Set-based Many-objective Evolutionary Optimization Integrating Preferred Regions

Many-objective optimization problems(MaOPs) are very common and important in real-world applications. Because of MaOPs involving a large number of objectives, it is very difficult to solve. At present, MaOPs is one of hot topics in the community of evolutionary optimization. Set-based evolutionary optimization based on the performance indicators is one of the effective methods to solve MaOPs. In addition, integrating a decision-maker's preferences into solve MaOPs can enhance the selection pressure and guide the evolution of a population. Thus, in this thesis, a set-based evolutionary optimization algorithm integrating preferred regions is proposed to solve MaOPs.First, a set-based many-objective evolutionary optimization integrating preferred regions for MaOPs is presented. In the set-based evolution, the preferred region of a high dimensional objective space is dynamically determined based on an achievement scalarizing function, and the framework of set-based many-objective optimization integrating preferred regions is put forward. The proposed method of determining preferred regions is applied to a benchmark optimization problem, and the experimental results verify its rationality and effectiveness.Secondly, the above framework of set-based many-objective optimization integrating preferred regions is specifically implemented, and a set-based many-objective optimization algorithm based on comparison and crossover operators on sets guided by preferred regions is presented. In this method, a comparison strategy on sets by combining the Pareto dominance on sets with preferred regions is designed, and the crossover operators on sets guided by preferred regions are developed to produce a Pareto front with superior performances. The proposed algorithm is used to solve several benchmark MaOPs. The experiments show its superiority by comparing with the other two typical methods.Then, on the basis of the above comparison and crossover operators on sets guided by preferred regions, the guidance of preferences to mutation operators on sets is further taken into account, and a set-based many-objective optimization algorithm based on mutation operator guided by preferred regions is presented. Under the framework of set-based evolution, the current generation's optimal solution set located in preferred region is taking as reference, and the population's mutation is guided toward it by an adaptive gaussian disturbance. The proposed algorithm is used to solve several benchmark MaOPs, and the experiments verify its superiority.Finally, by taking a car cab design as an example, the practical MaOP is analyzed. In addition, the proposed evolutionary optimization methods are applied into the practical engineering design, and the experimental results empirically verify their performance in feasibility and efficiency.The above research results enrich the theory of many-objective evolutionary optimization, and provide feasible approaches for solving MaOPs.