Hypersonic Vehicle Trajectory Prediction Algorithm Incorporating Multiple Attention and Coding-Decoding Structures

Near-space hypersonic vehicles are characterized by fast flight speed, good maneuverability and strong surprise capability, which make it a great challenge to predict and intercept such targets. Aiming at the problem of insufficient mining of semantic information and spatial structure information among features in the trajectory prediction process, a trajectory prediction algorithm integrating the multi-attention mechanism and the encoder-decoder structure based on LSTM is proposed. Firstly, the kinematic analysis of the hypersonic vehicle is carried out to construct the vehicle trajectory database under six-dimensional features, and then the encoder and decoder fusing the multi-head attention mechanism and LSTM neural network are designed. On this basis, the time-step data is changed to make the network output the trajectory sequence in the future period. The experimental results show that the model improves 29% and 13% in the two metrics of MAE and MAPE compared with similar methods, and realizes the effective prediction of multiple flight trajectories of hypersonic vehicles.