Error Prediction Method of Geomagnetic Model Based on Extreme Learning Machine

The high-precision geomagnetic field model is an important foundation for autonomous navigation of near earth satellites, but the improvement of navigation accuracy is constrained by observation errors, spherical harmonic truncation errors, and slow updates of the geomagnetic model. To solve this problem, this paper proposes a geomagnetic model error prediction method based on regularized extreme learning machine. The optimal estimation of the regularization coefficient C is achieved by using a subtraction mean algorithm, which reduces subjectivity and randomness in parameter tuning, improves learning efficiency and prediction accuracy. In addition, this method can effectively improve the error estimation accuracy when outliers exist in geomagnetic observation sequences. Then, a geomagnetic navigation method with model error compensation was proposed by integrating it with filtering algorithms, and simulation verification was conducted using real geomagnetic measurement data from in orbit satellites. The results show that the prediction accuracy of the method proposed in this paper is superior to several commonly used neural network prediction methods, and the navigation accuracy reaches 1.49km, indicating that the proposed error prediction model can effectively improve the performance of geomagnetic navigation.