The Mobile Robot GPS Position Based on Neural Network Adaptive Kalman Filter
Wei Wu, Wei Min · 2009
This paper presents a GPS positioning method based on neural network adaptive Kalman filter. Using the innovation vector which reflects the degree how the model fits the data, and real-timely accessing to the innovation vector's ratio of the theoretical variance to the actual of variance, we can get the working conditions of Kalman filter. Then track the change of system parameters through neural network, where the adaptive regulatory factors are generated which can correct the Kalman filter, improve the performance of the Kalman Filter, and prevent the filter divergence. Because neural network has a strong learning and adaptive ability, the system noise covariance matrix can be corrected real-timely, and can be adjusted online. The simulation results show that the error divergence produced by Kalman filter can be effectively suppressed, and the positioning accuracy of mobile robot GPS positioning system can be effectively improved.