Information Technology Reference
In-Depth Information
4.3
Gene Duplication
Besides the polynomial hypermutation, the omni-aiNet algorithm also incorpo-
rates a second technique of genetic variation, known as
Gene Duplication
.This
mutation consists of the duplication of parts of the elements in the DNA chain
during the chromosome reading. According to Ohno [12] and Holland
et al.
[10],
this mutation has an important role in the evolution of species.
This mechanism has already been proposed by de Franca
et al.
[7] as a relevant
operator in dopt-aiNet. Basically, it randomly selects a coordinate
i
of the an-
tibody and replaces every element in the remaining coordinates by
x
i
whenever
this replacement improves the performance of the antibody.
4.4
Suppression, Binary Tournament and Random Insertion
The main goal of the
Suppression
phase of the algorithm is to eliminate re-
dundancy among individuals in the population and to maintain diversity when
associated with the insertion of new randomly generated individuals in the pop-
ulation (
Random Insertion
).
In the Suppression phase, the Euclidean distance in the variable space among
every individual in the population is calculated and normalized with respect
to the maximum distance found so far. In this context, the individuals close
enough to each other according to a suppression threshold (defined by the user)
are subject to a
Binary Tournament
procedure and the worst one is eliminated
from the population.
This Binary Tournament follows basically the same criteria used in the ranking
procedure, which means that a given solution
i
is preferred to a solution
j
if (
i
)
i
is feasible and
j
is not feasible; (
ii
)
i
has a smaller Constraint Violation than
j
and both are not feasible; and (
iii
) both solutions are feasible and
i
-dominates
j
. If both solutions are feasible and there is no
-dominance among then, the
winner solution is randomly selected.
The Random Insertion is a mechanism that contributes to the diversity of
the population by inserting
N
rand
new individuals randomly generated into the
population (
N
rand
must also be defined by the user).
The Suppression and Random Insertion steps of the algorithm, together with
cloning and hypermutation phases, are also responsible for other important char-
acteristic of omni-aiNet: the dynamic variation of the population size. The algo-
rithm is then allowed to define a proper number of antibodies in the population
at each iteration, according to the specified suppression threshold.
5Comp i eAn ly s
This section presents a brief conceptual and comparative analysis between omni-
aiNet and Deb and Tiwari's omni-optimizer (DT omni-optimizer) [8]. Besides the
distinct bio-inspiration, the omni-aiNet and the DT omni-optimizer algorithms
present several conceptual differences that may lead each algorithm to perform
differently according to the characteristics of the problems being treated.
