The Research On Adaptive Niche Genetic Algorithms

The multimodal problems exist widely in the fields of multimodal and machine learning and industry application. The niching genetic algorithms (NGAs) are effective for resolving the multimodal problems. At present, the fitness sharing genetic algorithms which base on the function sharing are frequently-used in the niching genetic algorithms. Making an estimate of the solution space in advance is the limitation of the fitness sharing genetic algorithms. Because most of the real decision problems have complicated solution space, which is hard to estimate the parameters the algorithm needs.After reading a mass of the reference documentation, the paper has done the work as follows:1. The development history and the current state of the research on the genetic algorithms are introduced in brief. After that, it expatiated on the basic theories and concepts of the genetic algorithm. At last, it also analyzed the algorithm's limitation.2. Three kinds of niche genetic algorithms are introduced. It presented the idea and the steps of each niche genetic algorithms and made the qualitative analysis on the searching ability and the convergence speed.3. In the basement of studying on the fitness sharing genetic algorithms, a self-adaptive K means clustering fitness sharing method is proposed. The combination of the suggested method and the standard fitness sharing algorithms forms an integrated solution scheme for multimodal problems. The new algorithm made the improvement on the K means clustering method by introducing a minimal clustering distance which can control the number of the convergent peaks. There is no need to know the radius and the number of the peaks in advance for the improved algorithm. The results show that the new algorithm has good searching ability on the multimodal optimization with the evolutionary generation increases.