Error Compensation Algorithm for Dynamic Model Based on Neural Network

Abstract

The noise of GNSS navigation and positioning system is non-priori, while the optimal estimation of standard Kalman filter requires the establishment of accurate system model and observation model, which leads to the low accuracy of Kalman filter. Neural network has strong ability of denoising, learning, self-adapting and complex mapping. In order to improve the filtering accuracy, this paper proposes an algorithm to compensate the error of the dynamic model by using the neural network, and corrects the error of the dynamic model by using the RBF neural network in the filtering estimation part, which inhibits the contribution of the abnormal disturbance of the dynamic model to the navigation solution. The experimental results show that the algorithm can not only eliminate the positioning deviation in all directions, but also reduce the standard deviation in X, Y and Z directions by about 70%, 60% and 60% respectively, compared with the standard Kalman filter.