.png){kind=icon}
Return
Surrogate-assisted many-objective evolutionary optimization with angle and shift-based density estimation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.133513.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
When solving expensive many-objective optimization problems (MaOPs), the performance of existing surrogate-assisted evolutionary algorithms is usually limited by the adopted evolutionary frameworks. To make further progress in addressing expensive MaOPs, this paper integrates a new evolutionary paradigm based on angle and shift-based density estimation with Kriging models, and proposes a novel surrogate-assisted evolutionary algorithm called KAnD. In KAnD, Kriging models are constructed using the historically truly evaluated solutions in the database to approximate the original expensive MaOP. Then, we embed a nondominated-solution-led strategy into evolutionary search to drive the population evolution toward better convergence on the approximate problem. After that, we quantify the diversity and convergence contributions of each solution to the database via an angle indicator and a shift-based density estimation indicator, respectively. The solutions with the best Pareto optimality in terms of the two indicators are selected for expensive fitness evaluations and then stored into the database. The above process ensures both diversity and convergence in high-dimensional objective spaces with limited computational resources. Furthermore, by introducing a constraint-oriented mechanism, KAnD is extended to solve constrained expensive MaOPs, resulting in an extended algorithm named CKAnD. Extensive experiments on several benchmark test suites demonstrate the superiority of KAnD and CKAnD over state-of-theart methods.
Keywords:
Expensive many-objective optimization
problem
Evolutionary algorithm
Kriging model
Angle
Shift-based density estimation
Journal
IF:
7.5
Papers:
3.0W
Citations:
10.2W
Organization
Cited Papers
No cited papers available