Northwest Institute of Mechanical and Electrical Engineering
传统防空火控算法中的轨迹预测模型无法对复杂的无人机蜂群进行有效地轨迹预测，而现有针对无人机机动轨迹的预测研究往往只考虑单个无人机，且模型量级过大。为了准确且快速地预测无人机蜂群轨迹，本文提出一种面向蜂群的轨迹预测方法。在获得蜂群轨迹后，首先基于DBSCAN对其进行聚类，判断出蜂群中各个无人机的类别；然后基于分形算法，判断无人机轨迹是简单轨迹还是复杂轨迹；最后，用卡尔曼滤波进行简单轨迹的预测，用基于LSTM（Long Short-Term Memory）网络的方法进行复杂轨迹的预测。本文所提出的无人机蜂群轨迹预测方法的预测误差远远小于纯采用卡尔曼滤波方法预测的误差，且预测时间小于纯采用LSTM网络方法预测的时间,可以较为准确地预测蜂群中不同集群无人机的轨迹，并且兼顾了计算开销，为反无人机蜂群火控解算提供基础。
The trajectory prediction model of traditional weapon control algorithm can not effectively predict the complex trajectory of drones.Besides, the existing research on complex trajectory prediction usually consider only one UAV, with the huge amount of calculation. To predict the trajectory of drone swarm quickly and accurately, a method of trajectory prediction for UAVs is proposed. After obtaining the track of drone swarm, it is firstly clustered based on DBSCAN method; Then the trajectory is determined to be simple or complex. Finally,Kalman Filter prediction was used for simple trajectory prediction, while LSTM（Long Short-Term Memory）network was used for complex trajectory prediction. The proposed model is more precise compared with the model using Kalman Filter, as well as consume less computing time compared with the model completely using LSTM network,which can meet the real-time requirement, as well as provide scientific basis for anti-UAV swarm calculation.