基于POD-BPNN的热启动策略及其在气动代理优化中的应用

传统气动优化设计需要大量CFD 分析，而代理优化（SBO）方法能够有效降低CFD 分析次数，但该方法并没有改变单个CFD 分析时间。提出一种基于本征正交分解-反向传播神经网络（POD-BPNN）模型的热启动策略，并应用于气动代理优化。使用POD-BPNN 模型对SBO 中的初始样本建立从几何设计变量到流场数据的预测模型...

The hot-start strategy based on POD-BPNN and its application in aerodynamic shape optimization using surrogate-based approach

Traditional aerodynamic optimization design requires multiple time-consuming CFD analyses; while widely-used surrogate-based optimization (SBO) can effectively reduce the number of CFD analyses, but it cannot speed up a single CFD analysis. Hence, this paper proposes a hot-start strategy using the proper orthogonal decomposition-back propagation based neural network (POD-BPNN), and applies it in surrogate-based aerodynamic optimization. First, the POD-BPNN model is built from geometric design variables to flowfield data through the initial samples in an SBO. Then, during iteration of the SBO, the flowfield of a new sample is predicted by the built model, and is used as the initial flowfield of the hot-start CFD analysis for the new sample. At the same time, the new sample is used to update the POD-BPNN model until the end of the optimization. The results of the aerodynamic optimization design of the transonic airfoil indicate: the hot-start strategy using the POD-BPNN can reduce the time of a single CFD analysis by 68%, and improves the efficiency of the SBO by 37%.