Multi-Objective Differential Evolution Algorithm And Its Application On Data Clustering

In recent years,the research of many-objective optimization problems has achieved rapid development.Therefore,how to design an evolutionary algorithm that achieve a good balance between convergence performance and diversity is used to solve many-objective optimization problems become a difficult point in the field of evolutionary computation.As a data analysis tool,clustering analysis has widely used in various fields of research and application.Therefore,how to design an efficient automatic clustering algorithm has become one of the hotspots in the research of cluster analysis.Based on the above background,the research content of this paper is as follows:A many-objective differential evolution algorithm using a relaxed dominance relation is proposed.In the proposed algorithm,a relaxed domination relation is designed and incorporated to increase the selection pressure of individuals.Population is coevolved with an external archive,and the child population is generated by the mixed differential mutation operators.The fitness of each individual is evaluated based on an indicator method,and the population is updated.The archive is updated according to the Lp norm(0<p<l)distance based diversity maintenance strategy.The proposed algorithm is simulated with two other classical algorithms in a set of standard test functions,show that the proposed algorithm can generate a set of non-dominated solutions with better convergence and distribution in many-objective optimization problems.A multi-objective automatic differential evolution clustering algorithm based on a strategy of class-center density is proposed.In the clustering process,due to the randomness of the class-center selection leads to the selected class-center deviate from the data set,or the class-center is too concentrated and the error brought by clustering of this defect,the algorithm is used select class-center for two times.In order to make the algorithm get the optimal class-center quickly,an improved clustering criterion function is proposed to dynamically punish the number of clusters.The proposed algorithm is simulated with two other existing automatic clustering algorithms,as well as a classical clustering algorithm on UCI and artificial data sets.The experiment show that the proposed algorithm can obtain better clustering results.