The Research On Niche Genetic Algorithm Based On Clustering

Genetic algorithm,which is based on natural selection and genetic theory, has been successfully applied in combination optimization, machine learning, engineering optimization, image processing, artificial life, auto programming and other domain applications. As the research going deeply, the shortcomings of genetic algorithm in dealing with multi-modal optimization are gradually exposed. Simple genetic algorithm can't keep a good diversity in the whole searching, so it would be possibly immersed in some local optimal situation, which leads to premature convergence. On the other way, genetic algorithms tend to converge to only one peak of multi-peaks function, which can't satisfy the need of searching more than one peak. As an effective way of dealing multi-peak optimization problems, the niche method has been widely concerned as a hotspot in the genetic algorithm research area.Nowadays, without perfect theory and unified model, it is difficult to analyze the connection among niche forming method, population diversity maintaining and evolutionary searching efficiency in theory and experiment. This thesis studied the niche genetic algorithm based on clustering, and the main work and research results are as follows:Firstly, a multi-source diffusion ant colony niche genetic algorithm was proposed, which is based on ant colony pheromone diffusion. This algorithm selects and reserves several pheromone sources in order to maintain diversity and diffused the pheromone to improve the searching efficiency. The algorithm can effectively prevent premature convergence of genetic algorithm.Secondly, an improved niche genetic algorithm based on clustering was proposed, which applies clustering method into realization of niche. It is proved that the algorithm can converge to several optimums by the Markov Chain model. Thirdly, a niche genetic algorithm based on mountain clustering was proposed, which improves the diversity by the improved mountain clustering method.Experiment of multi-peak function shows the diversity of this algorithm is better than that of deterministic crowding niche and fitness sharing niche. And the proposed algorithm can explore out more peaks.