Abstract: Hypersonic glide vehicles (HGVs) have characteristics of wide strike range, fast flight speed and strong penetration ability, which bring severe challenges to accurate HGV trajectory prediction.To improve HGV trajectory prediction accuracy, a HGV trajectory prediction method based on time-domain convolution network (TCN) and bidirectional long short-term memory network (BiLSTM) was proposed. This method used causal dilation convolution of TCN to extract multi-scale dynamic features of HGV trajectory, and integrated bidirectional circulation mechanism of BiLSTM. It realized deep fusion of spatial correlation features by mining long-term dependence and context correlation of the trajectory. Finally, the method mapped prediction results to sample space through fully connected layer. Meanwhile, by introducing combination optimization mode of Bayesian optimization (BO) and grey wolf optimization (GWO), the method realized global optimization of network hyperparameters and established a HGV trajectory prediction model under deep learning framework.Numerical simulation results show that with complete training, the established prediction model can effectively predict future position state of HGV. Compared with four comparison models, the root mean square error of the proposed prediction model decreases by an average of 62.10%, and the average absolute error decreases by an average of 61.66%.