Research On Multimodal Genetic Algorithm And Multi-Objective Evolutionary Algorithm

With the development of application, the shortcomings of the genetic algorithm in multimodal optimization become more and more obvious. The niching method is effective and has been widely concerned as a hotspot in the research area. Currently, the theory of niching method is not perfect; as a result, there is no unified models, and it is difficult to compare or analyse different niching methods theoretically. Most of the niching methods are designed for either binary coding or real coding, but not both. As a result of the absence of the theory describing the localized character of the search space, some niching methods are not suitable for combination optimization. This paper discusses the character and performance of niching methods, and proposes a new multipopulation genetic algorithm aimed at multimodal optimization. C++ programs are used to carry out the simulated experiments and test performance of the algorithm.Optimization problems in reality are often concerned about more than one goal. They are so called multi-objective optimization problems. In this paper, we will propose a general framework of multi-objective evolutionary algorithms, and discuss its convergence.The main important work of this paper includes:(1) Design a new fitness transform to form niches. The fitness transform of fitness sharing model has disadvantages as follow: large amount of computation; difficulty in assigning a proper value to niching radius; assuming that the peaks are uniformly distributed. The new approach in this paper has overcomed these shortcomings.(2) Develop a new mode of gene communication. This paper sugguests that the ones complement of the best chromosome should be introduced to the population. This method works well, and a general concept, repulse function, is developed.(3) Discuss the convergence of a framwork of the multi-objective evolutionary algorithms.