A comparative study of the evolutionary many-objective algorithms

The many-objective optimization problem (MaOP) is widespread in real life. It contains multiple conflicting objectives to be optimized. Many evolutionary many-objective (EMaO) algorithms are proposed and developed to solve it. The EMaO algorithms have received extensive attentions and in-depth studies. At the beginning of this paper, the challenges of designing EMaO algorithms are first summarized. Based on the optimization strategies, the existing EMaO algorithms are classified. Characteristics of each class of algorithms are interpreted and compared in detail. Their applicability for different types of MaOPs is discussed. Next, the numerical experiment was implemented to test the performance of typical EMaO algorithms. Their performance is analyzed from the perspectives of solution quality, convergence speed and the approximation of the Pareto front. Performance of different algorithms on different kind of test cases is analyzed, respectively. At last, the researching statuses of existing algorithms are summarized. The future researching directions of the EMaO algorithm are prospected.