Surrogate-Assisted Differential Evolution With Search Space Tightening for High-Dimensional Expensive Optimization Problems

Abstract

High-dimensional expensive optimization problems (HEOPs) have posed significant challenges to current surrogate-assisted differential evolution algorithms (SADEs) because of the curse of dimensionality. To enhance the optimization efficiency and solution accuracy for HEOPs, Surrogate-assisted differential evolution with search space tightening (SADE-SS) is proposed in this article. There are three main contributions in SADE-SS: first, a novel parameter adaptation strategy is incorporated into the framework of SADE to improve its scalability by leveraging information from approximated fitness values. Second, a search space tightening strategy is proposed to strengthen the local exploitation capacity by identifying promising local search spaces. Third, a switching strategy is proposed to manage the global and local surrogate-assisted searches, aiming to balance exploration and exploitation capacities. Experiments on expensive benchmark functions with dimensions ranging from 30 to 400 were conducted to verify the effectiveness of SADE-SS for HEOPs. Moreover, ablation experiments were conducted to validate each proposed component. Comprehensive experimental results demonstrate that SADE-SS can secure highly competitive performance over state-of-the-art SAEAs for HEOPs.