Studies On The Evolutionary Algorithms For Complex Multi-objective Optimization Problems

Multi-objective optimization refers to optimize more than one conflicting objectives simultaneously,which is commonly seen in real-world and industrial applications.Hence it has important value to solve Multi-objective optimization problems effectively and efficiently.With the development of the industry,the complexity of multi-objective optimization problems is growing rapidly,and the existing algorithms cannot meet the actual demands.This thesis is aimed at proposing some methods for complex multi-objective optimization problems.As a heuristic random search method,the evolutionary algorithm has been widely used to solve multi-objective optimization problems for more than 20 years.However,in the face of some complex multi-objective optimization problems,the effect of the existing evolutionary algorithms is not satisfactory.For example,when handling many-objective optimization problems,it is difficult for evolutionary algorithms to balance between the convergence and diversity;when facing large-scale multi-objective optimization problems,the existing algorithms can hardly converge,while the computation cost of them are unbearable;when solving multi-objective optimization problems with complex Pareto solution sets,the existing algorithms also cannot approximate the whole Pareto front.Therefore,in this thesis we analyze the challenge of these problems and design relevant evolutionary algorithms based methods.The main study contents and contributions are summarized as follows:For many-objective optimization problems,in which more than three objectives are simultaneously optimized,we proposed a subregion division based evolutionary algorithm.In the proposed algorithm,the objective space is divided by the reference vectors to balance the convergence and the diversity.An effective mating selection strategy based on subregion division is also adopted for diversity maintenance.Experimental results have witnessed the significant advantage of our proposed algorithm.We also propose a framework to accelerate the computational efficiency of evolutionary algorithms on large-scale multi-objective optimization via problem reformulation.A bi-directional weight variable association strategy is adopted in the proposed framework.Experimental results have demonstrated that the proposed framework can significantly improve the computational efficiency of different algorithms.For the multi-objective problems with complicated Pareto sets,we propose a manifold learning based mating strategy to improve the performance of multi-objective evolutionary algorithms.Based on the idea of manifold learning methods,a method to compute manifold distances between solutions is proposed.The results of empirical studies have demonstrated that the proposed strategy can significantly improve the computational efficiency and performance of multi-objective evolutionary algorithms.