基于广义回归神经网络的积冰后三维机翼阻力增量预测方法  

Prediction of Wing Drag Coefficient in Icing Conditions Based on Generalized Regression Neural Network

作　　者：田牧垠 苏媛[1] 

机构地区：[1]北京航空航天大学航空科学与工程学院,北京100191

出　　处：《科技创新导报》2016年第22期14-18,共5页Science and Technology Innovation Herald

摘　　要：针对现有的积冰后阻力系数预测模型与实验数据误差较大,而且无法满足三维机翼的预测需求的问题,提出了一种基于广义回归神经网络的改进预测方法。基于现有的积冰理论和实验结果,分析了可能影响积冰外形的环境参数,并针对三维机翼的特点对上述参数进行修正,从中抽取了影响最大的一组影响因素作为网络的输入。为了达到最优的结果,采用交叉验证的方法预估平滑参数。仿真结果表明,该方法对于无后掠机翼的预测结果的精确度显著高于现有的Bragg,Gray,HPC模型,同时对于有后掠的机翼依旧能保持较高的精度。For the large erro rs between exis ting iced wing drag coefficient prediction models and results of experiment, animproved prediction method used generalized regression neural network, also known as GRNN, was developed. Based on the previous research resul ts, those critical icing condition parameters werechosen and adjusted to 3D ice accretion. With all the parameters defined, they became the input of the neural network. In order to obtain the best fitting result , K-fold cross validation was used to determine the smoothness parameter. The preliminary resul ts show that the calculated drag coefficient matched well wi th experiment data from various sources and this method performed bet ter than existing model.

关 键 词：飞机积冰 飞行力学 广义回归神经网路 阻力预测 

分 类 号：V24[航空宇航科学与技术—飞行器设计]