.png){kind=icon}
.png){kind=icon}
Return
A surrogate-assisted differential evolution for high-dimensional expensive constrained optimization problems with mixed-integer variables
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.126729.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recently, surrogate-assisted evolutionary algorithms (SAEAs) received a lot of attention due to their excellent performance in handling expensive constrained optimization problems (ECOPs). However, most of them can only be used for solving problems that are low-dimensional and with only continuous variables. Therefore, a surrogate-assisted differential evolution for high-dimensional ECOPs with mixed-integer variables (SADE-HDMI) is proposed in this paper. Firstly, a Multiple Local Extremum based Sampling (MLES) method is designed, in which two sampling strategies focusing on constraints and objective functions are utilized alternatively based on iterative information, so that the feasible region and high-quality feasible solutions can be efficiently located. Secondly, a Diverse Population Generation Operation for Mixed-Integer Variables (DPMI) is devised to avoid the population from falling into a local optimal region, where the diversity of the population is maintained by selecting solutions with more diversity and limiting the number of solutions with the same integer variables in the population. Convergence and diversity can be well balanced under the help of these two operations. Finally, the performance of SADE-HDMI is validated on fifteen benchmarks and a real-world optimization problem. The optimization results demonstrate that SADE-HDMI can locate feasible solutions with 100% probability on these 16 problems, and it is superior to or similar to other three state-of-the-art algorithms on 15 out of 16 problems.
Keywords:
Differential evolution
Expensive constrained optimization problems
Surrogate-assisted evolutionary algorithms
High-dimensional
Mixed-integer
Multiple local extremum-based sampling
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W