基于POD-RBFN降阶模型的串列叶栅流场预测

通过传统实验或CFD 手段获取流场信息的方法往往需要耗费大量资源或时间，这在需要快速获取大量流场信息时产生的成本是无法接受的，发展比传统CFD 更快速的流场预测方法具有重要意义。采用本征正交分解（POD）方法对样本流场进行模态分解，提取流场的主导模态；而后采用径向基函数神经网络（RBFN）响应POD 基函数的系数，实现流场降阶预测模型的构建，并在模型中采用基于函数响应偏差的自适应抽样方法；通过某串列叶栅非定常流场数据对预测模型进行验证。结果表明：本文构建的POD-RBFN 混合模型可以快速准确地预测出串列叶栅的流场参数分布；与静态采样相比，本文采用的自适应采样方法在采样效率上表现出明显优势，同样重构精度所需的样本数降低了25% 左右。

Tandem Cascade Flow Prediction by POD-RBFN Reduced Order Model

The cost of obtaining a large amount of flow field structure by traditional experiment or CFD is unacceptable, so it is significance to develop faster forecasting calculation methods. Proper Orthogonal Decomposition (POD) was used in this paper to extract the dominant mode of the tandem flow field. Radial Basis Function Network (RBFN) is used to respond to the coefficients of the POD basis functions to realize the construction of the reduced-order prediction model of the flow field. Then the adaptive sampling method was developed for the reduced order model. The prediction model is verified by the unsteady flow field data of a cascade cascade. It is concluded that the hybrid method can be utilized to accurately predict the aerodynamics parameters and flow field of tandem cascade. Compared with static sampling, the number of samples required for adaptive sampling to achieve the same reconstruction accuracy is reduced by about 25%.