基于支持向量回归代理模型的气动力优化设计

Aerodynamic Optimal Design of Surrogate Models Based on Support Vector Regression

目前,气动力优化设计中通常基于经验风险最小化原则构建代理模型,预测精度的提高需要更多的训练样本,计算代价较大,同时盲目降低代理模型的训练误差难以避免过学习问题。针对上述问题,首先提出采用支持向量回归(SVR)方法基于结构风险最小化原则构建代理模型的思路,然后对测试函数和翼型阻力进行预测,最后对某型运输机机翼进行优化设计试验。结果表明:与其他代理模型对比,基于SVR的代理模型在小样本情况下具有较好的泛化能力,并且能够快速准确地预测气动特性,在飞机优化设计中,可以提高工作效率,优化结果可靠、可控。

The common surrogate models used in aerodynamic optimization are mostly established based on em- pirical risk minimization. There are two problems in this method. Firstly, the precision of prediction is highly dependent on sample population size, however the computational cost is large. Secondly, the over fitting prob- lem cannot be avoided due to reducing training error blindly. In order to solve the problems, the idea of building a surrogate model based on the principle of structural risk minimization with support vector regression（SVR） is proposed. An aerodynamic optimization is clone for a transport＇s wing based on SVR surrogate model. The method has better generalization ability and can avoid the complexity for high dimension problem, which is proved by comparing with other surrogate model in case of less samples. It indicates that the surrogate model by SVR method can predict the aerodynamic characteristics quickly and accurately, the efficiency of the aerodynam- ic optimization can be improved, and the optimal results are reliable and controllable.