A Two-Stage Multi-Evolutionary Sampling Optimization Framework for Expensive Optimization Problems
Juan Du, Hanyu Hu, Jian Wang
Abstract
Evolutionary Algorithms (EAs) are powerful global optimizers, but their application to real-world problems is hindered by computationally expensive fitness evaluations. While Surrogate-Assisted Evolutionary Algorithms (SAEAs) reduce this cost, designing efficient sampling strategies under a limited budget remains a key challenge. This paper proposes a Two-Stage MultiEvolutionary Sampling Optimization Framework (TSMESOF) to address this issue. The framework dynamically divides optimization into an initial exploration phase using two global search methods, followed by an exploitation phase that transitions to two local search methods to refine solutions. A novel global search method integrating opposition-based learning and competitive swarm optimization is employed in the early stage. TS-MESOF was comprehensively compared with five state-of-the-art SAEAs on seven benchmark functions and a real-world reservoir scheduling case. Results demonstrate that the proposed framework significantly outperforms the other algorithms, showcasing its effectiveness and potential for solving expensive optimization problems.