The Research On Many-objective Evolutionary Algorithms Based On Fuzzy Dominance

In recent years,There is a considerable number of multiobjective optimization problems(MOPs)in people' s life and production.Those problems are characterized by having more than one object to be optimized simultaneously,and in general,they conflict with each other.Unlike the evolutionary algorithms applied to a single objective,the multiobjective evolutionary algorithms(MOEAs)are used to solve the multiobjective optimization problem,and the final results will be selected to obtain a series of Pareto optimal solutions for decision makers.Applying Pareto dominance relation to multiobjective optimization greatly promotes the development of optimization algorithm,and it has achieved effective results in solving optimization problems with 2 to 3 objectives.However,in the reality of production and life,the high dimension multiobjective optimization problem accounts for a large proportion.The traditional Pareto based evolutionary algorithm solves such problems because it lacks sufficient selection pressure in the process of fitness evaluation,so it faces Pareto dilemma.In addition,Pareto hierarchical sorting will consume a lot of time,and it is difficult to balance convergence and diversity.It is also a common problem of Pareto based domination evolutionary algorithm in dealing with high-dimensional multiobjective optimization problems.In order to improve the performance of evolutionary algorithm on high-dimensional multiobjective optimization problems,and effectively improve the convergence and set distribution of the algorithm,the fuzzy dominance rule(FDR)based on fuzzy logic is proposed in this thesis.Based on the framework of classical Pareto dominance-based evolutionary algorithms,this thesis uses a new fitness evaluation mechanism to distinguish the optimal solution into several different optimization levels.As a case of study,the concept of fuzzy domination is first adopted in the third generation fast non-dominated sorting genetic algorithm based on reference points,which seems to be more advanced and frequently used,and the original Pareto dominance-based fitness evaluation rules are replaced by fuzzy fitness evaluation mechanism.The improved algorithms are compared with the currently used several advanced evolutionary algorithms and show better performance on DTLZ standard test function sets.The results of simulation experiments show that the algorithms proposed in this thesis has advantages over other algorithms.In the next part,this thesis combines the fuzzy theory with the decomposition strategy inmultiobjective optimization.In the classical decomposition-based multiobjective evolutionary algorithm MOEA/D,the calculation rule of the fuzzy dominance level is adopted,and an improved decomposition-based fuzzy dominance many-objective evolutionary algorithm MOEA/D-FPD is proposed.In order to verify the performance of the modified algorithm,the MOEA/D-FPD algorithm and several commonly-used multiobjective evolutionary algorithms with nice performance are simulated on the WFG series standard test function set,and the optimization performance of the evolutionary algorithms on the test problems with different problem features and different numbers of objective is carried out.The comparative analysis shows that the adoption of fuzzy dominance rules can improve the diversity and convergence of the solution set.