RT-LSTM
A dual-LSTM real-time RTT prediction model proposed within the StarRTT framework; the history branch encodes historical observations and real-time satellite features, while the future branch uses TLE propagation features, achieving high-accuracy prediction under strict real-time backtesting.
Timeline (last 90 days)
RT-LSTM adopts a dual-LSTM structure: the history LSTM encodes historical RTT observations and real-time relative features of the serving satellite and gateway; the future LSTM processes future features, derived from TLE propagation, of the identified serving satellite up to the next scheduling instant.
In strict real-time backtesting, RT-LSTM achieved an average MAE of 1.405 ms across all 15 prediction horizons and 15 reference slot indices.
Compared with existing baseline models, RT-LSTM reduces MAE by 4.3% (vs. LSTM), 11.6% (vs. StarNet), and 22.1% (vs. the attention-based spatiotemporal model ASTM).
At the scheduling instant, RT-LSTM's error reduction relative to LSTM reaches 12.5%.
All Facts (4)
RT-LSTM adopts a dual-LSTM structure: the history LSTM encodes historical RTT observations and real-time relative features of the serving satellite and gateway; the future LSTM processes future features, derived from TLE propagation, of the identified serving satellite up to the next scheduling instant.
50%UnverifiedIn strict real-time backtesting, RT-LSTM achieved an average MAE of 1.405 ms across all 15 prediction horizons and 15 reference slot indices.
50%UnverifiedCompared with existing baseline models, RT-LSTM reduces MAE by 4.3% (vs. LSTM), 11.6% (vs. StarNet), and 22.1% (vs. the attention-based spatiotemporal model ASTM).
50%UnverifiedAt the scheduling instant, RT-LSTM's error reduction relative to LSTM reaches 12.5%.
50%