Next: Extensible Genetic Optimizer (EGO) Up: 4.2 Optimization Strategies Previous: 4.2.3 DIRECT Search Optimization Contents
The genetic optimization algorithm uses a similar approach to the storage of genetic information in DNA (deoxyribonucleic acid) [259,260,261]. The retrieval and transfer of genetic information from the parents to the children can be modeled for optimization purposes introducing mathematical operators which an equivalent to the natural mutation of the DNA and are equivalent to the inheritance of properties by constructing a new DNA for a child out of the DNA information from the parents. The DNA consists of a certain amount of chromosomes which are represented by a set of free parameters. The sets of free parameters are also called designs. A population consists of a certain number of individuals (designs). According to their fitness function4.8, the individuals remain alive or are discarded in favor of new individuals.
The designs can be altered according to the different operators mutation, inheritance, crossover, and selection. The mutation operator changes a small number of parameters with in one design to search for an improvement in the near neighborhood in terms of genetic information. Good parent DNA can be inherited and thus the information of a ``good'' design can be kept for the next iteration. The crossover operator uses the DNA information of two parent designs to form a new design. A crossover can be performed using one-point, two-point, or multiple-point crossover operations. The best individuals of a particular population survive analog to nature. On the basis of these best individuals, the genetic operations are performed. If the number of maximal designs has not been reached, the remaining designs will be initialized by randomly selected parameter sets.
Each of the genetic operations are applied with a certain probability which can be configured at the optimization set-up. According to the different implementations of the genetic algorithms, some operators can be given priority or even switched off. In the worst case the genetic optimizer degenerates to a random generator only.
The termination criteria for the genetic optimization can be classified into three categories: the maximum number of population evaluations has been reached, the maximum of computational time has been exceeded, or an interruption of the user has been initiated. However, the result of a genetic optimization is a certain number (typically 10) of the best designs (individuals) of all evaluated populations.