Data-Driven Surrogate Model and Admissible-Region-Based Orbital Uncertainty Propagation

With fast-growing space objects but limited sensor resource, optical-sensor-based short-arc uncertainty propagation is necessary for space object tracking. Recently, the admissible-region-based orbit initialization method has been shown to be effective for uncertainty propagation when the number of observations is very small. However, it is challenging to describe the uncertainty distribution in the admissible region. In this paper, a data-driven kriging method is combined with the admissible region to build an accurate surrogate model for orbital uncertainty propagation instead of using the Monte Carlo method directly. In addition, an active learning strategy is utilized to choose sample data from the admissible region more efficiently for incrementally building the surrogate model. The main advantage of the kriging method is that no a priori probability density function needs to be assumed for the uncertainty distribution. Two numerical examples including a low Earth orbit and a geosynchronous Earth orbit propagation are used to demonstrate the effectiveness of the kriging model. It compares favorably to the polynomial chaos-based uncertainty propagation in terms of accuracy. It is also shown that the proposed method can achieve close performance to the Monte Carlo results but with much less sampling points and higher computational efficiency.