Design And Application Of Multi-objective Evolutionary Algorithm Based On Cooperative Decomposition

In recent years,decomposition-based many-objective evolutionary algorithm has become a popular algorithm in the field of many-objective optimization due to its excellent selection pressure.However,the existing conical decomposition method and parallel decomposition method are still very sensitive to the shape of the front of the many-objective optimization problem,and it is difficult to maintain good distribution uniformity and diversity at the same time when capturing convex,concave,or steep fronts.Especially for the many-objective optimization of irregular complex fronts,such as fracture or deletion,the maintenance of population diversity is more important to obtain as wide a front as possible,which poses a higher challenge to the existing decomposition-based many-objective evolutionary algorithms.In this thesis,combining the advantages of conical decomposition and parallel decomposition in the detection ability and distribution of objective space,a collaborative decomposition method is designed,and a collaborative decomposition-based many-objective evolutionary algorithm(Co DEA)is proposed.Adaptive rotation coefficient and adaptive weight coefficient are introduced into the scalar objective aggregation function of the decomposition method.The adaptive rotation coefficient adaptively inherits the uniformity advantage of the parallel decomposition method and the diversity advantage of the conical decomposition method according to the different positions of the reference points of each subproblem in the hyperplane.The adaptive weight coefficient is designed to balance the advantages of the two decomposition methods in the generation of single-layer and double-layer reference points,and to cope with the phenomenon of edge missing that gets worse as the number of objects increases.For optimization problems with 5 or more objectives,to effectively alleviate the phenomenon of population center aggregation caused by double layers reference points,Co DEA designed an angle-based scalar objective aggregation function for inner-layer subproblems,which makes the individuals associated with the inner-layer reference points as far away from the central region of the population as possible.The experimental results of DTLZ,Convex?DTLZ,and WFG tests show that,compared with the other five popular many-objective optimization algorithms,the Co DEA algorithm proposed in this thesis has the overall optimal performance in the convergence,uniformity,and diversity of population distribution.In addition,to solve the many-objective optimization problem of the irregular complex front,this thesis introduced a Pareto-dominance-based population archive into Co DEA,so as to obtain candidates with good distribution by combining the advantages of both collaborative decomposition and Pareto-dominance,update and eliminate crowded individuals in the population archive,and maintain the diversity of the population.Experimental results on 18 Ma F instances and two engineering practice problems show that the population archive updating mechanism designed in this thesis can effectively improve the performance of Co DEA in solving irregular complex front problems,and obtain a more complete and uniform approximate front.