On-orbit identification algorithms for dynamics parameters of gravity satellite

Due to remote orbital maneuvers, payload separation and fuel consumption, the mass properties and dynamic parameters of gravity satellites change in real-time. These changes make it challenging to maintain high-precision stability in attitude control of the spacecraft, affecting the execution of space missions. This paper establishes an attitude dynamics model for gravity satellites that explicitly includes on-orbit variable parameters. Furthermore, backpropagation (BP) neural network is used to identify the dynamic parameters of gravity satellites. Finally, BP neural network is compared with the recursive least squares algorithm (RLS) and the random weight particle swarm optimization (PSO) algorithm. The results indicate that the identification error mean of the BP neural network can achieve 10^-3 orders of magnitude, and the root mean square error can achieve 10^-1 orders of magnitude. The identification accuracy and robustness are superior to those of the RLS algorithm and the random weight PSO algorithm. This paper compares the different high-precision on-orbit identification algorithms for dynamic parameters of gravity satellite, laying a foundation for the identification algorithms design and selection of the next generation gravity satellites.