A Novel Self-Learning GNSS/INS Integrated Navigation Method

GNSS / INS integrated navigation is widely used in dynamic navigation and positioning. However, in the case of GNSS signal failure for a long time, the accuracy of GNSS / INS integrated navigation system will drop sharply due to the error accumulation of inertial components over time, and the anti-jamming ability and reliability can’t be guaranteed. In order to solve this problem, this paper proposes a navigation algorithm with self-learning function, which combines data denoising and Long Short-Term Memory (LSTM) neural network in the traditional GNSS / INS integrated navigation. The general process of the algorithm is that when the GNSS signal is available, the velocity, yaw of INS and preprocessed IMU output are used as the input of LSTM model, the output is the position increment of GNSS. When GNSS signal is lost, the preprocessed IMU data and INS information are input into LSTM model to generate pseudo GNSS position and send it to extended Kalman filter (EKF) to correct INS navigation results. In order to obtain reliable and accurate positioning information during GPS outages, this paper proposes a data pre-processing method combining empirical mode decomposition (EMD) and wavelet threshold filtering (WTF) to process the original IMU raw measurement. First, the EMD adaptively decomposes the noisy IMU signal into a series of intrinsic mode functions (IMFs) according to amplitude and frequency. Then, the wavelet threshold filtering is applied to high-frequency IMFs, which separates the useful information in the high-frequency IMFs. Wavelet threshold filtering using heuristic threshold and soft threshold function. Finally, these IMFs are added with IMFs of low frequency and residual signal to achieve de-noising signal. The experimental results show that the proposed navigation algorithm can significantly improve the accuracy and reliability of GNSS / INS integrated navigation during GNSS outages.