A collaborative neural network-based framework for distributed disturbance compensation in satellite constellations

This study proposes a collaborative neural network (NN-Shared) framework for real-time distributed disturbance compensation in satellite constellations, based on the RGT-Walker closed-form solutions. Unlike conventional leader-follower or independent schemes, all satellites in the constellation share full structural and disturbance homogeneity, allowing for real-time collective learning and exchange of disturbance and error information. The closed-form RGT-Walker formulation offers strongly regular and periodic orbital patterns, and is demonstrated, both theoretically and experimentally, to provide superior prediction and generalization capacity for deep learning compared to classic J 2 -based models. Numerical simulations confirm that the NN-Shared approach greatly outperforms conventional controllers such as Kalman, PID, and hierarchical methods in terms of control effort, fuel consumption, and trajectory error at the constellation level. These results provide practical and technical foundations for deploying distributed neural collaborative control and information sharing as essential elements for next-generation autonomous, high-precision satellite constellation operations in the real space environment.