Maneuvering trajectory prediction of air combat targets based on self attention mechanism and CNN LSTM

DOI:	10.11809/bqzbgcxb2023.07.028
Keywords:	maneuvering trajectory prediction; analysis of air combat data; multi level time series; Self Attention; multi step trajectory prediction
Abstract:	An air combat target maneuver trajectory is a multi dimensional time series with rich temporal and spatial characteristics, and high degrees of complexity and uncertainty. At present, establishing kinematic models for trajectory prediction is very difficult. Besides, it is also difficult for time series prediction methods to extract temporal and spatial features, and only single sequential training can be achieved from T to T+1. In this view, this paper proposes model CNN LSTM ATT, which combines Self Attention (ATT) mechanism with Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM). This model is trained offline, and the obtained optimal model can realize high precision prediction of target maneuver trajectories. Compared with CNN LSTM and LSTM models for single step prediction, this model has a good single step prediction performance and different overload maneuvering prediction abilities. Considering the errors and missing of the transmitted data caused by electromagnetic interference and complex environment, the 5 step target trajectory prediction is carried out, and the prediction results and evaluation indexes are better than those of CNN LSTM and LSTM models.
Issue:	Vol. 44 No. 7 (2023)
Published:	2023-07-28
PDF HTML