Abstract: The experiment design technology of orthogonal array and latin cube, and the approximation models of polynomial response surface, kriging model and the radial basis function neural network were separately used to approximate the mapping relationship between the design variables and the performance parameters for the wings’ optimization. Considering the cost and accuracy of the approximation, the surrogate model based on radial basis function neural network was selected to replace the high precision aerodynamic analysis model to participate in the aircraft wings’ multidisciplinary optimization to realize the disciplines approximation and reduce the operation cost.

Key words: Aircraft wing; Collaborative optimization; Surrogate model; Aerodynamic calculation

摘要： 文章采用正交表、拉丁方等实验设计方法，结合多项式响应面、Kriging 模型、径向基函数神经网络等近似技术，分别拟合了飞行器翼面优化所需的设计变量与性能参数的映射关系。综合考虑了计算成本和精度后，确定以基于神经网络的代理模型来代替气动学科的高精度分析模型，参与飞行器翼面的多学科优化，从而实现了各学科的模型近似并降低了运算成本。

关键词: 飞行器翼面；协同优化；代理模型，气动计算