Efficient gravity field modeling method for small bodies based on Gaussian process regression

The detection of small bodies is very important, and studying the gravity fields of small bodies is a key requirement when designing exploration missions. This paper proposes an efficient gravity field modeling method based on Gaussian process regression from the novel perspective of data statistical and law mining. In contrast to classical gravity field modeling methods, the proposed method establishes a direct mapping relationship between the gravity field and the field point from a statistical point of view, which avoids the complex modeling calculation process. This method can accurately obtain the gravitational acceleration at a checkpoint while improving the calculation efficiency. In order to ensure the quality of the data samples required for the Gaussian process regression, a training set is obtained using a high-precision polyhedron modeling method. The kernel functions are selected, and the hyperparameters are optimized to establish a gravity field training model. Numerical simulations demonstrated that the mean relative error was only 1.27% when this approach was applied to the nearby gravity field modeling of four target small bodies with different sizes. In terms of the runtime requirements, the magnitude of computing time was reduced from 10 ⁴ s for the polyhedron method to 10 ⁻¹ s.