Study Of Sampling Strategy Based On Differential Evolution In Estimation Of Distribution Algorithms

Evolutionary algorithm is a heuristic optimization algorithm based on population which is inspired by biological evolution.Evolutionary algorithms have been widely applied to industrial design,bibliography,economy,power engineering and social science,etc.Estimation of distribution algorithm is a kind of popular evolutionary algorithm in evo-lutionary computation.Unlike traditional evolutionary algorithms,there is no crossover or mutation operations.Generally,estimation of distribution algorithm consists of three steps,modeling,sampling and selection.Sampling is a crucial step to estimation of distribution algorithm,as it is related to whether to generate more promising offspring population.The promising offspring population is significant to the final solutions.Differential evolution has attracted many researchers from various fields since proposed in 1995.Differential evolution is a kind of simple but effective stochastic optimization algorithm.Due to the advantages of differential evolution,it has been widely utilized in different fields.There have been quantities of significant work of differential evolution on route plan,electronic engineering,communication engineering,engineering system design,etc.This paper mainly studies the sampling strategy of estimation of distribution algorith-m.Traditional estimation of distribution algorithms utilize the built probabilistic model to sample.Inspired by differential evolution,this paper proposes a sampling strategy based on differential evolution,namely DES.And this strategy is utilized in estimation of distribution algorithm to study its performance.This paper combines DES with estimation of distribu-tion algorithm,and studies its performance on single objective optimization problems and multiobjective optimization problems respectively.According to the comprehensive experi-mental analysis,it can be seen that DES is impressively significant to improve the sampling of estimation of distribution algorithm with dramatical potential.The main contents of this paper are single objective estimation of distribution algo-rithm based on DES and multiobjective estimation of distribution algorithm based on DES.For single objective optimization problems,based on the framework of DE/EDA,utilize a differential evolution based on eigenvector to sample.Meanwhile,expensive Local Search is applied to improving the quality of solutions.Hence,single objective estimation of distri-bution algorithm based on DES is finally proposed.On the other hand,for multiobjective optimization problems,utilize DES to improve the sampling of RM-MEDA,then propose multiobjective estimation of distribution algorithm based on DES with impressive perfor-mance.This paper applies DES to single objective estimation of distribution algorithm and multiobjective estimation of distribution algorithm.The comprehensive and system-atic experimental analysis indicates the significance of DES in estimation of distribution algorithms.