Genetic Algorithms and Multimodal Search - Advancing Artificial Intelligence Through Biological Process Applications - page 236

It should be borne in mind that a traditional
GA performs its search considering only one
sub-region (the whole of the search space). Here
the search space will be divided into different sub
regions or intervals according to the number of
genetic individuals in the genetic pool.
Since the individuals in the genetic pool have
restrictions in their viable gene values, one of
these individuals would not be provided with a
valid solution. In addition to the genetic pool,
another population is then used (the secondary
population), where a classical GA would develop
its individuals in an interactive fashion with those
individuals of the genetic pool.
Differently from the genetic pool, the genes
of individuals of secondary population provide
solutions as they may adopt values throughout the
whole of the search space; whereas, the genetic
pool would act as a support, keeping search space
homogeneously explored.
Next, both populations, which are graphi-
cally represented in Figure 2, will be described
in detail.

When offering a solution, traditional GA may
have any valid value, whereas in the proposed GA,
the range of possible values is restricted. Total
value range is divided into the same number of
parts than individuals in genetic pool, so that a
sub-range of values is allotted to each individual.
Those values that a given gene may have will
remain within its range for the whole of the per-
formance of the proposed GA.
In addition to all that has been said, every
individual at the genetic pool will be responsible
of the genes that correspond to the best found
solution up to then (meaning whether they belong
to the best individual at secondary population).
This Boolean value would be used to avoid the
modification of those genes that, in some given
phase of performance, are the best solution to
the problem.
Furthermore, every one of the genes in an
individual has an I value associated which
indicates the relative increment that would be
applied to the gene during a mutation operation
based only on increments and solely applied to
individuals of the genetic pool. It is obvious that
this incremental value should be lower than the
maximum range in which gene values may vary.
The structure of the individuals at genetic pool
is shown at Figure 3.
As these individuals do not represent global
solutions to the problem that has to be solved,
their fitness value will not be compulsory but it
will reduce the complexity of the algorithm and,
of course, it will increase the computational ef-
ficiency of the final implementation.

The Populations: Genetic Pool

As it has been previously mentioned, every one
of the individuals at the genetic pool represents
a sub-region of the global search space. There-
fore, they should have the same structure or gene
sequence than when using a traditional GA. The
difference lies in the range of values that these
genes might have.

Figure 3. Genetic pool individual i

Gene Values

…

G iN

G i1 G i2 G i3

… …

B i1 B i2 B i3 B iN I i1 I i2 I i3 I iN

Control: Best Solution? Mutation Increment