On SINS/GPS Integrated Navigation Filtering Method Aided by Radial Basis Function Neural Network

Abstract

Both Strapdown Inertial Navigation System (SINS) and GPS are nonlinear systems. Kalman Filter (KF) is frequently used as the fusion data technology of the SINS/GPS system. Before using KF for nonlinear system, to linearize this system will bring large errors. Moreover, during GPS outages, the integrated system cannot get the observations for KF algorithm. So, the navigation errors will grow rapidly with time. Aiming at the two problems and considering Radial Basis Function Neural Networks (RBFNN) can approximate nonlinear systems with arbitrary accuracy, we propose a SINS/GPS integrated system filtering method aided by RBFNN in the paper. When the GPS signal is locked, the trained RBFNN assists KF to predict the difference between ideal state errors and KF posteriori estimate errors, and then compensate the estimate errors of KF. During GPS outages, in order to estimate GPS outputs at the current filtering moment, the trained RBFNN is adopted to predict the increments of GPS observations. And then KF measurement is provided to damp the rapid accumulation of navigation errors. The simulation results indicate that the algorithm can improve the KF estimate accuracy when satellite signal is locked, and the navigation accuracy of the system is significantly improved during GPS outages.