible solution to the problem.
2. Randomly select 3 mutually exclusive parameter vectors from
the current population for each target vector.
3. Trigonometric mutation:
a. Evaluate the tness f(x) of the 3 parameter vectors selected
for each target vector.
b. Calculate the 3 weighting coefcients according to Eq. (17).
c. Create the donor vector for each target vector according to
Eqs. (21) and (22).
4. Reproduction:
a. Perform binomial crossover according to Eq. (23) to form the
trial vectors.
5. Selection:
a. Evaluate the tness f(x) of both the target and the trial vectors.
b. Generate the new population of each iteration by performing
selection according to Eq. (24).
6. Evaluate the tness f(x) of the new population.
7. Execute the algorithm again using the new population.
8. Terminate the algorithm if stopping criteria is met. Otherwise,
continue steps 37.
6. Test system
The IEEE 14 and 30 bus test systems were used in this paper even
though they are transmission systems. This is because there are
no standard interconnected sub-transmission test systems, so the
transmission systems were used. But the voltajes were adjusted to
sub-transmission and distribution levels. The ultimate reason why
using the IEEE 14 and 30 bus test system is to exploit the interconnected electrical system in order to test the performance of the
different algorithms.
6.1. The IEEE 14 bus test system
The IEEE 14-bus test system was chosen to test and compare the
performance among the three algorithms. The system consists of 32
phase relays. The voltages were selected to be 34.5 kV for buses at
high voltage side of transformers and 22 kV for buses at low voltage
side of transformers. All relays are considered to have very inverse
time characteristic curve as is presented in Table 1.

In the present work, coordination of phase relays will be studied, instantaneous operation and ground relays will be omitted.
As instantaneous operation and ground protection are individual
problems apart from coordination, the study here will concentrate
on the complicated coordination of phase relays.
The relay names are not assigned as a number, as was done conventionally, but are instead generated automatically by the on-line
algorithm as a string of numbers. These relay names (string of numbers) consist of 3 digits. The rst digit is the name of the nearby
bus, the second digit is the name of the remote bus and the third
digit represents the number of parallel lines between two buses.
For example, the relays between buses 1 and 2 that are near bus 1
are assigned as [121] and [122], while the relays that are near bus
2 are assigned as [211] and [212] respectively.
The two lines between buses 1 and 2 have the same impedance
value. Therefore the relays [121], [122], [211] and [212] sense the
same amount of maximum load currents of 815 A, but due to the n1 contingency analysis described in section II-A, the maximum load
currents of these relays are 1849 A. The current values are based on
minimum load operation.
The CTI is proposed to be 0.3 s. Both dial and k are considered
continuous in GA and DE, their range are [0.5:2.0] and [1.4:1.6]
respectively. However, dial and k are considered as discrete values
in ACO, their range and step are [0.5:2.0], [1.4:1.6] and 0.05, 0.01
respectively. Other parameters of the ACO are Q = 100, R = 5. Parameters of the DE such as  , F and Cr are 0.5, 0.8 and 0.5, respectively.
The only stopping criterion is to stop when the algorithms have
reached the maximum iteration of 1000. Although each algorithm
has its own stopping criteria, they were disabled in order to be
comparable among them. The GA and DE are simulated with 500
chromosomes and the ACO is simulated with 500 ant agents.
The fault currents are calculated with the remote end opened.
This was done due to two considerations: to obtain the maximum
fault current that the relay senses and the very small probability for
the remote end relay to malfunction. Note that as the operation of
the elements or the network topology changes, load ow and fault
analysis must be computed again through on-line algorithm.

6.2. The IEEE 30 bus test system

The IEEE 30-bus test system was chosen to test and analyze the
dynamic operations in the system. The system consists of 72 relays
not considering DOCRs as protections for transformers. The voltages were selected to be 34.5 kV for buses at high voltage side of
transformers and 22 kV and 13.8 kV for buses at low voltage side
of transformers. All relays are considered to have very inverse time
characteristic curve as is presented in Table 1.
The two lines between buses 1 and 2 have the same impedance
value. Therefore the relays [121], [122], [211] and [212] sense the
same amount of maximum load currents of 929 A, but due to the
n 1 contingency analysis described in section II-A, the maximum
load currents of these relays are 2002 A. The current values are
based on minimum load operation.
The CTI is proposed to be 0.3 s. Both dial and k are considered
continuous in DE, their range are [0.5:4.0] and [1.4:1.6] respectively. Parameters of the DE such as  , F and Cr are 0.5, 0.8 and
0.5, respectively. The only stopping criterion is to stop when the
algorithm has reached the maximum iteration of 5000. Although
the contribution of results have minimum improvement after 1000
iterations, this maximum iteration number was selected due to the
reason that the most important issue in the on-line analysis at this
very moment is the result quality among the different cases and
not the speed. This is to proof whether or not the on-line coordination enhances the relay operation time and sensitivity with more
accuracy than using just 1000 iterations.

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

129

Now DE is simulated with 500 individuals due to the need of

speed. Besides, the number 500 is tested and empirically seems to
be a very suitable number of individuals in searching quality results
in a reasonable computational time.
The fault currents are calculated with the remote end opened.
7. Results and discussion
7.1. On-line coordination and comparison among GA, ACO and
DE on the IEEE 14 bus test system
The performance of each algorithm is tested in this section using
the IEEE 14-bus test system; advantages and disadvantages of each
algorithm are highlighted. Their performance is based on the execution time, result quality, robustness and convergence ability of
each algorithm.
The test system consists of 62 relay coordination pairs before
sensitivity lter. But this varies accordingly to the variation of load
demand.
The main objective of this section is to compare and choose
the most suitable algorithm for on-line coordination. The analysis
of relay operation time and sensitivity are temporarily omitted in
this section. Therefore, execution time of the algorithms and tness
values are the interested topics in this section.
7.1.1. Simulation cases
7.1.1.1. Case 1. The 14 bus test system is simulated using GA, ACO
and DE with the corresponding parameters presented in previous
section at maximum load. There are a total of 39 relay coordination
pairs after the sensitivity lter. All three algorithms were simulated
ten times. The convergence of each algorithm was averaged using
the best tness of each iteration of the ten simulations.
7.1.1.2. Case 2. The 14 bus test system is simulated again using
GA, ACO and DE with the corresponding parameters presented in
previous section at minimum load. There are a total of 48 relay
coordination pairs after the sensitivity lter. All three algorithms
were simulated ten times. The convergence of each algorithm was
averaged using the best tness of each iteration of the ten simulations.
7.1.1.3. Case 3. Even though other type of protection principle is
used for transformers, overcurrent relays are implemented here for
the purpose of coordination study in this section. This increased the
number of coordination pairs.
The 14 bus test system is simulated using GA, ACO and DE
with the corresponding parameters presented in previous section
at minimum load. There are a total of 102 relay coordination pairs
before sensitivity lter and 73 after. All three algorithms were simulated ten times. The convergence of each algorithm was averaged
using the best tness of each iteration of the ten simulations.

Fig. 3. Averaged execution time of the three algorithms for the four cases in seconds.

Fig. 4. Averaged tness value of the three algorithms for the four cases.

same initial settings as presented in the previous section. The DE

outperformed GA and ACO.
Case 4 evaluated the convergence ability of the three algorithms
by using an open range of dial as search space with no prior elimination. The DE outperformed GA and ACO. DE found good quality
results in reasonable amount of computational time.
These can be observed from the averaged results of the 4 cases
in Figs. 36.
The overall performances (result quality, robustness, and convergence) of the three algorithms of all four cases are averaged and
presented in Fig. 6.
From Fig. 3 it is observed that the execution time of GA increased
almost double in case 3 due to the increment of coordination pairs,
and almost 5 times in case 4 due to the increment of iterations compared to cases 1 and 2. On the other hand, both ACO and DE did not
increase signicantly in case 3 (are not affected signicantly) due

7.1.1.4. Case 4. In this case, the 14 bus test system is simulated at

minimum load. Overcurrent relays are not considered as protection
for transformers in this case. The new idea in this section is the open
range of dial parameter in the coordination process. So limits of dial,
iterations are modied for studying the behaviors of the algorithms.
The range of dial is now [0.5:10.0] and maximum iteration is now
5000.
7.1.2. Results and comparison
The purpose of cases 1 and 2 is to observe the speed and result
quality of the algorithms. The DE outperformed GA and ACO.
Case 3 evaluated the robustness of the three algorithms by
increasing the number of coordination pairs and maintaining the

Fig. 5. Averaged violations of the three algorithms for the four cases.

130

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

Fig. 6. Overall performance of GA, ACO and DE.

to the increment of coordination pairs. But in case 4, ACO increased

more than 5 times compared to cases 1 and 2, this is because the
open dial search increased the size of AS-graph, therefore decreased
the speed of the algorithm dramatically. The DE increased reasonably in case 4 compared to cases 1 and 2.
From Fig. 4 it is observed that as there are more coordination pairs in the system, the tness values increases. Case 3 > case
1 > case 2. This is reasonable because as there are more coordination pairs, there are more coordination constraints and therefore
more difcult to coordinate. GA and ACO seem not to be as robust
as DE in case 3 due to the increment of coordination pairs. In case
4, it is observed that DE tends to have much better convergence
ability than GA and ACO because its tness value approaches the
tness of case 2.
From Fig. 5 it is observed that as coordination pairs increases the
number of violation increases as well. This can be easily observed
from case 3 compared to cases 1 and 2. The number of violations
increased for all three algorithms because the number of coordination pairs was increased but the number of individuals/ant agents
or number of iterations was not increased.
From Fig. 6 it is observed that the overall performance of DE
(5.78) outstand GA (15.63) and ACO (11.15). DE tends to converge
to the minimum during the 5000 iterations, on the contrary, GA and
ACO tends to converge to a xed result and maintained throughout
the 5000 iterations.
7.2. On-line coordination using DE on the IEEE 30 bus test system
The DE was chosen to coordinate DOCRs on more cases of online coordination because it was demonstrated to be the best out
of the three algorithms. The IEEE 30 bus test system is selected to
study the behavior of the system under dynamic changes.
The test system consists of 141 relay coordination pairs before
sensitivity lter. But this varies accordingly to the dynamic changes
of the network topology.
The main objective of this section is to analyze the impact of
on-line coordination on the relay operation time and sensitivity.
Therefore, the execution time of the algorithm and the tness value
are not of interest in this section.

Fig. 7. Comparison between case 1 and case 2, tp, tb and sen.

7.2.1.3. Case 3. The 30 bus test system is simulated using DE with

the corresponding parameters presented in previous section at
maximum load referring to case 1 with increment of load on bus
30 of 15 MW. There are a total of 93 relay coordination pairs after
the sensitivity lter.
7.2.1.4. Case 4. The 30 bus test system is simulated using DE with
the corresponding parameters presented in previous section at
maximum load referring to case 1 with increment of line between
buses 1 and 3. There are a total of 151 relay coordination pairs
before the sensitivity lter and 106 after.
7.2.2. Results and discussion
Cases 1 and case 2 are compared, mainly their relay operation
time and sensitivity.
Case 3 demonstrates the inappropriate relay operations (inappropriate operation), if on-line coordination is not implemented.
Case 4 demonstrates the possible improvements of sensitivity,
if on-line coordination is implemented.
From Fig. 7 it is observed that the averaged relay operation
time (primary and backup) of case 2 are smaller than case 1. This
demonstrates that the on-line coordination does enhance the relay
operation time. Also the relays sensitivities are increased in case 2
as well.
From Fig. 8 it is observed that there is an increment of number of coordination pairs (NCP) in case 2. This is because of the
increase of relays sensitivities; hence this demonstrates that the
on-line coordination enhance sensitivity of relays. Another observation is the decrease of CTI in case 2; as the CTI narrows, the backup
relay operation becomes faster and better.
From Fig. 9 it is observed that after the re-coordination of the
system due to the increment of load at bus 30 as presented in
case 3, there is greater averaged relay operation time (primary and
backup), and smaller sensitivity. Apparently, the on-line coordination did no good, but made things worse. Let us take a closer view of

7.2.1. Simulation cases

7.2.1.1. Case 1. The 30 bus test system is simulated using DE with
the corresponding parameters presented in previous section at
maximum load. There are a total of 99 relay coordination pairs after
the sensitivity lter.
7.2.1.2. Case 2. The 30 bus test system is simulated using DE with
the corresponding parameters presented in previous section at
minimum load. There are a total of 107 relay coordination pairs
after the sensitivity lter.

Fig. 8. Comparison between case 1 and case 2, NCP and CTI.

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

131

Table 4
Sensitivity before the newly constructed line between buses 1 and 3.
Relay name

Sensitivity

131341

2.84

Table 5
Sensitivity after the newly constructed line between buses 1 and 3.
Relay name

Sensitivity

131341

3.60

7.3. Observations
Fig. 9. Comparison between case 1 and case 3, tp, tb and sen.

Table 2
Pickup current with contingency analysis before the increment of load at bus 30.
Relay name

Ipickup

27 30 1
30 27 1

358.16
358.16

Table 3
Load current without contingency analysis after the increment of load at bus 30.
Relay name

Iload

27 30 1
30 27 1

438.34
438.34

the effect of this increment of load by analyzing the load current. As

the increment was located at bus 30, we analyze the nearby relays,
since they are the most affected by this change.
From all the relays it was observed that these two relays [27 30
1] and [30 27 1] as shown in Tables 2 and 3 have an increase of
load current without contingency analysis after the increment of
load at bus 30 which is greater than the maximum pickup current
calculated considering contingency analysis before this increment
of load. In other words, as the load at bus 30 is increased, these two
relays will operate inappropriately because the actual load current
is greater than de pickup current.
From Fig. 10 it is observed that after a new line is constructed,
the averaged relay operation time (primary and backup) decreased.
The sensitivity increased. These changes may not be very notable
on the averaged value, but it has a great impact on the DOCR [131],
which functions as backup for relay [341]. These are relays that
originally exist in the system. They are presented as follows.
From Tables 4 and 5 it is observed that there is a signicant
increase of sensitivity for this coordination pair.

Fig. 10. Comparison between case 1 and case 4, tp, tb and sen.

All three algorithms showed the ability to encounter reasonable results for the tested system. But due to the need of speed
in on-line coordination, the number of individuals and ant agents
were reduced to only 500 for the complicated problem. This lead
to bad results for GA, but at the same time also showed the outstanding performance of DE. DE encountered the best results with
500 individuals out of the three algorithms, and is also the fastest.
DE showed the outstanding performance of execution time, result
quality, robustness and convergence ability. Therefore, DE is the
most suitable candidate for on-line coordination.
The on-line coordination surely increased the sensitivity of
DOCRs in some cases and at the same time reduced the sensitivity of DOCRs in other cases. Also, relay operation time are reduced
in some cases and increased in other cases. These are all due to the
operation of network. But despite of the reduction of sensitivity
or increment of relay operation time, these are due to the latest
network operation condition. Therefore, the system is prepared for
another unknown contingency. This made the system to stay strong
at all time.
The on-line coordination did brought advantages such as reducing relay operation time, increasing relay sensitivity and the ability
to withstand another unknown contingency.
8. Conclusions
This work proposed an on-line coordination for directional
overcurrent relays using ACO and DE algorithms. Solving relay
coordination for the test systems showed the feasibility of the algorithms. The aim is to nd an on-line solution (to send new settings
to each relay) to ensure the sensitivity and selectivity of relays for
the current topology, for the future changes of topology considering
an n 1 contingency, and for variable load demand current.
References
[1] A.J. Urdaneta, R. Nadira, L.G. Perez, Optimal coordination of directional overcurrent relays in interconnected power systems, IEEE Transactions on Power
Delivery (1988) 903911.
[2] A.J. Urdaneta, H. Restrepo, S. Marquez, J. Sanchez, Coordination of directional
overcurrent relay timing using linear programming, IEEE Transactions on
Power Delivery (1996) 122129.
[3] B. Chattopadhyay, M.S. Sachdev, T.S. Sidhu, An on-line relay coordination
algorithm for adaptive protection using linear programming technique, IEEE
Transactions on Power Delivery (1996) 165173.
[4] M.M. Mansour, S.F. Mekhamer, N.E.-S. El-Kharbawe, A modied particle swarm
optimizer for the coordination of directional overcurrent relays, IEEE Transactions on Power Delivery (2007) 14001410.
[5] R. Mohammadi, H.A. Abyaneh, F. Razavi, M. Al-Dabbagh, S.H.H. Sadeghi, Optimal relays coordination efcient method in interconnected power systems,
Journal of Electrical Engineering (2010) 7583.
[6] M. Singh, B.K. Panigrahi, Optimal overcurrent relay coordination in distribution
system, in: International Conference Energy, Automation, and Signal (ICEAS),
2011, pp. 16.
[7] D.K. Singh, S. Gupta, Optimal coordination of directional overcurrent relays: a
genetic algorithm approach, in: IEEE Students Conference on Electrical, Electronics and Computer Science, 2012.

132

M.Y. Shih et al. / Electric Power Systems Research 110 (2014) 122132

[8] D. Uthitsunthorn, T. Kulworawanichpong, Optimal overcurrent relay coordination using genetic algorithms, in: International Conference on Advances in
Energy Engineering, 2010.
[9] A.S. Noghabi, J. Sadeh, H.R. Mashhadi, Considering different network topologies
in optimal overcurrent relay coordination using a hybrid GA, IEEE Transactions
on Power Delivery (2009) 18571863.
[10] A. Motie Girjandi, M. Pourfallah, Optimal coordination of overcurrent relays
by mixed genetic and particle swarm optimization algorithm and comparison
of both, in: International Conference on Signal, Image Processing and Applications, 2011.
[11] A.J. Urdaneta, L.G. Perez, H. Restrepo, Optimal coordination of directional overcurrent relays considering dynamic changes in the network topology, IEEE
Transactions on Power Delivery (1997) 14581464.
[12] A.S. Noghabi, J. Sadeh, H.R. Mashhadi, Optimal coordination of directional overcurrent relays considering different network topologies using interval linear
programming, EEE Transactions on Power Delivery (2010) 13451348.
[13] J. Lewis Blackburn, T.J. Domin, Protective Relaying, Principles and Applications,
3rd ed., 2006.
[14] K.Y. Lee, J.G. Vlachogiannis, Optimization of power systems based on ant colony
system algorithms: an overview, ISAP (2005) 2235.
[15] Y.-H. Hou, Y.-W. Wu, L.-J. Lu, X.-Y. Xiong, Generalized ant colony optimization
for economic dispatch of power systems, in: International Conference on Power
System Technology Proceedings, 2002.
[16] S.-J. Huang, Enhancement of hydroelectric generation scheduling using ant
colony system based optimization approaches, IEEE Transactions on Energy
Conversion (2001).

[17] M. Dorigo, M. Birattari, T. Stutzle, Ant colony optimization: articial ants as a

computational intelligence technique, IEEE Computational Intelligence Magazine (2006) 2839.
[18] S. Das, P.N. Suganthan, Differential evolution: a survey of the state of the art,
IEEE Transactions on the Evolutionary Computation (2011) 431.
[19] H. Saadat, Power System Analysis (McGraw-Hill Series in Electrical and Computer Engineering), 1999, ISBN 0-07-561634-3.
Meng Yen Shih received the B.Sc. degree in electrical engineering from Instituto
Tecnolgico de Chetumal (ITCH) in 2010, and he is pursuing the M.Sc. degree in
electrical engineering. His research interest is power system protection.
Arturo Conde Enrquez received the B.Sc. degree in mechanical and electric engineering from Universidad Veracruzana, Veracruz, Mxico, in 1993 and the M.Sc. and
Ph.D. degrees in electric engineering from de Universidad Autnoma de NuevoLen,
Nuevo Len, Mxico, in 1996 and 2002, respectively. Currently, he is a Professor of
the Universidad Autnoma de Nuevo Len and a member of the National Research
System of Mxico, Mxico.
received the B.Eng. degree in electronics with specialty
Luis Martn Torres Trevino
in digital systems and the M.Sc. from Universidad de San Luis Potos. Ph.D. degree
in articial intelligence from Instituto Tecnolgico de Estudios Superiores de Monterrey Campus Monterrey 2004. Currently, he is a Researcher and Professor of the
Universidad Autnoma de Nuevo Len, Mexico.