Trajectory prediction algorithm based on maneuver analysis for hypersonic glide vehicles in a deep neural network

A long time series prediction of hypersonic glide vehicles (HGVs) is urgently needed to fulfill the increasing defense demand since trajectory prediction is crucial for fight intention inference, threat assessment, and vehicle interception in near space. To address this issue, a trajectory prediction algorithm based on maneuver analysis under a neural network is proposed for HGVs. From a dynamic perspective, trajectory prediction can be mapped as a time series prediction by maneuvering parameters. First, a new set of maneuvering parameters is extracted from the maneuvering mode analysis, which aids the fast Fourier transform (FFT) method in the trajectory classification implementation. Subsequently, the equilibrium glide trajectory can be forecasted by employing a novel trajectory modeling algorithm based on the autoregression method. Moreover, the trajectory with a large number of maneuvers that exceeds the proper threshold value is addressed via the error compensation method and the intelligent prediction algorithm. The error compensation method eliminates the disadvantages arising from the limited window length but low frequency of the trajectory data under periodic variation and the anisotropy of the prediction capacity in different dimensions. Finally, the advanced self-attention and distilling mechanism and generative style encoder implementations are introduced to improve the prediction capacity in the Transformer framework, which ensures the time sensitivity of the improved prediction algorithm to meet the defense purpose. The extensive flight scenarios of the HGVs demonstrate that the novel algorithm outperforms the existing trajectory prediction algorithms.