Behavior Analysis of Constrained Multiobjective Evolutionary Algorithms using Scalable Constrained Multi-Modal Distance Minimization Problems
Behavior Analysis of Constrained Multiobjective Evolutionary Algorithms using Scalable Constrained Multi-Modal Distance Minimization Problems
复制标题
DOI：
10.23919/wac55640.2022.9934365
复制
发表时间：
2022-10
期刊：
影响因子：
--
通讯作者：
Maaya Yano;Naoki Masuyama;Y. Nojima
中科院分区：
文献类型：
--
作者：
Maaya Yano;Naoki Masuyama;Y. Nojima
本文提出了可扩展约束多模态距离最小化问题，以针对现实世界优化问题中经常出现的多模态属性和约束来评估算法行为。我们之前的研究提出了二维约束多模态距离最小化问题（CMDMP），其中包括上述特征。本文将 CMDMP 扩展到可以定义任意数量决策变量的可扩展问题。它们可用于检查决策变量数量对约束多目标进化算法 (MOEA) 搜索性能的影响。在计算实验中，我们使用所提出的 CMDMP 评估了两种 MOEA，即 NSGA-II 和 DNEA，以及三种约束处理方法，即 CDP、IEpsilon 和 SP。
This paper proposes scalable constrained multimodal distance minimization problems to evaluate algorithm behaviors against a multi-modal property and constraints that often appear in real-world optimization problems. Our previous study proposed two-dimensional constrained multi-modal distance minimization problems (CMDMPs), which include the above characteristics. This paper extends CMDMPs to scalable problems which can define any number of decision variables. They can be used to examine the effects of the number of decision variables on the search performance of constrained multiobjective evolutionary algorithms (MOEAs). In computational experiments, we evaluate two MOEAs, i.e., NSGA-II and DNEA, and three constraint handling methods, i.e., CDP, IEpsilon, and SP, using the proposed CMDMPs.