A self-adaptive model for selective pressure handling within the theory of genetic algorithms

In this paper we introduce a new generic selection method for Genetic Algorithms. The main difference of this selection principle in contrast to conventional selection models is given by the fact that it considers not only the fitness of an individual compared to the fitness of the total population in order to determine the possibility of being selected. Additionally, in a second selection step, the fitness of an offspring is compared to the fitness of its own parents. By this means the evolutionary process is continued mainly with offspring that have been created by advantageous combination of their parents' attributes. A self-adaptive feature of this approach is realized in that way that it depends on the actual stadium of the evolutionary process how many individuals have to be created in order to produce a sufficient amount of 'successful' offspring. The experimental part of the paper documents the ability of this new selection operator to drastically improve the solution quality. Especially the bad properties of rather disadvantageous crossover operators can be compensated almost completely.