摘要
针对尖楔前体(类乘波体)飞行器用嵌入式大气数据传感(FADS)系统存在建模困难及解算模型精度低的问题,首先采用BP神经网络建模代替传统的FADS系统空气动力学建模的方法,建立含有双隐含层的四层神经网络模型,然后通过合理选择网络结构参数及训练验证,对FADS系统的攻角进行解算,最后对驻点压力对算法精度的影响进行研究。结果表明,本文建立的含有双隐含层的四层神经网络模型精度较高,攻角测试误差小于0.25°;驻点压力是否作为输入参数对FADS系统神经网络算法求解精度影响较大,攻角测试误差相差达0.1°。在飞行器前缘半径允许的情况下,应尽量得到驻点压力用于解算攻角,提高解算精度。
In allusion to the difficuhy in the modeling of the flush air data sensing （FADS） system applied to the vehicle with sharp wedged fore-bodies （quasi-waverider vehicle） and its low solving accuracy, firstly, the neural network architecture with two hidden layers is designed and performed based on a back propagation neutral network algorithm replacing traditional theoretical aerodynamic model of the FADS system. Secondly, in connection with the characteristics of the FADS system applied to the vehicles with sharp wedged fore-bodies, angle of attack is solved based on the neural network model by choosing appropriate network architecture parameters and related training verification test. Finally, systematic comparison for angle of attack is researched whether treating the stagnation pressure as an input parameter. Numerical simulation results show that the precision of the neural network architecture with two hidden layers we have established in this paper is accurate enough to satisfy the demands, and the testing error for angle of attack is less than 0.25°. Therefore, the neural network model can replace the aerodynamic model to solve angle of attack. The stagnation pressure can greatly affect the accuracy of the neural network model. And the testing error can reach 0.1°whether treating the stagnation pressure as an input parameter. Allowing for the radius of the sharp wedged fore-bodies, the stagnation pressure should be used as an input parameter to solve the angle of attack and enhance the solving precision.
出处
《宇航学报》
EI
CAS
CSCD
北大核心
2016年第9期1072-1079,共8页
Journal of Astronautics
基金
航天十一院自主创新研发项目(80000800029921171206)
关键词
嵌入式大气数据传感(FADS)系统
尖楔前体
BP神经网络
驻点压力
攻角
Flush air data sensing （FADS） system
Sharp wedged fore-bodies
Back propagation neural network
Stagnation pressure
Angle of attack
