关键词: 高超声速滑翔飞行器, 轨迹预测, 多变量时间序列, 长短时记忆网络, 扩展卡尔曼滤波

Abstract:

To address the issue of inaccurate description of the motion state in the trajectory prediction process of hypersonic glide vehicles and the difficulty in effectively learning sequence features after multi-variable time series are input into the network, leading to unsatisfactory trajectory prediction accuracy, a trajectory prediction method based on multi-scale time series processing is proposed. Firstly, the maneuver state parameters are constructed as control quantities, and a maneuver prediction model is designed based on the dynamic equation. Secondly, a time series processing model is established, and multiple modules are introduced to conduct multi-scale analysis on the time series through convolution, projection and other operations. The bidirectional long short-term memory network is used to obtain the time series prediction of the state parameters. Finally, the trajectory prediction is achieved by combining the extended Kalman filter with the maneuver prediction model. Experimental results show that, compared with the existing trajectory prediction methods, this method has better comprehensive prediction performance.

Key words: hypersonic glide vehicles, trajectory prediction, multivariable time series, bidirectional long short-term memory network, extended Kalman filter