The Study On Many-objective Optimization Problems Based On Adapti've Neighborhood Selection

In the real life,Many problems need to be optimized for a large number of objectives,which is more than 3.We call the problems many-objective optimization problems,MaOPs.The MOEAs for solving such problems are called the many-objective evolutionary algorithms,MaOEAs.Generally,the conflict between convergence performance and distribution performance will deteriorate with the increase of the number of objectives.The Pareto dominance also loses its effectiveness in the high-dimensional problems.The selection mechanism of the traditional MOEAs can not provide enough selection pressure for distinguishing superior individuals from the objective space which is flooded a large number of individuals.Therefore,an effective selection method is needed to improve the selection pressure to balance convergence performance and distribution performance.This paper proposes a MaOEA,called "Adaptive neighborhood selection for many-objective optimization problems,ANS-MOEA",to deal with the MaOPs.We design two pieces of information for each individual;convergence information(CI)and distribution information(DI).In the critical layer selection,a well-converged individual(calculated by CI)is selected from the population firstly,and the individuals in the neighborhood(calculated by DI)are then placed in the neighborhood set(we defined it as neighborhood collection,NC).When NC is unsaturated,the neighborhoods of the individuals were selected into NC directly.When the NC is saturated,the neighborhood need to compare with the individuals,which are existing in the NC.The individual which has bigger DI value is retained into a population,and individuals with smaller DI value is selected in NC,When archive is saturated,the individuals in the NC are eliminated.In order to verify the effect of ANS-MOEA,this paper selects four state-of-the-art MaOEAs,NSGA-III,IBEA,BIGE,GrEA,as the comparison algorithms.The experimental results show that ANS-MOEA can effectively improve the selection pressure of the algorithm,so as to provide a set of Pareto optimal solution sets which can balance convergence and distribution in solving MaOPs.