Error prediction method of geomagnetic model based on extreme learning machine

Large inherent error of geomagnetic model and long updating time of coefficient are the main reasons that restrict the improvement of geomagnetic navigation accuracy. In order to solve this problem, an error prediction model based on regularized extreme learning machine is proposed in this paper. By establishing the mapping relationship between geomagnetic parameters and time information and geomagnetic field intensity vector elements, combined with real satellite magnetic field measurement data, the error estimation and prediction of geomagnetic field model are realized. Then, a geomagnetic navigation method based on the fusion of model prediction method and extended Kalman filter is proposed. The navigation results are simulated by using the geomagnetic measured data of the orbiting satellite. The results show that: Compared with several conventional neural network prediction methods, the position accuracy of geomagnetic navigation can reach 1.96km, indicating that the proposed error prediction model can effectively improve the performance and accuracy of geomagnetic navigation.