Research On Weight Vector Adjustment And Operator Selection In Decomposition-Based Multi-Objective Optimization Algorithms

Multi-objective evolutionary algorithm is a method that combines evolutionary algorithm and multi-objective optimization strategy to solve multi-objective optimization problems.It is widely used in industry and other fields.multi-objective evolutionary algorithm based on decomposition(MOEA/D)is an important branch of decomposition.Because the idea of decomposition can reduce the difficulty of the overall optimization and the complexity of the algorithm,it has been widely studied in the past decade.However,with the deepening of research and increasing application requirements,MOEA/D still needs to be optimized and improved in terms of balancing convergence and diversity.In this thesis,the weight vector and the operators of MOEA/D are studied and optimized,and the proposed method in this thesis is analyzed and verified by experiments.The research of this thesis is mainly divided into the following two aspects:(1)Proposes a weight grouping operators selection strategy(WGOS).WGOS divides the weight vector into several groups and assigns a different operator to each different group.In the process of evolution,the size of the group is adjusted by comparing the advantages and disadvantages of the subgenerations generated by different groups.The subgroup with better subgeneration will be appropriately expanded while the subgroup with worse subgeneration will be correspondingly reduced until the final game selects a group and corresponds to an operator.Since MOEA/D uses the solutions of adjacent subproblems to optimize the quantum problem,and different operators adopted by different parts of a population may deviate the development of solutions,MOEA/ DWGOS will quickly determine the optimal operator in the early stage of evolution,and test the operator in the evolution process.If the operator performance is not good,the second-best operator in the game process is replaced to ensure that a set of optimal solutions can be obtained.A great number of experiments show that this kind of strategy has good performance.(2)The strategy of weight vector adjustment is proposed.In the middle and late stages of evolution,the crowding degree of individuals is judged,and then the weight vector is properly adjusted according to the crowding degree of individuals.Some weight vectors are inserted in the sparse regions of individuals,and some weight vectors are deleted in the crowded regions of individuals,so that the solutions can converge to the Pareto Front better.A large number of experiments show that the decomposition class multi-objective evolutionary algorithm using this weight vector adjustment strategy has obvious advantages in terms of balancing convergence and diversity.