See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226762616

Lecture Notes in Computer Science

Conference Paper · July 2001

DOI: 10.1007/3-540-44719-9_9

CITATIONS READS
145 1,012

2 authors:

Carlos A. Coello Coello Gregorio Toscano Pulido

Center for Research and Advanced Studies of the Natio… Center for Research and Advanced Studies of the Natio…
809 PUBLICATIONS 61,311 CITATIONS 70 PUBLICATIONS 6,454 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Gregorio Toscano Pulido on 23 March 2015.

The user has requested enhancement of the downloaded file.

 A Micro-Genetic Algorithm for
Multiobjective Optimization

Carlos A. Coello Coello1 and Gregorio Toscano Pulido2

1
CINVESTAV-IPN
Depto. de Ingenierı́a Eléctrica
Sección de Computación
Av. Instituto Politécnico Nacional No. 2508
Col. San Pedro Zacatenco
México, D. F. 07300
ccoello@cs.cinvestav.mx
2
Maestrı́a en Inteligencia Artificial
LANIA-Universidad Veracruzana
Xalapa, Veracruz, México 91090
gtoscano@mia.uv.mx

Abstract. In this paper, we propose a multiobjective optimization ap-

proach based on a micro genetic algorithm (micro-GA) which is a genetic
algorithm with a very small population (four individuals were used in our
experiment) and a reinitialization process. We use three forms of elitism
and a memory to generate the initial population of the micro-GA. Our
approach is tested with several standard functions found in the special-
ized literature. The results obtained are very encouraging, since they
show that this simple approach can produce an important portion of the
Pareto front at a very low computational cost.

1 Introduction

A considerable number of evolutionary multiobjective optimization techniques

have been proposed in the past [3,20]. However, until recently, little emphasis
had been placed on efficiency issues. It is well known that the two main processes
that consume most of the running time of an EMOO algorithm are: the ranking
of the population (i.e., the comparisons required to determine non-dominance)
and the mechanism to keep diversity (evolutionary algorithms tend to converge
to a single solution because of stochastic noise; it is therefore necessary to avoid
this with an additional mechanism).
Recent research has shown ways of improving the efficiency of an EMOO
technique (see for example [14,6]). The main emphasis has been on using an
external file that stores nondominated vectors found during the evolutionary
process which are reinserted later in the population (this can be seen as a form
of elitism in the context of multiobjective optimization [10,17,22]).
Following the same line of thought of this current research, we decided to
develop an approach in which we would use a GA with a very small population
size and a reinitialization process (the so-called micro-GA) combined with an

E. Zitzler et al. (Eds.): EMO 2001, LNCS 1993, pp. 126–140, 2001.
c Springer-Verlag Berlin Heidelberg 2001
 A Micro-Genetic Algorithm for Multiobjective Optimization 127

external file to store nondominated vectors previously found. Additionally, we

decided to include an efficient mechanism to keep diversity (similar to the adap-
tive grid method of Knowles & Corne [14]). Our motivation was to show that a
micro-GA carefully designed is sufficient to generate the Pareto front of a multi-
objective optimization problem. Such approach not only reduces the amount of
comparisons required to generate the Pareto front (with respect to traditional
EMOO approaches based on Pareto ranking), but also allows us to control the
amount of points that we wish to obtain from the Pareto front (such amount is
in fact a parameter of our algorithm).

2 Related Work

The term micro-genetic algorithm (micro-GA) refers to a small-population ge-

netic algorithm with reinitialization. The idea was suggested by some theoretical
results obtained by Goldberg [8], according to which a population size of 3 was
sufficient to converge, regardless of the chromosomic length. The process sug-
gested by Goldberg was to start with a small randomly generated population,
then apply to it the genetic operators until reaching nominal convergence (e.g.,
when all the individuals have their genotypes either identical or very similar),
and then to generate a new population by transferring the best individuals of
the converged population to the new one. The remaining individuals would be
randomly generated.
The first to report an implementation of a micro-GA was Krishnakumar [15],
who used a population size of 5, a crossover rate of 1 and a mutation rate of zero.
His approach also adopted an elitist strategy that copied the best string found in
the current population to the next generation. Selection was performed by hold-
ing 4 competitions between strings that were adjacent in the population array,
and declaring to the individual with the highest fitness as the winner. Krishnaku-
mar [15] compared his micro-GA against a simple GA (with a population size
of 50, a crossover rate of 0.6 and a mutation rate of 0.001). Krishnakumar [15]
reported faster and better results with his micro-GA on two stationary functions
and a real-world engineering control problem (a wind-shear controller task). Af-
ter him, several other researchers have developed applications of micro-GAs [13,
7,12,21]. However, to the best of our knowledge, the current paper reports the
first attempt to use a micro-GA for multiobjective optimization, although some
may argue that the multi-membered versions of PAES can be seen as a form of
micro-GA1 [14]. However, Knowles & Corne [14] concluded that the addition of a
population did not, in general, improve the performance of PAES, and increased
the computational overhead in an important way. Our technique, on the other
hand, uses a population and traditional genetic operators and, as we will show
in a further section, it performs quite well.
1
We recently became aware of the fact that Jaszkiewicz [11] proposed an approach
in which a small population initialized from a large external memory and utilized it
for a short period of time. However, to the best of our knowledgem this approach
has been used only for multiobjective combinatorial optimization.
128 C.A. Coello Coello and G.T. Pulido

3 Description of the Approach

In this paper, we propose a micro-GA with two memories: the population

memory, which is used as the source of diversity of the approach, and the
external memory, which is used to archive members of the Pareto optimal
set. Population memory is respectively divided in two parts: a replaceable and
a non-replaceable portion (the percentages of each can be regulated by the
user).
The way in which our technique works is illustrated in Fig. 1. First, an initial
random population is generated. This population feeds the population memory,
which is divided in two parts as indicated before. The non-replaceable portion
of the population memory will never change during the entire run and is meant
to provide the required diversity for the algorithm. The initial population of
the micro-GA at the beginning of each of its cycles is taken (with a certain
probability) from both portions of the population memory as to allow a greater
diversity.
During each cycle, the micro-GA undergoes conventional genetic operators:
tournament selection, two-point crossover, uniform mutation, and elitism (re-
gardless of the amount of nondominated vectors in the population only one is
arbitrarily selected at each generation and copied intact to the following one). Af-
ter the micro-GA finishes one cycle (i.e., when nominal convergence is achieved),
we choose two nondominated vectors2 from the final population (the first and
last) and compare them with the contents of the external memory (this memory
is initially empty). If either of them (or both) remains as nondominated after
comparing against the vectors in this external memory, then they are included
there. All the dominated vectors are eliminated from the external memory. These
two vectors are also compared against two elements from the replaceable por-
tion of the population memory (this is done with a loop, so that each vector is
only compared against a single position of the population memory). If either of
these vectors dominates to its match in the population memory, then it replaces
it. The idea is that, over time, the replaceable part of the population memory
will tend to have more nondominated vectors. Some of them will be used in the
initial population of the micro-GA to start new evolutionary cycles.
Our approach uses three types of elitism. The first is based on the notion that
if we store the nondominated vectors produced from each cycle of the micro-GA,
we will not lose any valuable information obtained from the evolutionary process.
The second is based on the idea that if we replace the population memory by
the nominal solutions (i.e., the best solutions found when nominal convergence
is reached), we will gradually converge, since crossover and mutation will have
a higher probability of reaching the true Pareto front of the problem over time.
This notion was hinted by Goldberg [8]. Nominal convergence, in our case, is
defined in terms of a certain (low) number of generations (typically, two to five
in our case). However, similarities among the strings (either at the phenotypical
2
This is assuming that we have two or more nondominated vectors. If there is only
one, then this vector is the only one selected.
 A Micro-Genetic Algorithm for Multiobjective Optimization 129

Fig. 1. Diagram that illustrates the way in which our micro-GA works.

or genotypical level) could also been used as a criterion for convergence. The
third type of elitism is applied at certain intervals (defined by a parameter called
“replacement cycle”). What we do is to take a certain amount of points from
all the regions of the Pareto front generated so far and we use them to fill in
the replaceable memory. Depending on the size of the replaceable memory, we
choose as many points from the Pareto front as necessary to guarantee a uniform
distribution. This process intends to use the best solutions generated so far as
the starting point for the micro-GA, so that we can improve them (either by
getting closer to the true Pareto front or by getting a better distribution). This
also avoids that the contents of the replaceable memory becomes homogeneous.
130 C.A. Coello Coello and G.T. Pulido

To keep diversity in the Pareto front, we use an approach similar to the adap-
tive grid proposed by Knowles & Corne [14]. The idea is that once the archive
that stores nondominated solutions has reached its limit, we divide the search
space that this archive covers, assigning a set of coordinates to each solution.
Then, each newly generated nondominated solution will be accepted only if the
geographical location to where the individual belongs is less populated than the
most crowded location. Alternatively, the new nondominated solution could also
be accepted if the individual belongs to a location outside the previously spefi-
cied boundaries. In other words, the less crowded regions are given preference
so that the spread of the individuals on the Pareto front can be more uniform.
The pseudo-code of the algorithm is the following:

function Micro-GA
begin
Generate starting population P of size N
and store its contents in the population memory M
/* Both portions of M will be filled with random solutions */
i=0
while i < Max do
begin
Get the initial population for the micro-GA (Pi ) from M
repeat
begin
Apply binary tournament selection
based on nondominance
Apply two-point crossover and uniform mutation
to the selected individuals
Apply elitism (retain only one nondominated vector)
Produce the next generation
end
until nominal convergence is reached
Copy two nondominated vectors from Pi
to the external memory E
if E is full when trying to insert indb
then adaptive grid(indb )
Copy two nondominated vectors from Pi to M
if i mod replacement cycle
then apply second form of elitism
i=i+1
end while
end function

The adaptive grid requires two parameters: the expected size of the Pareto
front and the amount of positions in which we will divide the solution space for
each objective. The first parameter is defined by the size of the external memory
 A Micro-Genetic Algorithm for Multiobjective Optimization 131

and it is provided by the user. The second parameter (the amount of positions in
which we will divide the solution space) has to be provided by the user as well,
although we have found that our approach is not very sensitive to it (e.g., in most
of our experiments a value of 15 or 25 provided very similar results). The process
of determining the location of a certain individual has a low computational cost
(it is based on the values of its objectives as indicated before). However, when
the individual is out of range, we have to relocate all the positions. Nevertheless,
this last situation does not occur too often, and we allocate a certain amount of
extra room in the first and last locations of the grid to minimize the occurrence
of this situation.
When the external memory is full, then we use the adaptive grid to decide
what nondominated vectors will be eliminated. The adaptive grid will try to
balance the amount of individuals representing each of the elements of the Pareto
set, so that we can get a uniform spread of points along the Pareto front.

4 Comparison of Results

Several test functions were taken from the specialized literature to compare our
approach. In all cases, we generated the true Pareto fronts of the problems using
exhaustive enumeration (with a certain granularity) so that we could make a
graphical comparison of the quality of the solutions produced by our micro-GA.
Since the main aim of this approach has been to increase efficiency, we ad-
ditionally decided to compare running times of our micro-GA against those of
the NSGA II [6] and PAES [14]. In the following examples, the NSGA was run
using a population size of 100, a crossover rate of 0.8 (using SBX), tournament
selection, and a mutation rate of 1/vars, where vars = number of decision vari-
ables of the problem. In the following examples, PAES was run using a depth of
5, a size of the archive of 100, and a mutation rate of 1/bits, where bits refers
to the length of the chromosomic string that encodes the decision variables. The
amount of fitness function evaluations was set such that the NSGA II, PAES
and the micro-GA could reasonably cover the true Pareto front of each problem.

4.1 Test Function 1

Our first example is a two-objective optimization problem defined by Deb [5]:

Minimize f1 (x1 , x2 ) = x1 (1)

g(x2 )
Minimize f2 (x1 , x2 ) = (2)
x1
132 C.A. Coello Coello and G.T. Pulido

where:

(  2 ) (  2 )
x2 − 0.2 x2 − 0.6
g(x2 ) = 2.0 − exp − − 0.8 exp − (3)
0.004 0.4

and 0.1 ≤ x1 ≤ 1.0, 0.1 ≤ x2 ≤ 1.0.

Fig. 2. Comparison of results for the first test function.

The parameters used by the micro-GA for this example were: size of the ex-
ternal memory = 100, size of the population memory = 50, number of iterations
= 1500, number of iterations of the micro-GA (to achieve nominal convergence)
= 2, number of subdivisions of the adaptive grid = 25, crossover rate = 0.7,
mutation rate = 0.029, percentage of non-replaceable memory = 0.3, population
size (of the micro-GA) = 4, replacement cycle at every 50 iterations.
Our first test function has a local Pareto front to which a GA can be easily
attracted. Fig. 2 shows the true Pareto front for this problem with a continuous
line, and the results found by the NSGA II, PAES and our micro-GA are shown
with points. Similar fronts were found by the three approaches. For this exam-
ple, both the NSGA II and PAES performed 12,000 evaluations of the fitness
function. The average running time of each algorithm (over 20 runs) were the fol-
lowing: 2.601 seconds for the NSGA II (with a standard deviation of 0.33555913),
1.106 seconds for PAES (with a standard deviation of 0.25193672) and only 0.204
seconds for the micro-GA (with a standard deviation of 0.07764461).
 A Micro-Genetic Algorithm for Multiobjective Optimization 133

4.2 Test Function 2

Our second example is a two-objective optimization problem proposed by

Schaffer [18] that has been used by several researchers [19,1]:



 −x if x≤1

−2 + x if 1<x≤3
Minimize f1 (x) = (4)

 4 − x if 3<x≤4

−4 + x if x>4

Minimize f2 (x) = (x − 5)2 (5)

and −5 ≤ x ≤ 10.

Fig. 3. Comparison of results for the second test function.

The parameters used for this example were: size of the external memory =
100, size of the population memory = 50, number of iterations = 150, number
of iterations of the micro-GA (to achieve nominal convergence) = 2, number of
subdivisions of the adaptive grid = 25, crossover rate = 0.7, mutation rate =
0.056 (1/L, where L=18 bits in this case), percentage of non-replaceable memory
= 0.3, population size (of the micro-GA) = 4, replacement cycle at every 25
iterations.
134 C.A. Coello Coello and G.T. Pulido

This problem has a Pareto front that is disconnected. Fig. 3 shows the true
Pareto front for this problem with a continuous line (the vertical line is obviously
not part of the true Pareto front, but it appears because we used linear segments
to connect every pair of nondominated points). We used points to represent the
solutions found by the NSGA II, PAES and our micro-GA.
Again, similar Pareto fronts were found by the three approaches. For this ex-
ample, both the NSGA II and PAES performed 1,200 evaluations of the fitness
function. The average running time of each algorithm (over 20 runs) were the fol-
lowing: 0.282 seconds for the NSGA II (with a standard deviation of 0.00014151),
0.107 seconds for PAES (with a standard deviation of 0.13031718) and only 0.017
seconds for the micro-GA (with a standard deviation of 0.0007672).

4.3 Test Function 3

Our second example is a two-objective optimization problem defined by Deb [5]:

Minimize f1 (x1 , x2 ) = x1 (6)

Minimize f2 (x1 , x2 ) = g(x1 , x2 ) · h(x1 , x2 ) (7)

where:

g(x1 , x2 ) = 11 + x22 − 10 · cos(2πx2 ) (8)

( q
f1 (x1 ,x2
1− if f1 (x1 , x2 ) ≤ g(x1 , x2 )
h(x1 , x2 ) = g(x1 ,x2 (9)
0 otherwise

and 0 ≤ x1 ≤ 1, −30 ≤ x2 ≤ 30.

This problem has 60 local Pareto fronts. Fig. 4 shows the true Pareto front
for this problem with a continuous line. The results obtained by the NSGA II,
PAES and our micro-GA are displayed as points.
The parameters used by the micro-GA for this example were: size of the ex-
ternal memory = 100, size of the population memory = 50, number of iterations
= 700, number of iterations of the micro-GA (to achieve nominal convergence)
= 4, number of subdivisions of the adaptive grid = 25, crossover rate = 0.7,
mutation rate = 0.029, percentage of non-replaceable memory = 0.3, population
size (of the micro-GA) = 4, replacement cycle at every 50 iterations.
Once again, the fronts produced by the three approaches are very similar.
For this example, both the NSGA II and PAES performed 11,200 evaluations
of the fitness function. The average running time of each algorithm (over 20
 A Micro-Genetic Algorithm for Multiobjective Optimization 135

Fig. 4. Comparison of results for the third test function.

runs) were the following: 2.519 seconds for the NSGA II (with a standard de-
viation of 0.03648403), 2.497 seconds for PAES (with a standard deviation of
1.03348519) and only 0.107 seconds for the micro-GA (with a standard deviation
of 0.00133949).

4.4 Test Function 4

Our fourth example is the so-called “unitation versus pairs” problem [9], which
involves the maximization of two functions over bit strings. The first function, f1
is the number of pairs of adjacent complementary bits found in the string, and
the second function, f2 is the numbers of ones found in the string. The Pareto
front in this case is discrete. We used a string length of 28, and therefore, the
true Pareto front is composed of 15 points.
The parameters used for this example were: size of the external memory =
100, size of the population memory = 15, number of iterations = 1250, number
of iterations of the micro-GA (to achieve nominal convergence) = 1, number
of subdivisions of the adaptive grid = 3, crossover rate = 0.5, mutation rate
= 0.035, percentage of non-replaceable memory = 0.2, population size (of the
micro-GA) = 4, replacement cycle at every 25 iterations.
Fig. 5 shows the results obtained by our micro-GA for the fourth test func-
tion. A total of 13 (out of 15) elements of the Pareto optimal set were found on
average (only occasionally was our approach able to find the 15 target elements).
PAES was also able to generate 13 elements of the Pareto optimal set on average,
and the NSGA II was only able to generate 8 elements on average.
136 C.A. Coello Coello and G.T. Pulido

Unitation versus pairs

28 1
27 0 0 1
26 0 0 0 0 1
25 0 0 0 0 0 0 1
24 0 0 0 0 0 0 0 0 1
23 0 0 0 0 0 0 0 0 0 0 1
22 0 0 0 0 0 0 0 0 0 0 0 0 1
21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
18 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
f2

13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0
2 0 0 0 0 0
1 0 0 0
0 0

0 1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627
f1

Fig. 5. Results of the micro-GA for the fourth test function.

For this example, both the NSGA II and PAES performed 5,000 evaluations
of the fitness function. The average running time of each algorithm (over 20 runs)
were the following: 2.207 seconds for the NSGA II, 0.134 seconds for PAES and
only 0.042 seconds for the micro-GA.
Borges & Barbosa [2] reported that were able to find the 15 elements of the
Pareto optimal set for this problem, using a population size of 100 and 5,000
evaluations of the fitness function, although no actual running times of their
approach were reported.
4.5 Test Function 5
Our fifth example is a two-objective optimization problem defined by Kursawe
[16]:

X
n−1  q 
Minimize f1 (x) = −10 exp −0.2 x2i + x2i+1 (10)
i=1

n
X

Minimize f2 (x) = |xi |0.8 + 5 sin(xi )3 (11)
i=1

where:

−5 ≤ x1 , x2 , x3 ≤ 5 (12)
 A Micro-Genetic Algorithm for Multiobjective Optimization 137

Fig. 6. Comparison of results for the fifth test function.

Fig. 6 shows the true Pareto front for this problem as points. The results
obtained by the NSGA II, PAES and our micro-GA are also shown as points.
It is worth mentioning that PAES could not eliminate some of the dominated
points in the runs performed.
The parameters used for this example were: size of the external memory =
100, size of the population memory = 50, number of iterations = 3000, number
of iterations of the micro-GA (to achieve nominal convergence) = 2, number
of subdivisions of the adaptive grid = 25, crossover rate = 0.7, mutation rate
= 0.019, percentage of non-replaceable memory = 0.3, population size (of the
micro-GA) = 4, replacement cycle at every 50 iterations.
For this example, both the NSGA II and PAES performed 2,400 evaluations
of the fitness function. The average running time of each algorithm (over 20 runs)
were the following: 6.481 seconds for the NSGA II (with a standard deviation of
0.053712), 2.195 seconds for PAES (with a standard deviation of 0.25408319) and
only 0.704 seconds for the micro-GA (with a standard deviation of 0.00692099).

5 Conclusions and Future Work

We have introduced an approach that uses a GA with a very small population

and a reinitialization process to generate the Pareto front of a multiobjective
optimization problem. The technique has a exhibited a low computational cost
when compared to the NSGA II and PAES in a few test functions. The approach
uses three forms of elitism, including an external file of nondominated vectors
138 C.A. Coello Coello and G.T. Pulido

and a refilling process that allows us to approach the true Pareto front in a
successive manner. Also, we use an adaptive grid (similar to the one used by
PAES [14]) that maintains diversity in an efficient way.
The approach still needs more validation (particularly, with MOPs that have
more decision variables and constraints), and needs to be compared with other
EMOO approaches (under similar conditions) using some of the metrics that
have been proposed in the literature (see for example [22]). We only provided
running times produced on a PC, but a more exhaustive comparison is obviously
lacking. However, the preliminary results presented in this paper, indicate the
potential of the approach.
Some other future work will be to refine part of the process, so that we can
eliminate some of the additional parameters that the approach needs. Since some
of them are not very critical (e.g., the number of grid subdivisions, or the amount
of iterations to reach critical convergence), we could probably automatically
preset them to a reasonable value so that the user does not need to provide
them.
Finally, we are also interested in using this approach as a basis to develop a
model of incorporation of preferences from the decision maker [4].

Acknowledgements. We thank the anonymous referees for their comments

that helped us improve the contents of this paper. The first author gratefully
acknowledges support from CONACyT through project 34201-A. The second
author acknowledges support from CONACyT through a scholarship to pursue
graduate studies at the Maestrı́a en Inteligencia Artificial of LANIA and the
Universidad Veracruzana.

References

1. P. J. Bentley and J. P. Wakefield. Finding Acceptable Solutions in the Pareto-

Optimal Range using Multiobjective Genetic Algorithms. In P. K. Chawdhry,
R. Roy, and R. K. Pant, editors, Soft Computing in Engineering Design and Man-
ufacturing, Part 5, pages 231–240, London, June 1997. Springer Verlag London
Limited. (Presented at the 2nd On-line World Conference on Soft Computing in
Design and Manufacturing (WSC2)).
2. Carlos C.H. Borges and Helio J.C. Barbosa. A Non-generational Genetic Algorithm
for Multiobjective Optimization. In 2000 Congress on Evolutionary Computation,
volume 1, pages 172–179, San Diego, California, July 2000. IEEE Service Center.
3. Carlos A. Coello Coello. A Comprehensive Survey of Evolutionary-Based Mul-
tiobjective Optimization Techniques. Knowledge and Information Systems. An
International Journal, 1(3):269–308, August 1999.
4. Carlos A. Coello Coello. Handling Preferences in Evolutionary Multiobjective
Optimization: A Survey. In 2000 Congress on Evolutionary Computation, volume 1,
pages 30–37, Piscataway, New Jersey, July 2000. IEEE Service Center.
5. Kalyanmoy Deb. Multi-Objective Genetic Algorithms: Problem Difficulties and
Construction of Test Problems. Evolutionary Computation, 7(3):205–230, Fall
1999.
 A Micro-Genetic Algorithm for Multiobjective Optimization 139

6. Kalyanmoy Deb, Samir Agrawal, Amrit Pratab, and T. Meyarivan. A Fast Eli-
tist Non-Dominated Sorting Genetic Algorithm for Multi-Objective Optimization:
NSGA-II. KanGAL report 200001, Indian Institute of Technology, Kanpur, India,
2000.
7. G. Dozier, J. Bowen, and D. Bahler. Solving small and large scale constraint sat-
isfaction problems using a heuristic-based microgenetic algorithm. In Proceedings
of the First IEEE Conference on Evolutionary Computation, pages 306–311, 1994.
8. David E. Goldberg. Sizing Populations for Serial and Parallel Genetic Algorithms.
In J. David Schaffer, editor, Proceedings of the Third International Conference on
Genetic Algorithms, pages 70–79, San Mateo, California, 1989. Morgan Kaufmann
Publishers.
9. Jeffrey Horn, Nicholas Nafpliotis, and David E. Goldberg. A Niched Pareto Ge-
netic Algorithm for Multiobjective Optimization. In Proceedings of the First IEEE
Conference on Evolutionary Computation, IEEE World Congress on Computa-
tional Intelligence, volume 1, pages 82–87, Piscataway, New Jersey, June 1994.
IEEE Service Center.
10. Hisao Ishibuchi and Tadahiko Murata. Multi-Objective Genetic Local Search Al-
gorithm. In Toshio Fukuda and Takeshi Furuhashi, editors, Proceedings of the 1996
International Conference on Evolutionary Computation, pages 119–124, Nagoya,
Japan, 1996. IEEE.
11. Andrzej Jaszkiewicz. Genetic local search for multiple objective combinatorial op-
timization. Technical Report RA-014/98, Institute of Computing Science, Poznan
University of Technology, 1998.
12. E.G. Johnson and M.A.G. Abushagur. Micro-Genetic Algorithm Optimization
Methods Applied to Dielectric Gratings. Journal of the Optical Society of America,
12(5):1152–1160, 1995.
13. Charles L. Karr. Air-Injected Hydrocyclone Optimization via Genetic Algorithm.
In Lawrence Davis, editor, Handbook of Genetic Algorithms, pages 222–236. Van
Nostrand Reinhold, New York, 1991.
14. Joshua D. Knowles and David W. Corne. Approximating the Nondominated
Front Using the Pareto Archived Evolution Strategy. Evolutionary Computation,
8(2):149–172, 2000.
15. K. Krishnakumar. Micro-genetic algorithms for stationary and non-stationary
function optimization. In SPIE Proceedings: Intelligent Control and Adaptive Sys-
tems, pages 289–296, 1989.
16. Frank Kursawe. A variant of evolution strategies for vector optimization. In
H. P. Schwefel and R. Männer, editors, Parallel Problem Solving from Nature.
1st Workshop, PPSN I, volume 496 of Lecture Notes in Computer Science, pages
193–197, Berlin, Germany, oct 1991. Springer-Verlag.
17. Geoffrey T. Parks and I. Miller. Selective Breeding in a Multiobjective Genetic
Algorithm. In A. E. Eiben, M. Schoenauer, and H.-P. Schwefel, editors, Parallel
Problem Solving From Nature — PPSN V, pages 250–259, Amsterdam, Holland,
1998. Springer-Verlag.
18. J. David Schaffer. Multiple Objective Optimization with Vector Evaluated Genetic
Algorithms. PhD thesis, Vanderbilt University, 1984.
19. N. Srinivas and Kalyanmoy Deb. Multiobjective Optimization Using Nondomi-
nated Sorting in Genetic Algorithms. Evolutionary Computation, 2(3):221–248,
fall 1994.
20. David A. Van Veldhuizen and Gary B. Lamont. Multiobjective Evolutionary Algo-
rithms: Analyzing the State-of-the-Art. Evolutionary Computation, 8(2):125–147,
2000.
140 C.A. Coello Coello and G.T. Pulido

21. Fengchao Xiao and Hatsuo Yabe. Microwave Imaging of Perfectly Conducting
Cylinders from Real Data by Micro Genetic Algorithm Coupled with Determinis-
tic Method. IEICE Transactions on Electronics, E81-C(12):1784–1792, December
1998.
22. Eckart Zitzler, Kalyanmoy Deb, and Lothar Thiele. Comparison of Multiobjective
Evolutionary Algorithms: Empirical Results. Evolutionary Computation, 8(2):173–
195, Summer 2000.

View publication stats