Diversity-based Offspring Selection Criteria for Genetic Algorithms
A. Scheibenpflug, S. Wagner, M. Affenzeller - Diversity-based Offspring Selection Criteria for Genetic Algorithms - Lecture Notes in Computer Science LNCS 9520, Las Palmas, Gran Canaria, Spain, 2015, pp. 393-400
Genetic algorithms can be affected by an early loss of diversity in their populations called premature convergence. To address this problem, this paper presents two extensions for the offspring selection genetic algorithm. Both extensions are based on diversity maintenance mechanisms applied when selecting offspring for the next generation. The first approach focuses on producing solutions that feature a predefined quality improvement as well as an appropriate structural distance from their parents. The second approach monitors the average diversity of the population and selects more diverse offspring if the population does not meet a predefined diversity. We show that these algorithms allow to control diversity and are useful methods for influencing the development of the population independent of the algorithms other parameters.