Subjects: Physics >> Geophysics, Astronomy, and Astrophysics Subjects: Information Science and Systems Science >> Basic Disciplines of Information Science and Systems Science submitted time 2024-06-09
Abstract: The precise estimation of the satellite clock bias (SCB) holds considerable importance in ensuring accurate timekeeping, navigation, and positioning. This studyintroduces a novel SCB prediction approach that integrates variational mode decomposition (VMD) and long short-term memory (LSTM) network techniques, combining signal decomposition with deep learning methodologies. Initially, the raw SCB data undergoespreprocessing, followed by decomposition using the VMD method to generate multiple intrinsic mode functions (IMFs). These decomposed IMFs serve as inputs for LSTM, where several independent LSTM models are established for training and prediction purposes. Subsequently, the predicted outcomes are aggregated and reconstructed to derive the finalSCB prediction. Experimental findings demonstrate notable advancements in clock bias prediction for the spaceborne hydrogen atomic clock for BDS, with prediction accuracies of 0.048 ns, 0.204 ns and 1.397 ns for 6 hours, 3 days and 15 days, respectively. These results exhibit significant enhancements compared to both the LSTM network and the Back Propagation (BP) neural network, with improvements of 56%, 84% and 83% for the aforementioned time intervals in comparison to LSTM, and enhancements of 59%, 82% and 83% relative to the BP neural network.
Peer Review Status:Awaiting Review