Trajectory Prediction of High-Mobility Targets Based on Self-Attention Mechanism and Bilstm

High maneuvering target trajectory has high complexity and uncertainty, and it is very easy to cause trajectory prediction accuracy to be difficult to satisfy due to model mismatch in the process of target tracking. To address this problem, a data-driven composite prediction method is proposed. Considering that the target high maneuvering trajectory is characterized by continuity, temporal sequence, and short duration, a convolutional neural network (CNN) and a bidirectional long and short time memory network (Bi-LSTM) are used to learn the spatial dimension features and temporal dimension features of the input data, respectively; the fusion of self-attention mechanism (AT) assigns weights to the outputs of the CNN-Bi-LSTM to dynamically adjust the degree of attention at different time steps according to the time series, thus focusing on the key information; using MAE as an indicator, the model hyperparameters are adjusted using the Dung Beetle Algorithm (DBO) to determine the optimal prediction model. The effectiveness of the proposed algorithm is illustrated by setting up comparative ablation experiments. The simulation results show that compared with the CNN model, the MAE of the proposed model is reduced by 44% on average, which can effectively improve the accuracy of high maneuvering trajectory prediction.