尖楔前体飞行器FADS系统的神经网络算法

摘要

With respect to the difficulty in the modeling of the Flush Air Data Sensing System (FADS)for vehicles with sharp wedged fore-bodies and low solving precision of the model, applications of the artificial neural network algorithm for FADS system to sharp wedged fore-bodies were investigated in this paper.Back-propagation (BP)neural network model were set up to replace traditional aerodynamic model of the FADS system.Regarding the characteristics of the FADS system for vehicles with sharp wedged fore-bodies,neural network architecture with single hidden layer and double hidden layers were designed and performed on the basis of reasonable validation testing and structural parameters of network. The comparison was systematically analyzed between the testing error distributions of these two models.Flight parameters such as angle of attack,angle of sideslip,the free stream static pressure,and the Mach number were determined according to the network algorithm.Numerical simulation results show that the developed BP neural network algorithm has good accuracy for the vehicle with sharp wedged fore-bodies.Moreover,the accuracy of the neural network model with double hidden layers is higher than that of the network model with single hidden layer.%对人工神经网络算法在尖楔前体飞行器用嵌入式大气数据传感系统(Flush Air Data Sensing System,FADS)中的应用进行了探讨.针对该FADS系统存在的建模困难及解算精度低的问题,采用BP神经网络算法代替传统的空气动力学模型,通过合理选择网络结构参数及训练验证,分别建立了FADS系统的含有单隐含层的三层网络模型及含有双隐含层的四层网络模型,对攻角、侧滑角、自由来流静压及马赫数等参数进行求解.数值仿真结果表明,建立的用于尖楔前体飞行器的FADS系统的神经网络算法求解精度较高,且含有双隐含层的网络模型精度优于单隐含层的模型精度.