A multi-step CNN-LSTM high-speed craft trajectory prediction method based on parametric modeling

In air defense operations, accurate prediction of high-speed craft trajectory is the core of achieving effective interception. Aiming at the insufficient processing capability of traditional methods for complex maneuvers and environmental disturbances, and the limitations of long short-term memory network (LSTM) in dynamic interaction feature exploration and prediction step adaptation, this paper proposes a Convolutional Neural Network-Long Short-Term Memory Network (CNN-LSTM) that integrates parameterized modeling and feature association enhancement. The model achieves space-time joint modeling of multi-dimensional features by constructing parametric motion features and combining CNN and LSTM, and then completes the accurate prediction of high-speed craft trajectory information. Experiments show that after offline training, the model demonstrates high-precision single-step prediction of target maneuvering trajectory; in the multi-step prediction scenario (step size k=15), considering the data error caused by electromagnetic interference and other complex environments, its prediction results and evaluation indexes are significantly better than those of LSTM model and traditional filtering model, which effectively verifies the model's robustness and prediction effectiveness.