Asynchronous Hierarchical Parallel Evolutionary Algorithm And Its Application In Fuzzy Clustering Analysis

Evolutionary computation is global stochastic search algorithm based on biology evolution mechanism, such as natural selection, heredity, mutation and etc. It is capable of finding the approximate global optimum solution without requiring continuous, differentiable and unimodal, thus it has been widely used in NP/NPC problems, neural network optimization problems, multi-objective optimization problems and many other fields. However, as the scale and complexity of problems grow, the search process of sequential evolutionary algorithm will be prolonged geminately. Thus, parallel evolutionary computation becomes important. Cantú-Paz classified it into four categories, including master-slave model, coarse-grained model, fine-grained model and hierarchical model. They are usually implemented in the parallel programming environments, such as MPI, PVM, OpenMP and etc.In order to avoid conquest and noneffect problems, overcome premature convergence, and improve efficiency, an asynchronous hierarchical parallel evolutionary algorithm (AHPEA) is presented in the thesis. In AHPEA, information of fitness is added into the extended fuzzy recombination operator, and a new fuzzy recombination operator based on standardized fitness (SF-FRO) is proposed to accelerate the convergence. Then, a heterogeneous model is constructed. Different exploration/exploitation degree is assigned to SF-FRO of each subpopulation which has distinct topological structure to get appropriate selection press. Finally, subpopulations are adequately connected and an asynchronous migration model is constructed. Thus, the conquest and noneffect problems are solved, premature convergence is avoided and efficiency is improved.In simulation research, the performance of AHPEA is tested with a set of representative problems. It is proceeded in three stages, namely, analysis of SF-FRO, heterogeneous model and asynchronous migration. The experimental results lead to three conclusions. (1) SF-FRO works well in speeding up the convergence for its information of fitness indicates the potential search direction and range. (2) The heterogeneous model outperforms the homogeneous model in solving massively multimodal problems. (3) The asynchronous migration model obviously improves the performance of evolutionary algorithm theoretically and experimentally.As a practical application, in the last part of the thesis, the theory foundation for fuzzy clustering analysis is studied and the optimization model of FKCN clustering...