Application of feed-forward neural network based on adaptive fading UKF in integrated navigation

摘要

Aiming at the problem of learning slowly and falling into local minimum easily, this paper proposed an improved neural network (NN) training method and applied it to the integrated navigation system. In this method, we employed adaptive fading unscented Kalman filtering (AFUKF) algorithm as a weights training tool for feed-forward NN. AFUKF is a real-time filtering algorithm based on standard UKF and avoids tremendous computational burden on the navigation's processor. Then, the trained NN by AFUKF was applied to the strapdown inertial navigation system/BeiDou navigation satellite system integrated navigation system and compared with NN method based on extended Kalman filter (EKFNN). Simulation results demonstrate that the proposed algorithm can effectively improve the positioning precision of navigation system, and the filtering performance of AFUKFNN is significantly superior to that of the EKFNN.