A dual-population and multi-stage based constrained multi-objective evolutionary

초록

The existence of constrained multi-objective optimization problems (CMOPs) in real-world applications motivate researchers to focus more on developing constrained multi-objective evolutionary algorithms (CMOEAs). Due to the presence of constraints, an efficient constraint handling technique (CHT) is required in CMOEA to balance the constraint satisfaction and optimization of objective functions. Recently, different fitness based, ranking based, multi-population and multi-staged evolutionary approaches are proposed to handle CMOPs. However, most of the approaches still struggle while handling CMOPs with discontinuous feasible regions or whose feasible regions consist infeasible barriers. To overcome these issues, we propose a novel Dual-Population and Multi-Stage based Constrained Multi-objective Evolutionary Algorithm which is termed as CMOEA-DPMS. In CMOEADPMS, two populations are used to explore the search space and feasible regions. Along with two populations, an archive is also employed to store feasible, well converged and distributed solutions. To employ appropriate mating selection and environmental selection strategies according to the evolution of the populations, evolutionary process is divided into several stages. A strategy decider mechanism is proposed to determine the appropriate mating and environmental selections depending on the status of the population. In addition, a novel CHT named decomposition based constraint non-dominating sorting (DCDSort) is proposed by combining decomposition based selection with traditional constraint non-dominating sorting to maintain feasibility, convergence and diversity. The proposed algorithm is evaluated on five recent and popular test suites along with 36 realworld constrained multi-objective optimization problems against eight state-of-the-art algorithms. The empirical results suggests that CMOEA-DPMS is significantly superior or comparable to the considered algorithms and can tackle all kinds of CMOPs.(c) 2022 Elsevier Inc. All rights reserved.

키워드