Evolutionary Algorithms - Self-Adaptive Heuristics for Evolutionary Computation - page 14

Fitness

The quality of a candidate solution is called fitness referring to the biological
concept of fitness. The fitness of a solution can be derived by
• real world features - the quality of a real-world system can be measured by a

set of defined features, which have to be translated into comparable values,

• simulation models - here, systems are simulated artificially and analog to

online optimization,

• mathematical models ,and

• test suite problems for experimental EC studies.

Algorithm Structure and Populations Models

Figure 2.2 shows the pseudocode of a generalized EA. In the first step a popu-
lation of so called individuals which represent a solution to an optimization or
search problem is initialized and evaluated. In an iteration loop the population
of solutions is changed, evaluated and selected until a termination condition is
fulfilled. Meanwhile, there exist various classes of EAs, partly with different his-
torical background. They differ in the kind of representation, genetic operators
and problem domains they are appropriate to.
The two main variants of panmictic population models are the generational
model and the steady-state model. In the generational model λ = μ individuals
are produced from the parental population of size μ using the genetic operators.
Afterwards, all offspring individuals replace the parents of the last generation.
In the steady-state model not the entire population is changed, but λ<μ new
individuals replace λ old ones. The ratio λ/μ is called generational gap. All
individuals in such a panmictic population model are considered to be potential
mates. Recently, many other population models were introduced, which are based
on neighborhood relationships. In such structured population models individuals
can only interact with certain neighbors determined by these relations. In the
island or multi population models the panmictic population is divided into several
smaller ones [154].

1 Start

2 t:=0;

3 Initialize P ( t );

4 Assign fitness to P ( t );

5 Repeat

Variation of P ( t ) →P ( t ) 6

Assign fitness to P ( t ); 7

Select P ( t +1) from P ( t ); 8

9 t := t +1;

10 Until termination condition

11 End

Fig. 2.2. Pseudocode of a generalized evolutionary algorithm