Clustering Multi-objective Evolutionary Algorithm And Its Application Research

Regardless of human life or production process,people can always encounter a lot of complicated multi-objective optimization problems(MOPs),such problems generally have multiple variables,multiple equations or inequality constraints and a number of nonlinear target volume and so on.The traditional method,such as weighting method and constraint method,can not deal with these problems well,but multi-objective evolutionary algorithm(MOEA)can not be limited by the characteristics of problem rules,and has many advantages and is able to obtain remarkable results.Multi-objective evolutionary algorithm is mainly composed of new solution generation and environment selection,At present,lots of the concerns and research works focus on the environment selection,and the new solution generation is not given so much concentration.Therefore,this thesis proposes to apply a popular method of machine learning,i.e.,clustering algorithm(CA)into the multi-objective evolution algorithms,taking full account of the rules of MOPs,and study efficient and improved way to produce new solutions,and that make the algorithm has better performance.Firstly,the multi-objective distribution estimation algorithms usually exist some drawbacks,such as the properties of the MOPs to be solved are not adequately considered,the outliers are not properly addressed,and the diversity of the produced new solutions are not enough,the modeling complexity is high.To deal with these drawbacks,an improved multi-objective distributed estimation algorithm(CEDA)based on clustering is studied.CEDA uses the agglomerative hierarchical clustering algorithm to analyze the population data in each iteration.Based on this structure,a multi-Gaussian model is established for all solutions.According to this model,the appropriate sample is selected to obtain the new individuals.In order to reduce the computational cost of modeling,adjacent individuals share the same covariance matrix to establish a Gaussian model.Based on the results of the standard test,CEDA can solve the complicated multi-objective optimization problems.Then,this paper proposes an improved multi-objective particle swarm optimization(CPSO)algorithm based on clustering technology,because the multi-objective particle swarm optimization(PSO)algorithm althought has high convergence speed but easy to lose the diversity of population.CPSO in the process of each iteration cycle to generate a new solution,the use of clustering algorithm for all individual cluster analysis,each individual matching individuals were identified at a given probability from the global or local population selection,and in order to better maintain the population solution diversity and the convergence speed of the algorithm,the adaptive solution of the new solution is a composite differential evolution algorithm with particle swarm optimization or diversity.Contrastive experiments based on standard test questions show that CPSO can also solve complex multi-objective optimization problems.Finally,two new clustering-based multi-objective evolutionary algorithms are applied to the optimization of return satellite cabin layout and optimization design of gear reducer for a light aircraft.The performance of the new algorithm in solving practical engineering problems is verified.