A hierarchical clustering-based cooperative multi-population many-objective optimization algorithm

Abstract

The increasing number of objectives poses a great challenge upon many-objective optimization algorithms (MaOOAs) when solving many-objective optimization problems (MaOOPs), since it is rather difficult to obtain well-distributed solutions with tight convergence. To efficiently improve the ability of solving MaOOPs, this paper proposes a hierarchical clustering-based cooperative multi-population many-objective optimization algorithm (C2MP-MaOOA). Specifically, a hierarchical clustering-based population division strategy is proposed in C2MP-MaOOA, which is able to effectively optimize different regions of the Pareto front (PF) regardless of its shape, so as to maintain population diversity and accelerate convergence. Any single-objective optimizer can be applied in C2MP-MaOOA to optimize a subpopulation. To comprehensively evaluate the performance of C2MP-MaOOA, it was compared with eight state-of-the-art existing algorithms and two variants of C2MP-MaOOA on 63 MaOOPs selected from DTLZ, MaF, and WFG benchmark suites. The results indicate that C2MP-MaOOA has the best overall performance for each benchmark suite, which demonstrates that C2MP-MaOOA is quite competitive in solving MaOOPs.