The Research On Many-objective Optimization Evolutionary Algorithm Based On Coordination Selection

Optimization problems with only two or three objectives and conflicting objectives are called multiobjective optimization problems(Multiobjective Optimization Problems,MOPs).When the number of objectives to be optimized is more than three,MOPs are defined as many-objective optimization problems(Many-objective Optimization Problems,MaOPs).In the classic multiobjective evolutionary algorithms(Multi-objective Evolutionary Algorithms,MOEAs),Pareto dominion,which is very effective,loses the ability to distinguish individuals when solving MaOPs.For example,classic MOEAs such as NSGA-II,SPEA2 and MOEA/D cannot search for the real Pareto front(Pareto Front,PF)when dealing with MaOPs.Therefore,it is an urgent challenge to ensure the convergence of the algorithm in high-dimensional environment.In addition,while ensuring the good convergence of the solution set obtained by the final algorithm optimization,MOEAs also needs to ensure that the final solution is more evenly distributed in the whole PF.How to balance the convergence and distribution are two aspects that must be considered in the design of MOEAs.This paper proposes two kinds of many-objective optimization evolution algorithms.They are "A many-objective Evolutionary Algorithm based on Rotation and Decomposition"(MaOEA-RD)and "A Many-objective Evolutionary Algorithm based on Staged the Coordination Selection"(MaOEA-SCS).To overcome the shortcomings of Pareto dominance in many-objective optimization,which cannot distinguish the advantages and disadvantages of individuals,the two methods have completely different evolutionary ideas.The former,like most MaOEAs,equalizes convergence and distribution in every iteration.The latter adopts a different evolutionary approach,that is,phased coordinated selection.Unlike the former,convergence and distribution focus on only one aspect of each iteration.MaOEA-RD improves the ability to identify individuals by rotating the coordinate system and calculating the distance between individuals and the constructed hyperplane.It is combined with a novel individual selection mechanism and reference vectors adjustment mechanism to ensure the convergence of the population.MaOEA-SCS divide the evolutionary stage into the convergence stage and the distribution stage,combining two different selection criteria to coordinate the population from generation to generation.In the experiments of both methods and three kinds of advanced evolutionary algorithms(RVEA,NSGA-III and VaEA)on DTLZ and WFG benchmark problems were compared,and according to the different characteristics of the two methods,MaOEA-RD and the other four outstanding evolutionary algorithms(RVEA*,ANSGA-III,MOEA/D-AWA and MOEA/D-URAW)were compared,MaOEA-SCS and three other excellent evolutionary algorithms(MaOEA-IGD,?-DEA and SPEA/R)were compared.The experimental results show that the two methods presented in this paper have excellent performance under different benchmark problems and are very competitive.