.png){kind=icon}
.png){kind=icon}
Return
Distance-aware cooperative Kriging-assisted evolutionary algorithm for expensive many-objective optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2026.102473.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In recent years, surrogate-assisted evolutionary algorithms (SAEAs) have received growing attention for addressing expensive many-objective optimization problems. However, the surrogate models commonly employed in these methods often fail to balance global trends with local details, leading to considerable prediction errors, especially near Pareto front boundaries and in highly complex regions. Meanwhile, most infill sampling strategies in existing SAEAs cannot simultaneously capture the distribution characteristics of promising regions and the convergence potential of the solution set, resulting in a lack of adaptive guidance during the search process. These limitations lead to low search efficiency and unsatisfactory performance of SAEAs, particularly for complex expensive many-objective optimization problems. To address these issues, this paper proposes a distance-aware cooperative Kriging-assisted evolutionary algorithm for expensive many-objective optimization (DACKEA). In DACKEA, a distance-aware cooperative Kriging model is designed. This model adaptively integrates a global Kriging model with several local ones via the proposed distance-aware weighting strategy, thereby enhancing prediction accuracy and robustness. Furthermore, a two-stage infill sampling criterion guided by multi-source information is developed to select promising solutions for model update, which comprehensively utilizes the distribution information, convergence potential, and prediction uncertainty of candidate solutions. This criterion employs a stage-switching mechanism driven by the rate of population convergence improvement, enabling adaptive transitions between stages to achieve precise responses to different optimization states and efficient allocation of limited computational resources. Extensive experiments on three benchmark suites and two real-world applications demonstrate that the proposed DACKEA is highly competitive against other state-of-the-art SAEAs in solving expensive many-objective optimization problems.
Journal
IF:
8.5
Papers:
2.1K
Citations:
1.0W