Font Size: a A A

Research On Co-evolutionary Algorithm For Handling LSGO Problem

Posted on:2020-11-01Degree:MasterType:Thesis
Country:ChinaCandidate:Y ChenFull Text:PDF
GTID:2428330590477229Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
In the real world,there are a large number of decision variables in many engineering and numerical problems.When the number of decision variables is more than 1000 dimensions,they are called large-scale global optimization(LSGO)problems,and with the development of science and technology,more and more LSGO problem with more and more decision variables needs to be solved.The d ifficulty of the LSGO problem lies in the “dimension disaster”.The increase of the dimension leads to an exponential increase in the search space,which causes the traditional evolutionary algorithm to fail when solving the LSGO problem.Co-evolutionary algorithm is one of the effective methods to deal with LSGO problem.Based on the idea of divide-and-conquer,the co-evolutionary algorithm has great advantages in dealing with LSGO problem.The realization of the co-evolutionary algorithm is mainly divided into three steps.First,a large-scale problem is decomposed into multiple low-dimensional sub-problems;then,under the cooperation of other subcomponents,each low-dimensional sub-problem is optimized separately;finally,all solution of sub-problems are merged.Existing algorithms for solving the LSGO problem are prone to premature convergence and loss of population diversity during evolution,leading to evolutionary stagnation.These phenomena are mainly due to the fact that the population is trapped in a local optimum,so that multiple individuals in the population overlap,resulting in missing dimensions.Since the existing grouping strategy does not divide the importance of the groups,the computing resources are equally divided,resulting in the algorithm not improving the search quality.In view of the above problems,the main work and research contents of this paper are as follows:(1)This paper introduces the basic concepts of the LSGO problem and the separability of the LSGO problem.Because of the co-evolutionary algorithm groups the variables of the LSGO problem and achieves the purpose of segmenting the search space,the difficulty of the problem can be greatly reduced.The co-evolutionary algorithm is selected to solve the LSGO problem.Several common grouping strategies in the co-evolutionary algorithm are analyzed and compared.The principle of several grouping strategies is introduced respectively.The grouping effect of each grouping strategy is tested through experiments.(2)In view of the evolutionary process,the evolutionary algorithm leads the individuals in the population to the local optimum,which leads to the stagnation of evolution and the lack of dimensions.A co-evolutionary algorithm based on dimension missing detection and recovery is studied.When the algorithm shows the phenomenon of “population degradation,evolution stagnation” in the evolution process,the theory of principal component analysis is used to detect the dimension of information missing in the population,and the individuals with poor fitness are selected to shrink and expand in the missing dimension To achieve the effect of recovery.The dimension missing detection operator and the dimensional recovery operator are designed in the algorithm,combined with the co-evolutionary algorithm framework,in order to reduce the complexity of dimension missing detection.Experimental analysis shows that the algorithm can reduce the complexity of degree loss detection,increase population diversity and improve algorithm performance.(3)Aiming at the non-uniform contribution of decision variables to the output in the LSGO problem,the algorithm divides the finite computing resources irrationally into the each sub-component,and proposes a co-evolution algorithm based on the importance and correlation metrics.Firstly,the sensitivity analysis method is used to identify the sensitivity of decision variables to the problem,and the contribution of decision variables to the problem is analyzed,and the importance of each decision variable is measured accordingly.The variables are stratified according to the importance of the variables,and the variables are guided according to the correlation between the variables,so that there is no correlation between the grouped groups,high correlation within the group,and allocating them computing resources according to the importance of the group,compared with the traditional The simple correlation grouping performed better.Experiments show that the proposed algorithm has obvious advantages in solving the LSGO problem.
Keywords/Search Tags:LSGO problem, Co-evolutionary algorithm, dimension missing monitoring, grouping strategy, sensitive analyze
PDF Full Text Request
Related items