Research On Coevolution Algorithm Based On Double Space

In nature,organisms evolve through survival of the fittest and genetic variation.The idea of evolution is added to the calculation to produce evolutionary calculation.This paper is mainly engaged in a sub direction in the field of Evolutionary Computing:coevolutionary computing.The difference between coevolutionary computing and general evolutionary computing mainly lies in the individual fitness evaluation method in the algorithm.In general evolutionary algorithms,the individual fitness is determined by itself;In coevolutionary algorithm,individual fitness is not only related to individual itself,but also related to other individuals.In contrast,coevolutionary algorithm is more similar to the real natural situation.It is a deeper simulation of natural evolutionary behavior and has great research significance.Based on this research,the main contributions of this paper are as follows.Firstly,an evolutionary algorithm framework based on double space is proposed.A new information space is organized outside the population space of the general evolutionary algorithm.The information space extracts state information,constraint information,historical information and other information from the population space,and completes the update evolution based on its own update operator with the evolution of the population.In addition,information space will play a guiding role in the process of population evolution.According to the combination analysis of various information,it will affect the evolutionary behavior of population space.Different evolutionary algorithms can be combined in the population space to complete the evolutionary solution process through the co evolution of information space and population space.Secondly,under the framework of double space evolutionary algorithm proposed above,two coevolutionary algorithms cooperative coevolutionary algorithm and competitive coevolutionary algorithm are studied.Combined with the newly proposed lion swarm optimization algorithm and some excellent mechanisms in recent years,two new coevolutionary algorithms are proposed.Firstly,under the framework of dual space coevolutionary computing,a cooperative coevolutionary lion swarm optimization algorithm(DS-CoopCLSO)is proposed,which avoids the premature phenomenon of the algorithm through the guidance of historical information,and uses 30 functions such as single peak,multi peak,combinatorial function and composite function proposed at cec2014 meeting,as well as the standard evaluation method formulated by the conference,The performance of the proposed algorithm is verified.By comparing with a variety of recognized typical evolutionary algorithms,the performance of DS-CoopCLSO is verified in different aspects.In addition,the algorithm is applied to multi threshold image segmentation,and the results show that the segmentation effect is good.Secondly,a double space competitive coevolution lion swarm optimization(DS-CompCLSO)based on double space is proposed.By giving the two populations a competitive relationship,they can complete coevolution in the process of competition.Each population is divided into young subgroups and adult subgroups according to their age.Through the independent evolution of subgroups and mutual learning and other behavioral processes,the local and global search ability of the algorithm is balanced.Finally,the proposed algorithm is used in deep convolution generation adversarial networks(DCGAN)to improve the stability of model training.By comparing the images generated on different data sets,the effectiveness of DS-CompCLSO in the optimization process is verified.