Research on Error Compensation Method of SINS/GNSS Integrated Navigation System Based on NARX Neural Network

This study proposes a new algorithm combining integrated navigation and neural networks, which combines a nonlinear autoregressive neural network (NARX) with an extended Kalman filter (EKF), based on the global navigation satellite system (GNSS) and strapdown inertial navigation system (SINS). The purpose is to use the NARX neural network to predict the measurement update of EKF during GNSS signal interruption, to compensate for the errors of SINS instead of EKF,to correct the divergence error of SINS, and to ensure the accuracy of the system. The NARX neural network uses the Levenberg-Marquard algorithm to train the neural network, which speeds up the neural network's convergence rate. As a dynamic neural network model, the NARX neural network takes into account the dependence of the current output on the past input and output. As a result, it uses the output at the current time as the feedback input at the next time. The simulation results demonstrate that employing the NARX neural network for error compensation during GNSS signal interruption can somewhat increase the location and speed accuracy of the SINS/GNSS system.