一种基于双向GRU的UAV飞行轨迹预测方法  

A Method for UAV Flight Trajectory Prediction Based on Bidirectional GRU

作　　者：张宗腾 张琳[1] 汪文峰[1] 滕飞 张搏[1] ZHANG Zongteng;ZHANG Lin;WANG Wenfeng;TENG Fei;ZHANG Bo(Air and Missile Defense College,Air Force Engineering University,,Xi'an 710000,China;No.95835 Unit of PLA,Malan 741000,China)

机构地区：[1]空军工程大学防空反导学院,西安710000 [2]中国人民解放军95835部队,新疆马兰741000

出　　处：《电光与控制》2022年第3期11-15,26,共6页Electronics Optics & Control

基　　金：中国博士后科学基金(2017M623417)。

摘　　要：由于三维轨迹是一个具有连续性和交互性的复杂时间序列,因此,针对无人机飞行轨迹预测问题,结合深度学习理论特点,提出了一种基于双向门控循环单元的无人机飞行轨迹预测方法,进一步提高了轨迹信息的利用率。首先,建立无人机飞行动力模型,仿真获得不同状态的飞行轨迹样本;其次,利用均方误差作为损失函数,确定了双向门控循环单元轨迹预测模型的隐藏层节点参数和迭代次数;最后,利用Adamax算法对双向门控循环单元模型进行优化,实现了无人机飞行轨迹的预测。实验结果表明,双向门控循环单元模型在X,Y,Z轴方向上预测结果的平均绝对误差均在5.0 m内,且轨迹预测平均用时约4.2 ms,与循环神经网络、门控循环单元相比,其预测效果更佳,具有良好的应用价值。Three-dimensional trajectory is a complex time series with continuity and interaction.Aiming at the problem of UAV flight trajectory prediction and by using the deep learning theory a method for UAV flight trajectory prediction based on bidirectional Gated Recurrent Unit(GRU)is proposed which can further improve the utilization rate of trajectory information.Firstly the UAV flight dynamics model is established and the flight trajectory samples in different states are obtained by simulation.Secondly the Mean Square Error(MSE)is used as the loss function to determine the parameters of hidden nodes and iterations of the bidirectional GRU model.Finally the Adamax algorithm is used to optimize the bidirectional GRU model for realizing the prediction of the UAV flight trajectory.Experimental results show that:1)The Mean Absolute Errors(MAE)of the prediction results of the bidirectional GRU model in the X Y and Z axis directions are all within 5.0 m and the average time for trajectory prediction is 4.2 ms approximately;and 2)Compared with RNN and GRU models our method has better prediction effect.Thus it will have fine application value.

关 键 词：无人机 轨迹预测 双向GRU 飞行仿真 时间序列 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]