Neural Networks Aided Unscented Kalman Filter for Integrated INS/GNSS Systems

Abstract

The integrated navigation system consists of Inertial Navigation System (INS) and receiver of Global Navigation Satellite System (GNSS). Aiming to improve position and velocity precision of the INS/GNSS system during GNSS outages, a novel method that combines unscented Kalman filter (UKF) and nonlinear autoregressive neural networks with external inputs (NARX) is proposed (namely NARX aided UKF). The NARX-based module is used to predict the measurements for UKF during GNSS signal outages. A new method for choosing inputs of NARX networks is suggested. This method is based on mutual information criterion (MI) for identifying the inputs that influence each of outputs and lag-space estimation (LSE) for investigating the dependency of these outputs on the past values of inputs and outputs. The performance of the proposed methodology is experimentally verified using data acquired from simulated flight trips, in which the measurement model of MEMS-based INS is used.