Neural Network-Aided Adaptive UKF for Integrated Underwater Navigation

Gravity aided navigation and geomagnetism aided navigation are equally important methods in the field of underwater navigation. However, the former is affected by terrain fluctuations, and the latter is sensitive to time-varying noise. Considering the characteristics that the gravity gradient vector can avoid the influence of time-varying noise and is less sensitive to terrain fluctuations, we propose to integrate the gravity gradient vector and geomagnetic vector together to achieve the merits of each aided navigation method. The gravity gradient vector and geomagnetic vector are used as measurement information from both local neural network-aided adaptive UKF filters, and then an information fusion algorithm based on weighted least squares estimation is used to combine the estimated values from each local filter to form an optimal estimated state value. Finally, the optimal estimated value is used to update the output values from each local neural network –aided adaptive UKF filter. Experimental results prove the feasibility of this integrated navigation method.