Clock Deviation Long-term Prediction for Autonomous Navigation Satellites with Neural Networks Supported by Inter-satellite Links

A prediction scheme of clock deviation in long-term supported by inter-satellite links (ISLs) is proposed for navigation satellites. A Synthesis Paper Time (SPT) with higher predictability is established firstly. Considering that the performance of polynomial model is affected by length of data arc and the order of models, a long-term prediction model with nonlinear weighting through neural networks of the SPT is further built. The satellite clock prediction is finally obtained from ISL time transmission. Results illustrate the 15-day accuracy of predicted each satellite clock can reach 0.63 ns, observably increased relative to the conventional prediction scheme.