Evolutionary many-objective optimization: A survey

Many-objective optimization problems (MaOPs) widely exist in industrial and scientific fields, where there are more than 3 objectives that are conflicting with each other (i.e., the improvement of the performance in one objective may lead to the deterioration of the performance of some other objectives). Because of the conflict between objectives, there is no unique optimal solution for MaOPs, but a group of compromise solutions need to be obtained to balance between objectives. As a class of population-based optimization algorithms inspired by biological evolution principles evolutionary algorithms have been proved to be effective in solving MaOPs, and have become one of the research hot spots in the field of multi-objective optimization. In the past 20 years, the research on many-objective evolutionary algorithms (MaOEAs) has made great progress, and a large number of advanced evolutionary methods and evaluation systems have been proposed and improved. In this paper, the research progress of evolutionary many-objective optimization (EMaO) is comprehensively reviewed. Specifically, it includes: (1) Describing the relevant theoretical background of EMaO; (2) Analyzing the problems and challenges faced by evolutionary algorithms in solving MaOPs; (3) Discussing the development of MaOEAs in detail; (4) Summarizing MaOPs and performance indicators in detail; (5) Introducing the visualization tools for high-dimensional objective space; (6) Summarizing the application of MaOEAs in some fields, and (7) Providing suggestions for future research in the domain.