.png){kind=icon}
Return
A domain transform-based evolutionary algorithm for noisy multi-objective optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2026.116300.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Noisy multi-objective optimization problems are prevalent in real-world applications, where noise in objective evaluations can distort dominance relations and mislead evolutionary search. To address this challenge, we propose a Domain Transform–based Non-dominated Sorting Genetic Algorithm II (DTNSGA-II), which is built on a dual-population, dual-domain cooperative framework specifically designed to balance noise suppression and diversity preservation. The key rationale behind DTNSGA-II is that noise predominantly manifests as high-frequency components in objective evaluations. Accordingly, a transform-domain population applies domain transform–based denoising via the Fast Fourier Transform to attenuate noise-dominated components, thereby providing more reliable fitness information and guiding the search toward stable Pareto fronts. In parallel, an original-domain population retains raw objective values to preserve population diversity and prevent excessive smoothing. Furthermore, a dynamic resampling mechanism for non-dominated solutions and an anti-over-smoothing strategy in later evolutionary stages are introduced to enhance robustness under varying noise levels. Experimental results demonstrate that DTNSGA-II consistently outperforms classical and state-of-the-art algorithms across a range of noisy scenarios. This study confirms the effectiveness of domain transform–based denoising in multi-objective evolutionary optimization and highlights the potential of cross-domain mechanisms for solving noisy optimization problems.
Keywords:
Noisy multi-objective optimization
Noisy evolutionary algorithm
Domain transform
Noisy problem
Journal
IF:
8
Papers:
5.7K
Citations:
3.5W
Organization
Cited Papers
No cited papers available