A fast parallel and multi-population framework with single-objective guide for many-objective optimization

Abstract

A large number of objectives pose challenges to many-objective evolutionary algorithms (MOEAs) in terms of diversity, convergence, and complexity. However, the majority of MOEAs are not able to perform well in all three aspects at the same time. To tackle this issue, this paper proposes a fast parallel and multi-population framework with single-objective guide for many-objective optimization. The general framework is able to enhance diversity via sub-populations and maintain convergence by the information sharing between sub-populations. The proposed framework is implemented on three representative MOEAs and is compared with original MOEAs on 64 many-objective benchmark problems. Experimental results show that the proposed framework is capable of enhancing the performance of original MOEAs with satisfactory convergence, diversity, and complexity.