.png){kind=icon}
.png){kind=icon}
Return
A constrained many-objective evolutionary algorithm with learning vector quantization-based reference point adaptation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2023.101359.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Constrained many-objective optimization problems (CMaOPs) are frequently encountered in real-world ap-plications, generally having constrained Pareto fronts (PFs) that are often incomplete and irregular in high-dimensional objective space. Due to the advantage of reference point-based decomposition algorithms on unconstrained many-objective optimization, they have been widely applied into dealing with CMaOPs. However, these algorithms may not be well suited for solving the constrained problems with irregular PFs, since their adopted uniformly distributed reference points may not properly fit the distribution of feasible regions. To address this issue, we propose a constrained many-objective evolutionary algorithm with reference point adaptation. To be specific, the learning vector quantization network is used to collect the feasibility information during the evolutionary search process in the supervised manner. By regarding the feasible and infeasible solutions as two classes of training samples, the network can gradually model the topological structure of the PF and adaptively generate the reference points in the feasible regions. In order to make full use of infeasible solutions to assist in the modeling of the PF, an adaptive e-truncation based constraint-handling technique is proposed to introduce the elite infeasible solutions during the evolutionary process. Experimental results on several public constrained many-objective optimization test suites demonstrate the competitiveness of the proposed algorithm in comparison with seven state of-the-art algorithms.
Keywords:
Constrained many-objective optimization
Evolutionary algorithm
Reference point adaptation
Learning vector quantization
Journal
IF:
8.5
Papers:
2.1K
Citations:
1.0W