Multi-OP: A Multilevel Model for Orbital Prediction Using Multi-Feature Fusion

To address the increasing computational demands and the need for enhanced stability in the orbit prediction of large-scale resident space objects (RSOs), this paper proposes a novel method grounded in orbital dynamics, which defines and computes environmental, dynamical, and fingerprint features. A feature fusion strategy is then applied to introduce a Multi-OP model comprising a shared backbone and a prediction layer tailored for orbit prediction tasks. Specifically, the shared backbone integrates environmental and dynamical features to capture public information, thereby enhancing the model’s generalization ability. The prediction layer, in contrast, leverages fingerprint features to generate individualized orbit predictions for each RSO. Experimental results demonstrate that the Multi-OP model not only enhances the accuracy of orbit prediction but also exhibits remarkable stability.