Direct mapping representation of small-body gravity fields under multidimensional features

Abstract

An accurate model of the gravity fields around irregular small bodies is significant for exploration missions. However, it is difficult for classical modeling methods to guarantee both calculation accuracy and efficiency. This paper proposes a direct mapping representation under multidimensional features for the rapid evaluation of small-body gravity fields. This representation establishes a mapping relationship between the features and corresponding gravitational acceleration using machine learning. And two implicit features are mined deeply to exploit more information about small bodies. Different feature-combination schemes with implicit features are investigated and compared to improve prediction accuracy and reduce training samples. The effectiveness of these schemes is validated by applying the presented method for the evaluation of gravitational accelerations at field points near comet 67P/C-G. Simulation results indicate that the scheme incorporating two implicit features is more accurate than the scheme based only on coordinates. Specifically, with the same training samples, average prediction errors are reduced by 30% compared to the latter; and under the same accuracy, the requirement for training samples shows a drop of 40%.