What is really important about genetic algorithms?
Genetic algorithms can be adapted in order to solve directly multi-
objective programming optimisation problems. We can implement a
parallel search of many Pareto-optimal solutions within the design
variables domain.
The fitness function has to be modified. Fitness can be selected
according to the non-dominated (Pareto-optimality) property. If we
find a non-dominated solution, we set a very high fitness value,
according to the formula. If the solution is dominated, the fitness value is
lower.
The population is divided into two sub-population. The first one,
denoted as rank 1, is constituted by the solutions that belong to
the approximated Pareto-optimal set. The remaining solutions are
sorted according to the definition, generating a second rank. And
so on for all the elements inside the population. This
implementation is generally more efficient than using non-linear
programming techniques after a scalarisation.
|
|