The Research On Large-scale Multi-objective Evolutionary Algorithm Based On Solutions Generation Strategy

Large-scale multi-objective optimization problems(LMOPs)exist widely in engineering and scientific research,which bring significant challenges.LMOPs are specifically characterized by hundreds of decision variables,high-dimensional search space,and complex relationship between decision variables and objective variables,which leads to deteriorate dramatically of classical multi-objective evolutionary algorithms(MOEAs).Therefore,various evolutionary algorithms(EAs)have been proposed by researchers for addressing LMOPs.Even so,balancing the convergence and diversity of population is the focus and difficulty for solving complex LMOPs.Based on the analysis above,it is very urgent to design effective solution generation strategy to solve LMOPs by balancing the convergence and diversity of the population.The main work of this paper is as follows:(1)A dual-stage large-scale multi-objective evolutionary algorithm with dynamic learning strategy is proposed.In order to improve the efficiency of addressing LMOPs,the algorithm divides the entire evolutionary process into two optimization stages,and employs different solution generation strategies to optimize the population.The first optimization stage speeds up the convergence of the population,the proposed algorithm adopts decision variables clustering strategy to cluster decision variables into convergence-related and diversity-related variables,and optimizes each category of decision variables separately instead of optimizing all decision variables simultaneously.The second optimization stage mainly focuses on improving the diversity of the population,a dynamic learning strategy is designed to lead solutions learning from solutions with good diversity.In addition,the population information is adopted to dynamically update the learning parameter during the evolutionary process to prevent the population from falling into local optimum.The proposed algorithm is verified on three LMOPs benchmark,and the experimental results show that the algorithm effectively improves the efficiency of addressing LMOPs.(2)A large-scale multi-objective evolutionary algorithm based on importance rankings and information feedback is proposed.First,a novel grouping strategy is proposed to alleviate the issue of uneven grouping of decision variables,which uses variables characteristics and importance rankings to grouping decision variables.Decision variables with same characteristics and greater importance rankings tend to be clustered into a same group,and each group of decision variables is evolved separately to accelerate the population to converge to the true Pareto front.Second,a solution generation strategy based on information feedback is designed to adjust the diversity of population in objective space.The algorithm utilizes the historical information obtained in the evolutionary process to adaptively update the evolutionary direction,thereby ensuring the further enhancement of the diversity of the population.The performance of the proposed algorithm is verified on two LMOPs benchmark,the experimental results demonstrate that the algorithm achieves significantly competitive performance.(3)A multi-operator-based large-scale multi-objective evolutionary algorithm is proposed.Two global solution generation strategies are involved in the proposed algorithm to enhance the global exploration ability and local exploitation ability.In order to improve the convergence of the algorithm,the first solution generation strategy leads the population to evolve in multiple orientations by constructing orientation vectors,and prefers to retain the solutions with better convergence.The second solution generation strategy is based on competitive learning strategy that explores the entire true Pareto front,which avoids the algorithm falling into a local optimum.Moreover,the algorithm proposes a comprehensive indicator to more accurately evaluate the convergence and diversity performance of solutions,which improve the ability to screen solutions.The performance of the algorithm is demonstrated on one LMOPs benchmark,the experimental results show that the algorithm effectively enhances the convergence ability of the algorithm.