基于变可信度代理模型的气动优化

Aerodynamic optimization based on multi-fidelity surrogate

对于复杂的多设计变量优化设计问题,构造高精度的代理模型需要大量的高可信度样本,并带来巨大的计算量。利用较小的计算代价构建高精度的代理模型具有很强的工程意义。采用Co-Kriging方法,基于两组相互独立的高、低可信度模型样本,构建了一种高效的变可信度代理模型。Co-Kriging变可信度代理模型通过构建高、低可信度模型之间的关系模型,充分利用低可信度模型信息来提高代理模型整体的预测精度,在保证预测精度的前提下,大大减少了构造代理模型所需的计算时间。使用两个函数算例分析了不同选样方案对近似精度的影响,并用一个基于不同代理模型的RAE2822气动优化设计的对比证明了该方法在工程设计优化中的可行性。

Constructing surrogate models need lots of high-fidelity sample data in complex multivariable simulation-based design optimization problems, and the calculation is very costly. Thereby constructing higher accuracy surrogate model with lower computational cost has great engineering significance. In this paper Co-Kriging method is used to construct multi-fidelity surrogate model based on two independent sample datasets. Thus a greater quantity of low-fidelity information can be used to enhance the accuracy of the highfidelity surrogate model by formulating a form of correction process which reflects the differences between the low and high fidelity model, and the total computational cost of sample data can be greatly reduced. Two function examples demonstrated the influence of approximate accuracy of sample number, and the feasibility of this methodology in engineering design problem is validated by a comparison of different surrogate model based aerodynamic optimization design of RAE2822.