DCGA: a diversity control oriented genetic algorithm
Hisashi Shimodaira · 1997
Genetic algorithms (GA's) are one of promising means for function optimization. Methods for function optimization are required to attain the global optimum without getting stuck at local optima. For multimodal functions, the power of the traditional GA is poor in this point. In order to achieve this goal, the appropriate diversity in the structures of the population needs to be maintained during the search so that local search and global searches are performed in a balanced way. In this paper, I propose a new genetic algorithm called DCGA (diversity control oriented genetic algorithm). In the DCGA, the structures of the population for the next generation are selected from a merged population of parents and their children eliminating duplicates based on a selection probability, which is a function of a hamming distance between the candidate structure and thc structure with the best fitness values and is larger for structures with lager hamming distances. Within the range of my experiments, the performance of the DCGA is remarkably superior to that of the traditional GA and conjectured to be a promising competitor of previously proposed algorithms.