基于改进反向传播算法的跨音速攻角补偿修正研究

跨音速飞行时攻角传感器误差显著增大，严重影响飞机的正常飞行。针对飞机试飞与正式装备时攻角传感器采取不同配置的特点，以及传统反向传播(back-propagation, BP)算法的不足，基于改进BP算法中Levenberg Marquardt (LM)算法,设计了一种跨音速攻角补偿修正算法。利用某型飞机的实际试飞数据对基于LM算法的BP神经网络进行了训练与测试,结果表明，经BP神经网络补偿修正后的攻角能够基本消除跨音速段原始测量攻角的剧烈波动，并与真实攻角吻合效果好。

Research on compensation-correction of transonic angle-of-attack based on improved back-propagation algorithm

In condition of transonic flight, the performance of the angle of attack (AOA) sensor declines remarkably which influences the normal flight seriously. Aiming at the different configurations of AOA sensors when the plane is in flight-test and in equipment and the shortage of tranditional back-propagation (BP) algorithm, an algorithm based on improved BP algorithm (Levenberg Marquardt algorithm, LM algorithm) is designed to compensate and correct the transonic AOA. The BP neural network based on LM algorithm has been trained and tested with actual flight data. The results show that the AOA, which is corrected by the BP neural network as compensation, could basically eliminate the violent fluctuations of the measured AOA in transonic phase. And its tendency coincides with the true AOA with a small error.