关键词: 代理模型, 元启发算法, 高维度, 计算昂贵优化问题, 探索与开发

Abstract: High-dimensional and expensive optimization problems are widely present in fields such as energy and resource optimization, urban and environmental planning, industrial design and manufacturing, aerospace, communication and information, etc. The increasing of dimension leads to an expansion of the search space, while the computational expense limits the times that the true solution can be evaluated. The aforementioned reasons render traditional optimization algorithms ineffective. Surrogate-assisted meta-heuristic algorithms, which use surrogate models to replace the expensive real fitness evaluations and utilize meta-heuristic algorithms to guide the optimization direction, can significantly reduce computational time and cost while maintaining optimization accuracy. This paper, focusing on the characteristics of high-dimension and expensive computation in optimization problems, organizes recent literature on surrogate-assisted meta-heuristic algorithms from six perspectives： generating initial sample points, constructing and updating surrogate models, using evolutionary algorithms, balancing exploration and exploitation, designing adaptability, and practical applications. The paper summarizes how surrogate-assisted meta-heuristic algorithms address these two major challenges. Finally, it proposes future research directions for aspects that are currently under-researched.

Key words: surrogate model, meta-heuristic algorithm, high-dimension, computationally expensive optimization problems, exploration and exploitation