A Study Of Cooperative Co-evolution Based On Bidirection-dectection Differential Grouping

Large-scale global optimization problems are typical problems in scientific research and engineering area.Compared with traditional mathematical methods and evolutionary algorithms,Cooperative Coevolution framework can effectively solve Large-Scale Global Optimization problems.Cooperative Coevolution framework is divided into two steps: problem decomposition and problem solving.Among them,Differential Grouping is an advanced problem decomposition method,which can accurately identify the association between a pair of decision variables,but there are still shortcomings of low decomposition accuracy and rapid increase of computational cost in the process of problem decomposition.Cooperative Coevolution solves large-scale global problems with static decomposition method or dynamic decomposition method,which is more scalable,but still has shortcomings.Cooperative Coevolution with static decomposition method is inefficient when solving fully non-separable problems,and Cooperative Coevolution with dynamic decomposition methods is computationally expensive when solving partially separable problems.To address the above problems,the main research of this paper is as follows.1.In order to improve the performance of the DG based methods for solving LSGO problems,a Bidirectional-Detection Differential Grouping(BDDG)method is proposed in Chapter 3.By designing a Bidirectional Detection Structure(BDS),BDDG is able to spend less computation than other DG-based methods.An Adaptive Perturbation Strategy(APS)is proposed to improve the decomposition accuracy of the BDS.Analytical methods are used to demonstrate that the complexity of BDDG is lower than that of the other state-of-the-art DG-based methods.Experimental results show that BDDG substantially reduces the computational cost for problem decomposition,improves the decomposition accuracy on some problems,and the computational cost used by BDDG grows more slowly with increasing problem dimensionality.2.To improve the performance of the CC with decomposition method in solving LSGO problems,a Two-Stage Decomposition Cooperative Coevolution(CCTSD)framework is proposed in Chapter 4.In the first decomposition stage,the problem is decomposed by a static decomposition method with archive.In the second decomposition stage,Known Information based Dynamic Decomposition(KIDD)method automatically identifies the type of subcomponents based on the archive,followed by a second decomposition of high-dimensional subcomponents and dynamic regrouping of incorrectly grouped decision variables.Experimental results show that CCTSD improves the performance of CC in the fully non-separable problems,and the optimization results are excellent in other problems.