An Underwater Multisource Fusion Anomaly Detection Navigation Algorithm Based on Factor Graph and LSTM

The innovative methodology of factor graph optimization (FGO) offers a promising avenue for integrating sensor data in the domain of integrated navigation for underwater vehicles, particularly autonomous underwater vehicles (AUVs). In complex underwater environments, the large-angle oscillations of the AUV and surface waves can cause the sensors on the AUV to generate observation anomalies, which can negatively affect the positioning accuracy and robustness of the AUV. This article proposes a novel anomaly detection factor graph method based on the improved stochastic outlier selection (SOS) for the combined inertial navigation systems (INSs)/Doppler velocity log (DVL)/pressure sensors (PSs) navigation process. The method constructs an INS/DVL/PS integrated factor graph model embedded with anomaly detection nodes and an anomaly-based long short-term memory (LSTM) prediction processing model for the anomalies. In addition, to rectify the position errors of the AUV on a periodic basis, the position state estimation is carried out based on the window historical data through the LSTM prediction module, thereby improving the positioning accuracy of the AUV. AUV surface experiments validate the robustness and accuracy of the method in detecting and rectifying anomalies in the combined INS/DVL/PS navigation observations of underwater vehicles.