Utilizing the Relationship Between Unconstrained and Constrained Pareto Fronts for Constrained Multiobjective Optimization

• Published in: IEEE transactions on cybernetics Vol. 53; no. 6; pp. 3873 - 3886
• Main Authors: Liang, Jing, Qiao, Kangjia, Yu, Kunjie, Qu, Boyang, Yue, Caitong, Guo, Weifeng, Wang, Ling
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.06.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Constrained multiobjective optimization; Constraints; evolutionary algorithm; Evolutionary algorithms; Evolutionary computation; Genetic algorithms; Learning; Linear programming; Multiple objective analysis; objective function; Optimization; Pareto optimization; Populations; relationship; Search problems; Shape; Sociology; Statistics
• ISSN: 2168-2267; 2168-2275; 2168-2275
• DOI: 10.1109/TCYB.2022.3163759

Constrained multiobjective optimization problems (CMOPs) involve multiple objectives to be optimized and various constraints to be satisfied, which challenges the evolutionary algorithms in balancing the objectives and constraints. This article attempts to explore and utilize the relationship between constrained Pareto front (CPF) and unconstrained Pareto front (UPF) to solve CMOPs. Especially, for a given CMOP, the evolutionary process is divided into the learning stage and the evolving stage. The purpose of the learning stage is to measure the relationship between CPF and UPF. To this end, we first create two populations and evolve them by specific learning strategies to approach the CPF and UPF, respectively. Then, the feasibility information and dominance relationship of the two populations are used to determine the relationship. Based on the learned relationship, specific evolving strategies are designed in the evolving stage to improve the utilization efficiency of objective information, so as to better solve this CMOP. By the above process, a new constrained multiobjective evolutionary algorithm (CMOEA) is presented. Comprehensive experimental results on 65 benchmark functions and ten real-world CMOPs show that the proposed method has a better or very competitive performance in comparison with several state-of-the-art CMOEAs. Moreover, this article demonstrates that using the relationship between CPF and UPF to guide the utilization of objective information is promising in solving CMOPs.