An effective LS-SVM/AKF aided SINS/DVL integrated navigation system for underwater vehicles

In order to combat the severity of the impact of short-term failure Doppler velocity log (DVL), we propose a machine learning (ML) aided method for strapdown inertial navigation system (SINS)/DVL integration solution. First, the inherent relationship between the underwater vehicle’s dynamics characteristic and the SINS’s velocity error is established through the learning methodology of the least square support vector machine (LS-SVM), and the prediction and compensation are performed during the failure period of the DVL. When the DVL signal is normal, the LS-SVM model is trained, the adaptive Kalman filtering (AKF) is adopted in the SINS/DVL integrated navigation system, the filtering estimation value is used to correct the SINS’s navigation calculation value. When the DVL signal is invalid, the variation of underwater vehicle movement is taken as the input of the LS-SVM model. Land vehicle field experiment is conducted to verify the feasibility and effectiveness of the LS-SVM/AKF algorithm aided SINS/DVL integrated navigation system. The results indicate that the proposed methodology can improve the accuracy of the SINS/DVL integrated navigation system during short-term failure of DVL.