节点文献

多学科设计优化代理模型技术的研究和应用

【作者】 穆雪峰

【导师】 姚卫星;

【作者基本信息】 南京航空航天大学 , 飞行器设计, 2004, 硕士

【摘要】 代理模型是多学科设计优化中一种包含试验设计和近似方法等多项内容的建模方法,它通过在系统中用数学模型来替代原有分析模块的方式,实现了分析模块与MDO计算框架的集成,使复杂系统的多学科设计优化变的切实可行。本文系统分析了多学科设计优化方法中代理模型的构造过程,比较了几种相关的试验设计方法和数学近似方法。通过对比分析,我们认为要构造代理模型就必须引入能够大大减少试验次数的试验设计方法和能够解决高维非线性问题的近似方法。为此本文中分别介绍了正交试验,均匀试验以及中心复合试验等几种试验设计方法,以及多项式响应面,Kriging,和径向基函数等几种数学近似方法。并且通过构造描述机翼展向升力分布的代理模型,对上述几种方法作了对比分析。算例的结果表明,试验设计方法的选取与要解决问题的规模直接相关。同时计算结果还显示,Kriging模型和径向基函数模型作为两种通用模型,可以很好的解决一般的高维非线性代理模型的构造问题;另外,从算例问题的物理本质出发所构造的广义多项式响应面模型也能够满足精度要求。

【Abstract】 Surrogate model is an important component of Multidiscipline Design Optimization (MDO). A surrogate model is created with experiment design methods and approximation approaches, and it can be easily integrated into a MDO Framework. This article describes the development process of surrogate models and introduces some experiment design methods and approximation approaches that can be used for a MDO surrogate model, they are Full Factorial Experiment Design, Orthogonal Experiment Design, Uniform Experiment Design, Central-Composite Experiment Design, and Polynomial Response Surface method, Kriging method, Radial Basis Function method and Artificial Neural Network. At the end of this article, a surrogate model is created to describe the lift distribution of a wing. Result shows that Kriging method and RBF method are two useful approximation approaches, especially for high dimension problem. In addition, this article created a generalized polynomial response surface model, and result shows that it can be also used to describe the lift distribution.

  • 【分类号】V221.92
  • 【被引频次】84
  • 【下载频次】1642
节点文献中: 

本文链接的文献网络图示:

本文的引文网络