An Underwater SINS/DVL Integrated System Outlier Interference Suppression Method Based on LSTM-EEWKF
Shuaishuai Zhang, Tao Zhang, Liang Zhang
Abstract
At present, strapdown inertial navigation system (SINS) and Doppler velocity logs (DVLs) integrated navigation is a common way for autonomous underwater vehicle (AUV) to conduct underwater operations. The complex and changeable Marine environment easily leads to unknown noise and outliers in the measurement information of DVL. When the DVL is disturbed by outliers, the positioning accuracy will be reduced. In order to solve this problem, a long short-term memory extended exponential weighted Kalman filter (LSTM-EEWKF) algorithm assisted by long short-term memory (LSTM) neural network is proposed. When the DVL works normally, the SINS/DVL integrated navigation system is iteratively learned by the LSTM neural network. When the DVL is disturbed by outliers, the trained LSTM neural network is first used to provide correction information for the system, and then, the extended exponential weighted Kalman filter (EEWKF) algorithm is used to make the system continue to work. The EEWKF algorithm realizes weight allocation for different historical moments by introducing an exponential weighting coefficient, which reduces the weight of the estimator when the SINS/DVL integrated system is disturbed by outliers. It can effectively reduce the impact of outliers and historical data on the positioning results of integrated navigation. The simulation and Yangtze River experiment show that the proposed algorithm has higher positioning accuracy than several hybrid algorithms. It can suppress outliers more effectively and make the system more robust.