Research on Modeling and Predicting Real-time Satellite Clock Offset Using VMD-LSTM Model

Real-time satellite clock offsets are prerequisite for real-time precise point positioning. However, disruptions in communication due to suboptimal network conditions are unavoidable, which consequently render real-time satellite clock offsets unreliable or unavailable. To maintain the continuity of real-time precise point positioning, a real-time satellite clock offset prediction model based on VMD-LSTM is proposed. Considering that the real-time satellite clock offsets are complex and variable, the proposed model initially decomposes the satellite clock offsets into multiple modal components with distinct frequencies, utilizing the variational mode decomposition (VMD), aiming to simplify the intricate nature of real-time satellite clock offsets. Then each modal component is individually modeled using the long short-term memory (LSTM) model to accurately capture the characteristics and variation of the clock offsets. Finally, the individual predictions made for each modal component are combined to generate a consolidated predict for the real-time satellite clock offset prediction. The experiment results show that the VMD-LSTM model significantly improves the prediction accuracy of real-time satellite clock offsets. Compared with the quadratic polynomial (QP), grey model (GM11), and LSTM models, the prediction accuracy of the VMD-LSTM model for 5-minute, 15-minute, and 30-minute is improved by (27.2, 15.8, 4.45)%, (15.0, 30.8, 7.85)%, and (38.6, 44.4, 17.7)% for BDS-2, (60.5, 50.8, 10.9)%, (69.9, 50.1, 14.4)%, and (71.8, 63.7, 24.2)% for BDS-3, and (1.83, 63.4, 3.47)%, (13.7, 58.5, 7.98)%, and (14.0, 48.7, 23.9)% for GPS, respectively.