Research On Dominance Relations Of Many-Objective Evolutionary Algorithms

Multi-objective optimization problems widely exist in various fields of real life applications and have become one of the research hotspots in optimization field.Different from the traditional single objective optimization problems,it is difficult to directly compare two solutions in multi-objective optimization problems.Therefore,it is much more difficult to solve than the traditional single objective optimization problem.When the number of objectives is more than 3,this kind of problem is usually named as many-objective optimization problem.The main method to solve this kind of optimization problem is many-objective evolutionary algorithm.Among them,the many-objective evolutionary algorithm based on Pareto dominance is one of the main algorithms to solve this kind of problem.However,with the increase of the number of objectives,the performance of Pareto dominance will decline rapidly.The reasons are as follows: 1)With the increase of the objectives,Pareto dominance cannot provide enough selection pressure,which leads to the phenomenon called domination resistance.2)The secondary selection criterion based on diversity in the algorithm will play a leading role in the environment selection,which leads to the algorithm difficult to converge.Therefore,this thesis studies the dominance relations in many-objective evolutionary algorithm.The specific research work is as follows.(1)Focus on the two archives in multi-objective optimization,this thesis designed a new two-archive evolutionary algorithm called two archive multi-objective evolutionary algorithm(TA-MOEA)is designed.Different from the traditional two archive algorithms which maintain convergence and diversity,the archives in TA-MOEA are used for global search and local search respectively.In the global search,genetic algorithm is used to improve the convergence of the algorithm.In local search,opposition-based learning is used to improve the diversity of the algorithm.The experimental results indicate that compared with the classical multi-objective evolutionary algorithms,the proposed algorithm has better performance in convergence and diversity.(2)Focus on Pareto domination cannot provide enough selection pressure,this thesis studies the dominance relations.a many-objective dominance relation namely convergence indicator based dominance relation(CDR)is designed.The convergence indicator and an adaptive parameter based on niche are used to balance the convergence and diversity of the population.Based on the dominance relation,a many-objective evolutionary algorithm named CDR-MOEA is designed.The experimental results indicate that the designed CDR dominating relation and CDR-MOEA show good performance in solving many-objective optimization problems.(3)Focus on many objectives in many-objective optimization,this thesis studies the evaluation methods in many-objective optimization and designed a many-objective evolutionary algorithm called angle-based bi-objective evolutionary algorithm(ABOEA).By transforming the many-objective optimization problem into a problem with only two objectives: minimizing convergence and maximizing diversity,the dominance probability between solutions is greatly improved.The experimental results indicate that the evaluation method can effectively evaluate the objectives values.Meanwhile,the comprehensive performance of ABOEA is greatly improved compared with the existing algorithms.