A low-cost GPS/INS integration based on UKF and BP neural network

Abstract

Nowadays, low-cost Global Positioning System (GPS)/inertial Navigation System (INS) integration is widely used. Numerous techniques based on Kalman Filter (KF) and Artificial Neural Networks (ANNs) are proposed to fuse the GPS and INS data. Kalman filter is an optimal real-time data fusion method for GPS/INS integration while GPS signal is available. But when GPS outages, Kalman filter cannot provide estimated position errors for INS. Without compensation, navigation accuracy will deteriorate badly along with time. ANNs are able to handle the problem of non-linearity and map input-output relationships without prior knowledge. In order to provide continuous, accurate and reliable navigation solution even during GPS outages, we proposed a novel model of combining UKF and BP neural network algorithms for INS errors compensation. UKF is an implementation of KF with great performance and used to ensure the high accuracy when GPS is available. BP is a most widely used method of training a multi-layer Feed-Forward Artificial Neural Networks (FFANNs). On the basis of enough training, it can predict INS position error when GPS signal is blocked. The model has been verified to have good performance for fusing GPS and INS data, even when GPS signal is unavailable.