Research On Evolutionary Algorithms For Solving Many-Objective Optimization Problems

Many-objective optimization problems(MaOPs)widely exist in the field of engineering applications scientific research and scientific research,and have become a research hotspot in the field of intelligent information processing.In recent years,various many-objective evolutionary algorithms have been proposed for MaOPs with discontinuous,degenerate,discrete,irregular Pareto front and constrained conditions,including the evolutionary algorithms based on convergence priority and uniformly distributed reference points.However,when dealing with MaOPs of irregular Pareto front and constraints,it often leads to a variety of problems,such as easy to lose part of the target area,the population easily converges to a certain sub-area of the Pareto front,and the diversity of the population becomes worse and so on.In order to further improve the performance of the algorithm,for MaOPs with irregular Pareto front,a manyobjective evolutionary algorithm with diversity-first based environmental selection is proposed.For MaOPs constraints,a decomposition-based evolutionary algorithm with reference vector adaptation for constrained many-objective optimization is proposed.The major work and contribution in this dissertation includes the following two aspects.(1)For the existing many-objective evolutionary algorithm usually adopts the environmental selection strategies with convergence-first,which leads to the population easily converge into a subregion of the Pareto front,a many-objective evolutionary algorithm with diversity-first based environmental selection(MaOEA-DES)is proposed.The environmental selection procedure of MaOEA-DES adopts a diversityfirst-and-convergence-second principle,which first selects the representative solutions that having better diversity and then considers using the well-converged solutions to replace them in subregions.This selection replacement strategy can maintain the diversity and make contribution to the convergence.Meanwhile,a selection criterion,termed adaptive angle penalized distance,is designed to judge whether the replacement is implemented or not.The proposed algorithm is compared with five state-of-the-art many-objective evolutionary algorithms on a large number of test problems with various characteristics.Experimental studies demonstrate that the proposed algorithm has competitive performance on many-objective optimization problems.(2)For the problems that the many-objective evolutionary algorithm based on reference point to solve the constrained MaOPs will lead to the population diversity variation,a decomposition-based evolutionary algorithm with reference vector adaptation for constrained many-objective optimization(C-MOEA/RVA)is proposed.Since the distribution of uniform reference points is not consistent with the distribution that of Pareto front,so that the distribution of the solution set is poor.C-MOEA/RVA uses a set of adaptive reference points to guide the generation of the solution,so as to improve the distribution of the population in the feasible region.C-MOEA/RVA uses Learning Vector Quantization(LVQ)to generate a set of reference points.The reference points generated by LVQ can well learn the topological shape of real Pareto front.Then,an adaptive constraint processing technique is introduced in the environmental selection process based on reference vector adaptation,in which the proportion threshold of infeasible solution can be adaptive adjusted according to the population evolution process,and the excellent infeasible solution is introduced dynamically to participate in the population iteration.Furthermore,a method of combining reference points was constructed to delete uniform reference points in the feasible region and retain LVQ reference points,so as to enhance the distribution of population in the feasible region.In order to verify the effectiveness of C-MOEA/RVA,the proposed algorithm is compared with five other algorithms were experimentally tested on the existing constraint test problems in experimental research.The experimental results show that C-MOEA/RVA has good versatility in constrained MaOEAs with different types of Pareto front.