A Dual-Stage Surrogate-Assisted Differential Evolution for Expensive Multimodal Optimization
提出一种双阶段代理辅助差分进化算法，早期用全局代理模型探索不确定区域，后期用局部代理模型精细搜索，在有限评估预算下高效定位多个全局最优解，适用于昂贵优化问题。
Expensive multimodal optimization problems present significant challenges due to the need to locate multiple global optima with high accuracy under strict evaluation budgets. To address this issue, this paper proposes a dual-stage surrogate-assisted differential evolution algorithm. In the early stage of the search, an adaptive phase-based global search strategy guided by a global surrogate model is employed. This strategy dynamically switches between exploring uncertain regions and exploiting areas near the current best seed, aiming to discover diverse global optima. In the later stage, a potential-then-exploit-based local search strategy based on multiple local surrogate models is introduced. It first perturbs each potentially promising solution via mutation and then applies surrogate-guided exploitation to enhance convergence. Experimental results on the CEC2013 multimodal benchmark suite and a real-world multiple competitive facility location and design problem demonstrate that the proposed algorithm significantly outperforms four state-of-the-art multimodal optimizers and five surrogate-assisted algorithms, especially under tight evaluation budgets.