为了解决优化设计中计算效率与高可信度信息获取之间的矛盾,从高、低可信度模型的物理机理出发,基于Kriging模型和拉丁超立方设计选样方法构造两模型差值的代理模型;并利用该代理模型对低可信度模型进行修正,构成了具有高可信度的修正模型.与直接对高可信度模型构造的代理模型相比,修正模型不但分析精度更高,而且所需的构造样本更小.文中分别以翼型气动力分析、机翼气动力分析和无人机隐身特性分析为例,从不同维数、不同学科的角度验证了修正模型特性,并进行了机理分析.%We construct a new and better metamodel, called by us LMIM (low-fidelity model improving method) model, by applying Kriging method and Latin Hypereube Design (LHD). It improves the fidelity of low-fidelity model utilizing the difference between high- and low-fidelity models. Sections 1 and 2 explain the construction of our metamodel and its application to three examples. Subsection 1.1 briefs LHD and subsection 1.2 briefs the Kriging model. Subsection 2.1 is entitled framework of LMIM metamodel; Fig. 1 shows its framework. Subsection 2.2 is enfified airfoil aerodynamic LMIM metamodel; Table 1 gives this example's design space; Figs. 3 through 6 and Table 2 give its calculation results. Subsection 2. 3 is entitled wing aerodynamic LMIM metamodel; Figs. 7 through10 and Table 3 give its calculation results. Subsection 2.4 is entitled unmanned aerial vehicle radar cross section LMIM metamodel; Figs. 11 and 12 and Table 4 give its calculation results. Subsection 2.5 is entitled the mechanism analysis of LMIM metamodel; Figs. 13 and 14 give its calculation results. Section 3, based on the calculation results of the three examples and their analysis, gives the two following preliminary conclusions: (1) LMIM metamodel has greater computational efficiency than high-fidelity model; (2) it has higher fidelity and smaller database than the respective ones of traditional metamodel.
展开▼
