Research On Differential Evolution Algorithm Based On Dynamic Neighborhood

The computational intelligence method is a general term for a class of algorithms that were inspired by the wisdom of nature and human intelligence.No matter in the field of scientific research or industrial production,people can not avoid meeting some global optimization problems.However,with the progress of the times,traditional optimization methods have been difficult to solve the increasingly complex optimization problems in various sciences and engineering.Because these complex optimization problems are often characterized by multi-objective,large-scale,constrained,dynamic,uncertain and so on.Therefore,the computational intelligence method has been widely concerned and developed greatly because of its excellent features such as simple structure,high efficiency and strong robustness.The differential evolution algorithm studied in this paper is an outstanding algorithm in the field of computational intelligence.The algorithm iterates,evolves and searches for the optimal solution in the solution space by maintaining a population of NP.It has characteristics of easy to implement,simple and efficient,strong robustness and so on,and has been successfully applied in the field of science and engineering to solve the optimization problems in various fields.However,the differential evolution algorithm also has some shortcomings.When solving some high-complex problems,it is prone to premature due to rapid convergence,and it is difficult to jump out the local optimum.At present,the actual scientific and engineering problems often have a large number of local optimal values,which makes it more difficult to obtain the global optimal solution.Differential evolution algorithm also proposes a variety of mutation strategies,but different strategies have different focuses.Selecting different mutation strategies for different optimization problems may produce a large difference results.The selection of mutation strategies has become a major problem in practical applications.Differential evolution algorithm is an algorithm based on population difference.In the later stage of the algorithm,the population difference gradually becomes smaller,the convergence speed of the algorithm becomes slower and slower,and the local exploration ability also weakens.Therefore,it is difficult to converge to the optimal solution with a limited computing time.Faced with the above-mentioned deficiencies,we have the following considerations: Differential evolution algorithm is an evolutionary algorithm based on population differences.The algorithm mainly relies on the individual interactions in the population to generate a new generation of populations,to gradual search and find the optimal solution.However,we find that most of the interactive individuals are selected with a completely random process in the original differential evolution.The population information has not been fully exploited and the mutation operator also has not played a guiding role.In this article,we will focus on how to exploit and utilize population information to improve the optimization performance of the differential evolution algorithm.Therefore,how to extract the useful information from the population and how to apply the effective population information to the evolutionary process of the algorithm search will be the two main aspects of this article.Based on the above considerations,this paper introduces the concept of dynamic neighborhood,hoping to fully utilize the dynamic neighborhood to mine the population information.At the same time,it also proposes mutation operation strategy based on neighborhoods to make full use of population information when selecting the parent vectors to guide the population search.This paper revolves around the dynamic neighborhood and proposes four kinds of differential evolutionary algorithm framework based on dynamic neighborhood.These frameworks strengthen the interaction of individuals and realizes the purposes of mining population information and utilizing useful information.(1)In order to break the restrictions on the interaction of population in the static neighborhood,avoid to fall into the local optimum,and also improve the neglect of population information by completely random selection mechanism,a dynamic neighborhood differential evolution algorithm based on random grouping is proposed.In the process of evolution,the population continuously rerandomly groups and dynamically changes neighborhoods of individuals to realize the mining of population information.Then combined with the neighborhood-based mutation strategy,this algorithm can also make full use of useful information.(2)In order to improve the defects of random grouping for population information mining,combined with the advantages of topology neighborhood,a dynamic neighborhood differential evolution algorithm based on adaptive neighborhood size is proposed.While utilizing the topology to mine population information,it also makes full use of individual's information to guide search.(3)To improve the defects that single topology can not adapt different problems in the different evolution process,and explore the synergy between multiple topologies,an adaptive multi-topological dynamic neighborhood differential evolution algorithm is proposed by combine with adaptive selection of topological operators to implement a dynamic neighborhood strategy between multiple topologies.(4)While fully exploiting population information with multi-topology,fully considering the utilization of individual population information,a multi-topological dynamic neighborhood differential evolution algorithm based on individual dependence is proposed to make the information mining and utilization of population closely related.In summary,this paper addresses the shortcomings of differential evolution algorithms,based on the concept of dynamic neighborhood,from random grouping to topology neighborhood,from a single topology to multi-topology collaboration,from adaptive selection operators to individual information utilization,proposes four layered progressive dynamic neighborhood strategy.At the same time,a large number of experimental evaluations prove that the proposed algorithms have a good optimization performance,and compare and analyze the respective advantages and disadvantages of these algorithmic strategies,to provide an effective reference for scientific research and engineering.