Relation model-assisted multi-region evolutionary algorithm for expensive constrained optimization

doi:10.1016/j.eswa.2026.131977

Relation model-assisted multi-region evolutionary algorithm for expensive constrained optimization

2026-03-14

0

PRE

AI

Y

Yuxi Sun Zhichao Huang

DOI：10.1016/j.eswa.2026.131977

Original

Original request for help

Save

Abstract

En 中文

Expensive constrained optimization problems (ECOPs) are inherently challenging due to complex landscapes, stringent feasibility requirements, and severely limited evaluation budgets. This article proposes a relation model-assisted multi-region evolutionary algorithm (RM-MREA). By partitioning the search space into multiple regions, RM-MREA leverages region-specific relation models to guide global exploration toward promising areas, which in turn directs regression model-assisted local exploitation more effectively. Extensive experiments on real-world constrained problems demonstrate that RM-MREA consistently outperforms state-of-the-art methods in solution quality, feasibility, and robustness. These results highlight that a well-guided global exploration not only identifies feasible regions efficiently but also substantially enhances the performance of local exploitation, yielding superior performance in complex constrained landscapes.

Keywords:

Expensive constrained optimization

Multi-region evolutionary algorithm

Relation model

Global exploration

Local exploitation

Relation model-assisted multi-region evolutionary algorithm for expensive constrained optimization

Abstract

Journal

Organization