En savoir plus
In contrast to exact mathematical optimization techniques the development of heuristic optimization algorithms is not based on a strict mathematical formalism or theory but on a creative process of observation, abstraction, and simulation. The variety of heuristic optimization methods is enormous and also various natural processes have been used for inspiration. So these techniques are usually treated as optimization black boxes and it is very hard for practitioners of other areas to obtain any deeper understanding of their internal functioning or to efficiently tune parameters to achieve high quality results. Besides an intensive study of Genetic Algorithms another approach is developed in this book to scrape the black paint of the box. By recapitulating the natural archetype of Genetic Algorithms - the natural evolution process - from a microscopic and macroscopic point of view, alleles are identified as the basic entities Genetic Algorithms work with. This allele-oriented approach is the basis for the introduction of new measurement values enabling practitioners to obtain more information about the internal state of the algorithm and to understand critical factors concerning its performance. In the experimental part of the book these new measurement values are used to analyze the Standard Genetic Algorithm and especially to highlight genetic diversity as one of the most important influencing factors concerning solution quality and to show the critical interplay between hyperplane sampling (crossover) and neighborhood-based search (mutation).
