Retrieval-augmented patch generation for geosynchronous satellite status forecasting.

Area of Science:

• Space Situational Awareness
• Satellite Dynamics
• Artificial Intelligence in Aerospace

Background:

• Accurate geosynchronous satellite status forecasting is crucial for space situational awareness.
• Current models struggle with non-stationary, weakly periodic high-orbit data, hindering long-range dependency learning and adaptation to evolving orbital dynamics.

Purpose of the Study:

• To propose RAPG (Retrieval-Augmented Patch Generation), a framework for geosynchronous satellite status analysis, including forecasting and intent classification.
• To address limitations in modeling complex orbital dynamics and improve temporal representation learning.

Main Methods:

• RAPG utilizes an adaptive frequency-domain patching mechanism to segment satellite status time series based on spectral characteristics.
• A retrieval-augmented generation module accesses historical data for reconstructing non-repetitive motion patterns.
• The framework links adaptive patching with retrieval-based augmentation for enhanced temporal learning.

Main Results:

• RAPG consistently outperformed strong baselines in both geosynchronous satellite status forecasting and intent classification tasks.
• The framework achieved the lowest forecasting errors.
• RAPG attained an F1-score of 0.9461 and an accuracy of 0.9443 for intent classification.

Conclusions:

• RAPG offers a robust and scalable solution for real-time analysis of geosynchronous satellites.
• The framework significantly enhances remote sensing-based space domain awareness and dynamic orbital intelligence.
• RAPG demonstrates improved generalization across diverse orbital conditions.