基于注意力机制的高超声速飞行器LSTM智能轨迹预测

临近空间高超声速飞行器具有非惯性轨迹形式和大范围、强机动的突防能力，对目标飞行轨迹的准确预测能够为反导拦截系统有效拦截提供有力技术支持。针对高超声速飞行器的滑翔式和跳跃式飞行轨迹预测问题，提出一种基于注意力机制的Seq2Seq轨迹预测模型，利用LSTM网络设计编码器和解码器，同时利用注意力机制提取的信息进行解码预测。该网络以目标轨迹的位置、速度、弹道倾角和攻角六维特征序列作为输入网络，网络输出为未来一段时间内的连续轨迹序列，利用弹道仿真模型获得的目标飞行器轨迹数据作为训练集对网络进行训练与优化。实验结果表明，该网络能够对高超声速飞行器的多种飞行轨迹进行有效的轨迹预测，预测误差小，能够为反导拦截系统提供有利参考。

LSTM Intelligent Trajectory Prediction for Hypersonic Vehicles Based on Attention Mechanism

The near-space hypersonic vehicle has a non-inertial trajectory form and large-scale and strong maneuvering penetration capabilities.The accurate prediction of the target's flight trajectory can provide strong technical support for the effective interception of the missile interception system.To address the problem of glide and skiptrajectory prediction of hypersonic aircrafts, this paper proposes a Seq2Seq trajectory prediction model based on attention mechanism, which employsthe LSTM network to design the encoder and decoder, and uses the information extracted by attention mechanism to performdecodingand prediction. The network takes the six-dimensional feature sequence of the target trajectory's position, velocity, trajectory inclination angle and attack angle as the input network, and the continuous trajectory sequence during a certain period in the futureis the network output. The trajectory data of the target aircraft obtained by the trajectory simulation model is used as the training setto train and optimize the network. The experimental results show that the proposed network can effectively predict the various flight trajectories of hypersonic aircrafts with small prediction errors, which can provide some insights into the missile interception system.