Genetic Algorithms (GA) were first introduced by HOLLAND
[99]. GAs are so-called population based search strategies.
They maintain a set of points, so-called genomes, in a function space and try to
make use of analogies to biological evolution by performing mutation and
crossover operations between the individuals of a population. Starting with an
initial population, the algorithm evolves by iteratively creating a new
generation of individuals based on an already existing one. New individuals are
introduced by a so-called crossover operation, where at least two5.4individuals of a generation are chosen as 'parents'. Their genomes are combined
to produce children. On some of the newly generated children a so-called
mutation operation is performed by replacing a part of the child with a random
value. Depending on the genetic algorithm at hand some individuals of the
parental generation might survive into the following generation(s). Individuals
are chosen for mating based on their fitness. The fitness is a metric to
indicate how good an individual represents a solution of the problem. It
directly corresponds to the target value of a gradient based optimizer. When the
optimization is started an initial population of genomes is chosen. The
parameters of the genomes are initialized randomly but within given bounds. The
fitness of the population is then computed by evaluating each individual by
means of a simulation run.
Subsections
2003-03-27
