基于Bi-LSTM 的无人机轨迹预测模型及仿真

传统轨迹预测模型存在模型简化较大、考虑因素较少等问题。结合飞行轨迹连续性、时序性、交互性 的特点,提出基于双向长短期记忆(Bi-LSTM)神经网络的轨迹预测模型,将入侵者的位置、姿态和两机的相对 信息同时作为轨迹预测模型的输入,更加符合真实轨迹变化规律;对建立的基于 Bi-LSTM 的轨迹预测模型采 用综合考虑动量和速度的自适应调整学习步长的学习算法进行训练;并与基于 Elman神经网络的轨迹预测模 型进行仿真对比分析。结果表明:与基于 Elman神经网络的轨迹预测模型相比,所提模型在不同方向预测200 个点的平均绝对误差不超过4m,三维预测效果更优,可以较为准确地进行轨迹预测。

UAV trajectory prediction simulation for autonomous collision avoidance

Aiming at the problems of traditional trajectory prediction methods, such as large model simplification and less consideration, combined with the characteristics of flight trajectory continuity, time series and interactivity, a trajectory prediction method based on bidirectional long short term memory neural network is proposed. The position, heading, pitch, roll and relative information of the intruder UAV are simultaneously used as the input of the trajectory prediction model, which is more in line with the true trajectory change law. A Bi-LSTM-based trajectory prediction model is established, which can simultaneously learn the implicit information in the forward and backward trajectories, and adopt the adaptive learning rate learning algorithm to train the model. The simulation results show that compared with the Elman neural network, the average absolute error of the model predicted by 200 points in different directions is less than 4m, and the 3D prediction effect is better, and the trajectory prediction can be performed more accurately.