Research On Evolutionary Algorithms Collaborating With Multi-information

In scientific research and engineering applications,many practical problems can be transformed into optimization problems to be solved.As an important class of optimization tools,evolutionary algorithms(EAs)have received close attention from many researchers in recent years.EAs are a class of heuristic random search algorithms that simulate biological evolution in nature.They have the characteristics of simple structure,excellent performance,and strong robustness.Compared with some classic mathematical optimization algorithms,such as algorithms based on gradient information,EAs do not have high requirements on the mathematical properties of optimization problems,and can even be directly used as black box optimization tools.Therefore,EAs have been widely used in many practical optimization problems.In recent years,with the increasing complexity of optimization problems,the performance of EAs has also been greatly challenged,which is mainly manifested in low solution accuracy and slow convergence speed.Therefore,how to improve the performance of EAs is an urgent problem.Based on the characteristics of EAs,this paper proposes the idea of using multi-information collaboration to improve the performance of the algorithms.In EAs,individuals in a population usually contain a variety of information: fitness information,location information,and neighborhood information.Three types of information have different characteristics.Fitness information can intuitively reflect the strengths and weaknesses of individuals,location information can represent individuals' degree of clustering and neighborhood information can describe the topological structure of the population.The idea of multi-information collaboration is to design corresponding improvement strategies by comprehensively using this information,so as to improve the solution accuracy obtained by algorithms and speed up the convergence speed of algorithms.In this paper,we designed improved strategies based on multi-information collaborative thinking for enhancing two representative EAs: Differential Evolution(DE)and Artificial Bee Colony(ABC)algorithm.The main research work is as follows:(1)The performance of classic DE is highly dependent on the mutation strategy and its control parameters.In order to improve the availability of DE,based on the idea of multi-information collaboration,this paper proposes a dynamic grouped multi-strategy DE(MIGDE for short).In MIGDE,firstly,the population is divided into three groups according to the fitness information and location information of the individuals;secondly,the three groups are respectively configured with mutation strategies and parameter values with different search capabilities;finally,in order to further develop affection of the middle group,based on the individual's neighborhood information,the neighborhood search operation was applied to the middle group of individuals.In order to verify the performance of MIGDE,experiments were performed on 22 widely used test functions,and compared with 4 classic DEs and 8 well-known improved DEs.The experimental results show that MIGDE has advantages in solution accuracy and convergence speed.In addition,this paper uses MIGDE to solve two practical optimization problems: parameter estimation for frequency-modulated sound waves(FM)and spread spectrum radar poly-phase code design(SSR).Experimental results show that compared with classic DE,the accuracy of MIGDE results can be improved by 88.7% and 6.7%,respectively.(2)The classic ABC has the problem of strong exploration but poor exploitation,which makes the slow convergence speed.Therefore,based on the idea of multi-information collaboration,this paper proposes an improved ABC(SILABC for short)based on superior information learning.In SILABC,firstly,the neighborhood search operation is applied based on the neighborhood information of the individual;secondly,the superior individuals retained in the neighborhood search operation are regarded as the providers of superior information and according to probability are learned by individuals which their fitness values are not good;finally,to perform superior information learning based on the fitness information and location information.In order to verify the performance of SILABC,experiments were performed on 32 widely used test functions,and compared with the classic ABC,9 well-known improved ABC,and other three EAs.The experimental results show that SILABC is more accurate in solution and faster in convergence.In addition,this paper uses SILABC to solve the node coverage optimization problem of wireless sensor networks.The experimental results show that the accuracy of results obtained by SILABC can be improved by 10.64%.