摘要
提出了一种基于深度学习的翼型气动系数预测方法,有效克服了以往方法依赖翼型设计参数以及算法复杂度随预测精度的提高呈指数级增长等缺点。首先,介绍了卷积神经网络(CNN)的基本原理、网络机构以及训练方法,给出了训练样本数、批量大小、批次数量、迭代次数、循环次数的关系;其次,设计了针对翼型图像处理的CNN结构,随机选择6000个样本对该网络进行了训练;最后,对561个翼型的法向力系数进行了预测,并与部分参数法方法的预测结果进行了比较。仿真结果表明,提出的图形化预测方法具有很高的预测精度。
To prevent the dependence of prediction methods on design parameters and the exponential increase of algorithm complexity with increasing prediction accuracy,an aerodynamic coefficient prediction method of airfoils based on deep learning is proposed.First,the fundamental theory,network structure and training method of Convolutional Neural Networks(CNN)are introduced.Then,according to the characteristics of airfoil image processing,the structure of CNN model is designed and the parameters are trained by 6000 random samples.Finally,the normal force coefficients of 561 airfoils are predicted and compared with those prediction of some other parameterization methods.The simulation results show that the proposed graphical prediction method has high prediction accuracy.
作者
陈海
钱炜祺
何磊
CHEN Hai;QIAN Weiqi;HE Lei(China Aerodynamics Research and Development Center,Mianyang 621000,China)
出处
《空气动力学学报》
CSCD
北大核心
2018年第2期294-299,共6页
Acta Aerodynamica Sinica
基金
国家自然科学基金(11532016)
中国博士后科学基金项目(2015M582810)
关键词
深度学习
卷积神经网络
翼型
气动系数
预测
回归
deep learning
Convolutional Neural Networks(CNN)
airfoil
aerodynamic coefficient
prediction
regression
