基于循环神经网络的飞行机动识别方法及仿真研究

A Flight Maneuver Recognition Approach and Simulation Study Based on Recurrent Neural Network

飞行机动识别技术在飞行训练考核、飞行数据挖掘等需要对飞行数据进行评估、理解的领域有重要的应用价值。针对现有飞行机动识别方法过于依赖领域知识的问题,考虑飞行数据的时序特征,提出一种基于序列模式识别的飞行机动识别的方法,实现飞行机动的实时自动识别。针对基于普通前向神经网络、循环神经网络和双向循环神经网络的飞行机动识别算法,通过仿真实验进行了识别效果对比分析。结果表明,循环神经网络方法能够识别出飞行数据中的序列模式,在降低领域知识依赖的同时,达到较高的识别精度。

Flight maneuver recognition technology has great application value in the field of flight training assessment,flight data mining,etc.To solve the problem that the existing methods rely too heavily on domain knowledge,a flight maneuver recognition method based on sequential pattern recognition is proposed considering the temporal characteristics of flight data,which is able to achieve real-time automatic recognition.Simulation experiments are conducted to compare the performance of three algrithms based on nerual network,recurrent nerual network and bi-direction recurrent nerual network,respectively.The results show that,the proposed method is able to identify sequential patterns in flight data.While reducing the dependence on domain knowledge,it performs higher recognition accuracy.