A Hybrid Method for INS/GPS Integrated Navigation System Based on the Improved Karman Filter and Back Propagation Neural Network

The navigation system for unmanned aerial vehicle is commonly integrated by the inertial navigation system (INS) and global positioning system (GPS). Nevertheless, errors of INS/GPS accumulate over time when GPS suffers from outages. In this paper, a comprehensive filter called MH∞-5thCKF is proposed for taking the place of Karman filter (KF). It combines superiorities of H-infinity filter and multiple fading filter to enhance filtering precision and robustness. In addition to MH∞-5thCKF, an optimized back propagation neural network (BPNN) is utilized for GPS outages. The ELSHADE-SPACMA algorithm is introduced to select BPNN parameters. When satellite signals are available, the improved BPNN uses angular rates, specific forces and GPS increments to train the model. Once satellite signals are lost, the improved BPNN predicts pseudo-GPS information so that MH∞-5thCKF continues compensating INS errors. Compared with the conventional KF and BPNN, simulation results demonstrate that proposed algorithms not only enhance filtering performance, but also avoid BPNN falling into local optimum to guarantee model stability when GPS fails to work.