Segregative Genetic Algorithms (SEGA): A hybrid superstructure upwards compatible to genetic algorithms for retarding premature convergence

Many problems of combinatorial optimization belong to the class of NP-complete problems and can be solved efficiently only by heuristics. Both, Genetic Algorithms and Evolution Strategies have a number of drawbacks that reduce their applicability to that kind of problems. During the last decades plenty of work has been investigated in order to introduce new coding standards and operators especially for Genetic Algorithms. All these approaches have one thing in common: They are very problem specific and mostly they do not challenge the basic principle of Genetic Algorithms. In the present paper we take a different approach and look upon the concepts of a Standard Genetic Algorithm (SGA) as an artificial self organizing process in order to overcome some of the fundamental problems Genetic Algorithms are concerned with in almost all areas of application. With the purpose of providing concepts which make the algorithm more open for scalability on the one hand, and which fight premature convergence on the other hand, this paper presents an extension of the Standard Genetic Algorithm that does not introduce any problem specific knowledge. On the basis of an Evolution-Strategy-like selective pressure handling some further improvements like the introduction of a concept of segregation and reunification of smaller subpopulations during the evolutionary process are considered. The additional aspects introduced within the scope of that variants of Genetic Algorithms are inspired from optimization as well as from the views of bionics. In contrast to contributions in the field of genetic algorithms that introduce new coding standards and operators for certain problems, the introduced approach should be considered as a novel heuristic appliable to multiple problems of combinatorial optimization using exactly the same coding standards and operators for crossover and mutation as done when treating a certain problem with a Standard Genetic Algorithm. Furthermore, the corresponding Genetic Algorithm is unrestrictedly included in all of the newly proposed hybrid variants under especial parameter settings. In the present paper the new algorithm is discussed for the traveling salesman problem (TSP) as a well documented instance of a multimodal combinatorial optimization problem. In contrast to all other evolutionary heuristics that do not use any additional problem specific knowledge, we obtain solutions close to the best known solution for all considered benchmark problems (symmetric as well as asymmetric benchmarks) which represents a new attainment when applying Evolutionary Computation to the TSP.