Recurrent networks for integrated navigation

Abstract

A novel neural-network-compensated Kalman filter for integrated navigation system was proposed. Based on the similarity of operation principle between Elman networks and non-linear ARMA model, the Elman network is employed as a compensating error estimator to improve accuracy of the Kalman filter. The proposed architecture is evaluated with the acquired data from a naval vessel. And the results show that the presented method can markedly attenuate the effect of interferes to Kalman filter, and improve the precision of the integrated navigation system. © Springer-Verlag Berlin Heidelberg 2005.