54
P.A. Whigharn . G.B. Fogel
nonlinear problems can be searched for near-optimal solutions. However, since
evolution represents a stochastic search through the solution space, no application
can guarantee the optimal solution. However, for many problems areal-time nearoptimal solution is quite satisfactory, and in fact often it is not possible to
determine when an optimal solution has been reached. Evolutionary algorithms
find good solutions to novel problems in complex situations - requirements that
suit the concepts of ecology and ecological modelling.
4.3.1
The Basic Evolutionary Aigorithm
The basic evolutionary algorithm (EA) comprises the following main components,
independent of the actual representation used for individuals:
A method for generating an initial population of individuals. Typically this is
made at random. The representation of the individual in the population is typically
correlated to the problem that is being addressed.
A fitness function. This function gives a measure of fitness that can be used to
score the worth of individuals in the population in terms of their performance to
the task at hand.
A method of selection, based on fitness. This selection pressure drives the
population towards better solutions. Common forms of selection are proportional,
where the probability of selection is direct1y proportional to the fitness of an
individual compared with the population as a whole, and tournament (roundrobin), where the selection is based on a fitness ranking between a random sub set
of the population.
A method of reproduction with heritable variation. Reproduction may mimic
various genetic operators, such as mutation and crossover, to produce new
individual behavior in the population with so me variety. These operators
commonly allow for both random changes within an individual's representation,
and the sharing of genetic information between two or more members of the
population. Alternatively, the variation operators can be applied to the phenotype
direct1y, avoiding a requirement for genetic representation.
A method of determining and maintaining population size. The basic
evolutionary algorithm uses a single population with a constant number of
individuals for each generation. Other strategies allow a population to gradually
change via a steady-state mechanism, or allow the population to grow and decay
based on some measure of external resources.
A termination criterion. Typically this is based on a performance measure (i.e.
a minimum desired fitness measure) or a total effort in terms of number of
generations, dock time, or computer processing time.
P.A. Whigharn . G.B. Fogel
nonlinear problems can be searched for near-optimal solutions. However, since
evolution represents a stochastic search through the solution space, no application
can guarantee the optimal solution. However, for many problems areal-time nearoptimal solution is quite satisfactory, and in fact often it is not possible to
determine when an optimal solution has been reached. Evolutionary algorithms
find good solutions to novel problems in complex situations - requirements that
suit the concepts of ecology and ecological modelling.
4.3.1
The Basic Evolutionary Aigorithm
The basic evolutionary algorithm (EA) comprises the following main components,
independent of the actual representation used for individuals:
A method for generating an initial population of individuals. Typically this is
made at random. The representation of the individual in the population is typically
correlated to the problem that is being addressed.
A fitness function. This function gives a measure of fitness that can be used to
score the worth of individuals in the population in terms of their performance to
the task at hand.
A method of selection, based on fitness. This selection pressure drives the
population towards better solutions. Common forms of selection are proportional,
where the probability of selection is direct1y proportional to the fitness of an
individual compared with the population as a whole, and tournament (roundrobin), where the selection is based on a fitness ranking between a random sub set
of the population.
A method of reproduction with heritable variation. Reproduction may mimic
various genetic operators, such as mutation and crossover, to produce new
individual behavior in the population with so me variety. These operators
commonly allow for both random changes within an individual's representation,
and the sharing of genetic information between two or more members of the
population. Alternatively, the variation operators can be applied to the phenotype
direct1y, avoiding a requirement for genetic representation.
A method of determining and maintaining population size. The basic
evolutionary algorithm uses a single population with a constant number of
individuals for each generation. Other strategies allow a population to gradually
change via a steady-state mechanism, or allow the population to grow and decay
based on some measure of external resources.
A termination criterion. Typically this is based on a performance measure (i.e.
a minimum desired fitness measure) or a total effort in terms of number of
generations, dock time, or computer processing time.