Looking Inside Genetic Algorithms

Heuristic optimization techniques are frequently treated as black boxes. Until now very few mathematical theory is available due to the high complexity of numerous benchmark and real world problems and also because of the numerous variaty of different algorithms . So it is difficult for a user to obtain any hints about the internal functioning of an algorithm and to tune parameters to achieve high quality results. The present monograph aims to overcome this problem for Genetic Algorithms by looking at them from a different angle. After studying the biological archetype and the area of population genetics, alleles are identified as the basic entities manipulated by Genetic Algorithms. This allele-oriented view leads to the introduction of new measurement values providing a deeper insight and helping the user to monitor the optimization process in more detail, to identify performance problems, to compare parameter settings, and to understand the fundamental functioning of Genetic Algorithms more clearly. Furthermore, these measurement values are used in a comprehensive set of test runs to highlight various aspects of the hyperplane sampling process and to shed light on the interplay between hyperplane sampling (crossover) and neighborhood search (mutation) in the Standard Genetic Algorithm.