Genetic Algorithm And Its Application On Cluster Analysis

Genetic algorithm based on natural selection and genetic theory,is a global optimization algorithm which combine the rules of the fittest survival and the chromosomes random permutation mechanism within groups in the process of biological evolution.It provides the common framework for a solution of nonlinear,multi-model, multi-objective optimization problem in the complex systems.It needs not depend on the specific areas of the problem,and has been widely successfully used in many fields of science and technology.However,traditional genetic algorithm has some shortcomings, Such as slow convergence,and premature convergence sometimes and so on.Genetic algorithm also needs further study.Cluster analysis is one of the core technologies of data mining as well as one of the main branch of multivariate statistical analysis.After years of development,cluster analysis has a solid theoretical foundation and formed a systematic approach system. However the traditional cluster analysis method has largely been confined to theoretical analysis and probability assumptions in the data distribution,fewer to consider the specific characteristics and differences among the actual data.Therefore,the problem how to overcome the dependence of probability hypothesis in the traditional method of cluster analysis has become an academic research hot spot problem in recent years.According the above-mentioned issues,this thesis will research the improvement of genetic algorithm and cluster analysis which based on the genetic algorithm.With the research of mechanism between general genetic algorithm and traditional method of cluster analysis,I present some methods for each other:fuzzy adaptive genetic algorithm, pseudo-parallel genetic cluster analysis method.The idea of fuzzy adaptive genetic algorithm which used population variance and entropy is dedicated to adjust the cross probability and mutation probability in the fuzzy inference system,so it will ensure the diversity of the population.The results of simulation experiments show that the performance of genetic algorithm has been largely improved,it is better to solve the general genetic algorithm problem of premature convergence;The idea of Pseudo-parallel genetic clustering analysis is that we could encode samples of each type with real-coded,through the air kind of identification and then repair the incorrect chromosome.Based on the introduction of discrete random mutation and optimization direction mutation operator,we can combine migration strategies and insert tactics, achieve the purpose of both local convergence speed and global convergence properties. Consequently,this method will overcome the initialization sensitive issue,as well as the issue of easy to fall into local extreme which owned by the traditional clustering algorithm based on the clustering criteria.The simulation experiment compared with K-means algorithm shows that the new clustering method based on the pseudo-parallel genetic algorithm is feasible and effective.