Research on Genetic Algorithms with Function Division Schemes

点击翻译按钮获取中文摘要

As early exploration of the search space gradually results in exploitation of a small number of fit regions later on in the search process, conventional genetic algorithms (GAs) tend to face difficulty in solving certain kinds of problem, such as highly multimodal functions, functions with high levels of epistasis , and deceptive features. On such problems, typical premature convergence occurs in conventional GAs, before the early exploration stage has had a chance to set the population in the right direction.To cope with this problem in genetic search, we proposed a "function division scheme" in GAs. In this scheme, we use separate populations for exploration and exploitation ; one of these populations, the explorer sub-GA, tries mainly to explore the whole search space and maintain a useful degree of global diversity, while the other population, the exploiter sub-GA, exploits the neighborhood of the best solution obtained so far. Thus the search function of the algorithm is divided into two functions, explorer and exploiter functions. The effectiveness of this scheme has been shown using several test functions. Further, we proposed several genetic operators that are useful for this scheme. They include multi-parent recombination operators and search space boundary extension operators.