A GNSS Positioning Algorithm Assisted by LSTM Neural Network and EKF

Abstract

In Global Navigation Satellite System (GNSS) positioning, the Kalman filter algorithm can effectively utilize the dynamic model characteristics and observation data of the mobile carrier to estimate the dynamic characteristics of the mobile carrier. However, due to the constantly changing motion state of the carrier, there is a mismatch between the preset dynamic model and the actual motion model, resulting in some deviation between the model and the real motion state. To address this issue, this paper proposes a GNSS positioning algorithm based on a Long Short-Term Memory (LSTM) neural network-assisted Extended Kalman Filter (EKF). The LSTM neural network tracks and learns the change in dynamic model error in the Kalman filter. The predicted error is used to correct the Kalman filter in order to improve its performance and achieve higher positioning accuracy. Simulation results show that compared with the algorithm using only the Kalman filter for positioning, the proposed algorithm improves the accuracy of the x, y and z directions by 48%, 13% and 34% respectively. In addition, the positioning results are more stable, having effectively suppressed the error divergence caused by the Kalman filter and significantly improved the positioning accuracy.