Research On Reference Vector Guided Many-objective Evolutionary Optimization

Multi-objective optimization is widely used in engineering applications and daily life,and multi-objective optimization problems are difficult to be solved with increasing number of objectives.Many-objective optimization problems are the multi-objective optimization problems involving more than three objectives.In recent years,evolutionary algorithms for solving many-objective optimization problems have attracted more and more research interest.In view of the current problems of many-objective optimization,this thesis proposes several many-objective evolutionary algorithms to improve the ability of evolutionary algorithms to solve many-objective optimization problems:(1)This thesis proposes a diversity ranking based many-objective evolutionary algorithm DREA.To address the issue that it is difficult to manage diversity when solving many-objective optimization problems,DREA proposes a diversity ranking method.The method can solve various shapes of Pareto fronts and obtains good diversity on different test problems.In addition,a reference vector adaptation method is proposed to search the range of Pareto front to assist diversity ranking method to manage population.The experimental results have demonstrated that the proposed algorithm can well solve many-objective optimization problems,and the diversity ranking method has been shown the effectiveness by embedding it into other algorithms.(2)This thesis proposes an adaptative reference vector based many-objective evolutionary algorithm ARVEA.To address the issue that the algorithms are sensitive to the Pareto front shapes when solving many-objective optimization problems,ARVEA proposes a reference vector adaptation method.The method introduces the hyperplane and adopts a set of predefined reference vector and archive to update reference vectors,which can adapt different shapes of Pareto fronts.The adapted reference vectors are introduced to partition the population and the ASF is adopted as the selection criterion in each subpopulation to balance the convergence and diversity of population.The empirical results have shown that the proposed algorithm can well solve many-objective optimization problems,which has better performance than the most many-objective evolutionary algorithms.(3)This thesis proposes an R2 indicator and reference vector based many-objective evolutionary algorithm R2-RVEA.To address the problem that the most many-objective evolutionary algorithms have poor versatility on solving optimization problems,R2-RVEA proposes a many-objective evolutionary algorithm combining decomposition method and performance indicator by analyzing the merit of decomposition method and performance indicator.The algorithm adopts Pareto dominance to select the non-dominated solutions to guide the evolution of population,it will further introduce population partition strategy and R2 indicator selection strategy to manage the diversity when the number of non-dominated solutions is greater than population size.The experimental results have shown that the proposed algorithm has good performance in solving different types of many-objective optimization problems.