Enhancing the reliability of shipborne INS/GNSS integrated navigation system during abnormal sampling periods using Bi-LSTM and robust CKF

Developing a reliable navigation system is a crucial challenge for intelligent ships. Integrating Inertial Navigation Systems (INS) with shipborne Global Navigation Satellite Systems (GNSS) provides a viable solution. However, the system's reliability heavily relies on GNSS availability. This paper proposes a method to enhance the reliability of shipborne INS/GNSS integrated navigation systems during GNSS abnormal sampling periods. The proposed method consists of two parts. Firstly, using specially designed input and output parameters, a pseudoGNSS prediction model based on a Bidirectional Long Short-Term Memory Recurrent Neural Network (Bi-LSTM) is constructed to generate predicted pseudo-GNSS measurements during periods of GNSS loss, effectively filling in the gaps left by the absence of actual GNSS data. Secondly, Huber's M-estimation-based Robust Cubature Kalman filter (RCKF) is integrated into the data fusion algorithm, and a switching mechanism between RCKF and the Cubature Kalman filter (CKF) is established based on the source of GNSS data utilized for measurement update, to mitigate the impact of non-Gaussian distribution prediction noise introduced by pseudo-GNSS prediction model. Experimental validation using real ship navigation data in a natural marine environment confirms the proposed method's capability to maintain navigation accuracy during GNSS loss for shipborne INS/GNSS integrated navigation systems. The algorithm's superiority becomes more evident as the duration of GNSS loss increases, underscoring its effectiveness in enhancing shipborne integrated navigation system reliability.