Biomedical Engineering Reference
In-Depth Information
performed by means of a classical GA; also the
crossover and mutation operations will be applied
as usual but the descendants will be discarded if
they do not fulfil the criteria required for belong-
ing to the species of their progenitors, meaning if
their genotype is too different from the genotypes
of the species.
The natural distribution of individuals into
species and their separate evolution is tried to
be mimicked. In nature, there are individuals
adapted to cold, dry or hot environments; each
group keeps its life into a given environment by
means of adapting specific characteristics that
distinguish it form other groups.
Nevertheless, this technique is not free of
disadvantages; certain requirements are needed
for a good functioning and they are not directly
achieved during the initial arrangement of the
problem. For instance, it should be desirable
that the population could be perfectly distributed
throughout the whole search space, as well as that
the groups were fairly distributed along that space
and in a quantity in accordance with the number
of total solutions of the problem. Unexplored ar-
eas might exist if these characteristics were not
present; in contrast, there might be another areas
highly explored where, depending on the group-
ing procedure, several species might coexist. The
most of these problems can be avoided by means
of an automated mechanism for the number of
existing species. It seems clear that if the number
of solutions is different for different problems, the
GA, the responsible for finding solutions, should
be the one who manages the number of evolved
species in each generation. In order to do this, the
GA will be allowed to expand the starting number
of species as generations advance.
The increase of the species number will be
achieved by performing crossover on individu-
als from different species throughout several
generations. As the resulting descendants mix the
knowledge from the species of their ancestors, a
new species can be created in a different location
from the ones of their progenitors. In this way, the
species stagnation can be avoided and the explora-
tion of new areas, together with the appearance
of new knowledge, may be obtained; in short, the
environment diversity is achieved. As individuals
might migrate or be expelled and afterwards create
new species with other compatible individuals,
the performance of individuals is again modelled
in their natural environment.
The crossover operations between the species
are applied similarly to what happens in biology.
It origins, on one hand, the crossovers between
similar individuals are preferred (as it was done at
the previous step using GAs) and on the other, the
crossovers between different species are enabled,
although in a lesser rate.
From an initial population generated for these
techniques to be implemented, some steps - fol-
lowing described and graphically represented in
Figure 6- take place:
1.
Random creation of the genetic popula-
tion
2.
Organisation into species (a species com-
prises individuals that have high similarities
among their genotypes)
3.
Application of the Genetic Algorithm on
every one of the defined species
4.
Introduction of new individual coming from
the crossover on different species. These
individuals are located in another area of the
search space. A variant of the functioning
implies the elimination of the individuals
that, after several generations, do not create
a species large enough.
5.
Verifying whether the number of evolutions
reaches the maximum allowed or if the
population reaches a top level of individuals
(defined at the beginning of the execution).
If some of these conditions are not fulfilled,
the algorithm execution finishes; if they are
so, the individuals of the existing population
will be again arranged into species (step
2).
Search WWH ::
Custom Search