- Research Article
- 10.1109/tetci.2026.3670690
A Review of Surrogate-Assisted Evolutionary Multi-Objective Optimization
- Jan 01, 2026
- IEEE Transactions on Emerging Topics in Computational Intelligence
- Qing Wang + 6 more +6
Expensive multi-objective optimization problems (EMOPs) are widely encountered in science and engineering, but solving them often requires expensive evaluation costs. To alleviate the computational burden, surrogate-assisted multi-objective evolutionary algorithms (SAMOEAs) have emerged as an efficient paradigm by integrating evolutionary search with surrogate models. This survey provides a systematic overview of the field from multiple perspectives. First, EMOPs are classified based on their attributes, including the structure of objective space, Characteristics of decision variables, dynamics and constraints, and augmented optimization methods. Second, we analyze the core algorithmic components, including surrogate model management, infill criteria, and other acceleration mechanisms. Furthermore, we highlight the design of SAMOEAs tailored for special scenarios, such as multimodal optimization, heterogeneous objectives, and inverse modeling. In addition, applications across benchmark problems and real-world domains are also reviewed. Finally, we outline future promising research directions, including the development of more efficient and robust surrogate models, parallel and distributed frameworks, LLM-assisted SAMOEAs, privacy-preserving SAMOEAs and theoretical analysis of SAMOEAs. This survey aims to establish a systematic perspective on SAMOEAs, bridging methodological advances with practical challenges, and offering a roadmap for future developments.
Read more