Evolutionary Algorithms - Self-Adaptive Heuristics for Evolutionary Computation - page 13

see also chapter 4 where in particular the second rule is discussed. The strength
of mutation is also known as step size or mutation rate and is a measure for the
level of change of the solutions.

Crossover/Recombination

The role of crossover, also called recombination, is discussed intensively. Recom-
bination aims at the inheritance of genetic material from at least two parental
solutions. In chapter 6, the role of recombination is discussed. For string repre-
sentations, a crossover point is chosen ( one-point crossover ) and two offspring
solutions are created by exchanging the two parts of the divided strings. For
numerical representations, crossover can be implemented as the arithmetic me-
dian of the continuous objective variables. The building block hypothesis (BBH)
[47] claims that recombination contributes to the identification of useful solution
constructing blocks. In contrast to the BBH, the theory of the genetic repair ef-
fect claims that recombination identifies and combines common features of the
solution.

Selection

Selection is the antagonist to the variation operators. It gives the evolutionary
search a direction. Only selected individuals are allowed to survive or inherit
their genetic material to the offspring. There are two types of selection opera-
tors of which the first one mentioned here is parent selection . In order to produce
offspring via recombination, a number of parents have to be selected. Parent se-
lection is equivalent to mate selection in biology. Its implementations can reach
from random selection to fitness based selection schemes. A typical selection
mechanism is called fitness proportional selection , where the probability p a that
an individual a is selected depends on its absolute fitness compared to the ab-
solute fitness of the rest of the population

f a
j f j

f a
fn

p a = = (2.2)

n j f j For the case of standard GAs,
where the population size is fixed, the expected number of copies of each indi-
vidual becomes

1

with the average population fitness f =

n a = p a n = f a /f. (2.3)

The second type of selection operators is survivor selection .Aftertheoffspring
population has been generated, survivor selection gives a direction to the opti-
mization process by copying the selected individuals into the population of the
next generation. This concept is equivalent to Darwin's principle of survival of
the fittest.