A Pseudo-GNSS Reconstruction Method of INS Based on the AT-DCLSTM

Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) together overcome their individual shortcomings, providing accurate navigation results in GNSS/INS integrated systems. However, the entire system positioning accuracy deteriorates over time during GNSS outages. In this paper, a pseudo-GNSS measurement is predicted by the combination of algorithm based on a dual channel long short-term memory (DCLSTM) neural network and attention mechanism (AT) model for aiding INS. Considering the differences in the propagation characteristics of longitude and latitude errors, the normal LSTM neural network is optimized by channel-wise to predict longitude and latitude errors respectively and get the pseudo-GNSS as the observation. In addition, the attention mechanism is used to improve the ability of LSTM network to capture the characteristics of long-term sequence information. Then the pseudo-GNSS measurement update model for Kalman Filter (KF) is established to incorporate these predictions into GNSS/INS integration during GNSS outages. Finally, experiments based on sea trials are implemented to evaluate the proposed method. Trial results show that the proposed method can effectively improve the performance of the integrated navigation system during GNSS outages.