航空学报 > 2020, Vol. 41 Issue (5): 623352-623352   doi: 10.7527/S1000-6893.2019.23352

飞行器气动外形数值优化与设计专栏

一种适用于气动优化的高效自适应全局优化方法

李春娜, 张阳康   

  1. 西北工业大学 航天学院空天飞行技术研究所, 西安 710072
  • 收稿日期:2019-08-08 修回日期:2019-08-25 出版日期:2020-05-15 发布日期:2019-09-27
  • 通讯作者: 李春娜 E-mail:chunnali@nwpu.edu.cn
  • 基金资助:
    国家自然科学基金(11502209)

An efficient adaptive global optimization method suitable for aerodynamic optimization

LI Chunna, ZHANG Yangkang   

  1. Aerospace Flight Vehicle Design Key Laboratory, School of Astronautics, Northwestern Polytechnical University, Xi'an 710072, China
  • Received:2019-08-08 Revised:2019-08-25 Online:2020-05-15 Published:2019-09-27
  • Supported by:
    National Natural Science Foundation of China (11502209)

摘要: 随着设计空间的增大和优化问题非线性程度的提高,基于代理模型的优化(SBO)过程收敛越来越慢,并且在局部勘测上呈现不足。本文发展了一种高效自适应全局优化方法,在整个样本细化迭代过程中采用变设计空间取样:即在每一步样本细化迭代过程中,利用当前设计空间中的样本建立代理模型,并且根据样本的内部特征,利用模糊聚类算法将该设计空间分割成几个子空间,然后在每个子空间内通过最大化目标函数的期望提高函数和最小化模型预测目标来增加新的样本,之后对子空间进行融合更新设计空间。6个解析测试算例的结果表明,所发展的方法相比于一般的代理模型优化方法,具有更好的鲁棒性以及全局探索和局部勘测能力,更适用于具有强非线性和多极值的优化问题。RAE2822气动优化实例表明,所发展的方法在处理工程实际问题时,仍然能够保持很好的效率、鲁棒性和自适应性。

关键词: 气动优化, 变设计空间, 模糊聚类算法, 代理模型, 自适应

Abstract: With the increase of design space and nonlinearity, the Surrogate-Based Optimization (SBO) process converges more slowly, and shows deficiency in local exploitation. This paper proposes an efficient adaptive global optimization method, of which infill samples are selected within a variable design space. In each refinement cycle, the current design space is divided into several subspaces by a fuzzy clustering algorithm, with respect to the inherent characteristics of samples in the current design space. Thus new infill samples are generated in each of the subspaces by maximizing expected improvement function and minimizing surrogate prediction, and the subspaces are then merged to form a new design space. The proposed method is validated by six analytical tests. In comparison with general SBO method, the proposed method shows better robustness and performance in global exploration and local exploitation, which is suitable for optimization problems with strong nonlinearity and many optima. The application by minimizing drag of RAE2822 airfoil indicates the proposed method performs well in solving engineering problems, and can maintain good efficiency, robustness and adaptability.

Key words: aerodynamic optimization, variable design space, fussy clustering algorithm, surrogate model, adaptive

中图分类号: