Gradient-based Kriging approximate model and its application research to optimization design  

作　　者：XUAN Ying XIANG JunHua ZHANG WeiHua ZHANG YuLin 

机构地区：[1]College of Aerospace and Material Engineering,National University of Defense Technology,Changsha 410073,China

出　　处：《Science China(Technological Sciences)》2009年第4期1117-1124,共8页中国科学（技术科学英文版）

基　　金：Supported by the National High Technology Research and Development Program of China ("863" Program)

摘　　要：In the process of multidisciplinary design optimization, there exits the calculation complexity problem due to frequently calling high fidelity system analysis models. The high fidelity system analysis models can be surrogated by approximate models. The sensitivity analysis and numerical noise filtering can be done easily by coupling approximate models to optimization. Approximate models can reduce the number of executions of the problem's simulation code during optimization, so the solution efficiency of the multidisciplinary design optimization problem can be improved. Most optimization methods are based on gradient. The gradients of the objective and constrain functions are gained easily. The gra- dient-based Kriging (GBK) approximate model can be constructed by using system response value and its gradients. The gradients can greatly improve prediction precision of system response. The hybrid optimization method is constructed by coupling GBK approximate models to gradient-based optimiza- tion methods. An aircraft aerodynamics shape optimization design example indicates that the methods of this paper can achieve good feasibility and validity.In the process of multidisciplinary design optimization, there exits the calculation complexity problem due to frequently calling high fidelity system analysis models. The high fidelity system analysis models can be surrogated by approximate models. The sensitivity analysis and numerical noise filtering can be done easily by coupling approximate models to optimization. Approximate models can reduce the number of executions of the problem’s simulation code during optimization, so the solution efficiency of the multidisciplinary design optimization problem can be improved. Most optimization methods are based on gradient. The gradients of the objective and constrain functions are gained easily. The gradient-based Kriging (GBK) approximate model can be constructed by using system response value and its gradients. The gradients can greatly improve prediction precision of system response. The hybrid optimization method is constructed by coupling GBK approximate models to gradient-based optimization methods. An aircraft aerodynamics shape optimization design example indicates that the methods of this paper can achieve good feasibility and validity.

关 键 词：KRIGING APPROXIMATION FUNCTION gradient-based APPROXIMATE model hybrid OPTIMIZATION method MULTIDISCIPLINARY design OPTIMIZATION 

分 类 号：O224[理学—运筹学与控制论]