基于LSTM深度神经网络的飞行机动识别方法研究

Research on Flight Maneuver Recognition Method Based on LSTM Deep Neural Network

飞行机动是空中博弈任务OODA环成型的重要组成部分,涉及态势观察、战术决策、飞行员控制执行等多项关键技术。面向近距飞行任务中的机动识别问题,开发了一种基于长短期记忆深度神经网络的机动识别方法。该方法能够有效降低机动识别对领域知识的依赖性,实现了对近距飞行任务中5类机动的自动识别。基于高逼真飞行模拟仿真环境采集的飞行数据对机动识别方法性能进行测试,结果表明,提出的机动识别方法在5类近距飞行机动平均识别准确率为95%,平均识别准确率超过基于随机森林与支持向量机的基线机动识别方法,证明了提出的方法在解决近距飞行任务机动识别任务的有效性与先进性。

Flight maneuvering is an important part of the OODA loop formation of the air game mission,involving many key technologies such as situation observation,tactical decision-making,and pilot control execution.For the problem of ma-neuver recognition in short-range flight missions,a maneuver recognition method based on LSTM(Long Short Term Memo-ry)deep neural network has been developed.This method can effectively reduce the dependence of maneuver recognition on domain knowledge and achieve automatic recognition of the five main types of maneuvers in short-range flight missions.The performance of the maneuver recognition method has been tested based on flight data collected in a high-fidelity flight sim-ulation environment.The results show that the proposed maneuver recognition method have achieved an average maneuver recognition accuracy of 95%in five types of close flight missions,and the average recognition accuracy exceeds the baseline maneuver recognition method based on random forest and support vector machine,proving the effectiveness and advancement of the method in solving maneuver recognition tasks in short-range flight missions.