The Near-Earth environment is becoming increasingly congested due to the rapid growth of commercial satellite constellations and service-oriented missions. As a consequence, an efficient Space Traffic Management (STM) framework becomes necessary to ensure sustainable and safe operations. Within this context, the Space Surveillance and Tracking (SST) programs play a crucial role by continuously processing tracking measurements and maintaining up-to-date catalogs of Resident Space Objects (RSOs). However, the inherent limitations of the widely used Simplified General Perturbation Model (SGP4) in orbital prediction, particularly due to its reliance on mean orbital elements and simplified force modeling, lead to significant inaccuracies over extended propagation periods. This work leverages a data-driven approach to enhance orbit prediction by integrating the SGP4-based propagation model with a Long Short-Term Memory (LSTM) neural network. The proposed hybrid methodology is designed to correct the systematic errors of SGP4 approximated states by learning correction patterns from high-fidelity ephemeris data. The effectiveness of the hybrid SGP4-LSTM model is assessed on multiple objects. Experimental results demonstrate a substantial reduction in position and velocity prediction errors, achieving up to a 98% improvement over the standard SGP4 propagation. This approach is non-intrusive and computationally efficient by seamlessly integrating into the existing SST workflows, thus improving orbital state estimation without requiring modifications to the traditional SGP4-TLE pipeline. By refining orbit predictions and quantifying uncertainty, this approach represents a step forward in the advancement of data-driven techniques for space situational awareness (SSA).
L’ambiente spaziale in prossimità della Terra sta diventando sempre più congestionato a causa della rapida crescita delle costellazioni satellitari commerciali e delle missioni orientate ai servizi. Di conseguenza, diventa sempre più fondamentale sviluppare un framework efficiente per la gestione del traffico spaziale al fine di garantire operazioni sicure e sostenibili. In questo contesto, i programmi di Sorveglianza e Tracciamento dello Spazio svolgono un ruolo cruciale, elaborando continuamente misurazioni di tracciamento e mantenendo aggiornati i cataloghi degli oggetti spaziali. Tuttavia, il modello semplificato di perturbazioni generali (SGP4), ampiamente utilizzato per la previsione orbitale, presenta limitazioni intrinseche dovute all’uso di elementi orbitali medi e a una modellazione semplificata delle forze perturbative, portando a imprecisioni significative nelle previsioni su lunghi periodi di propagazione. Questo lavoro sfrutta un approccio basato sull’apprendimento automatico per migliorare la predizione orbitale, integrando il modello di propagazione SGP4 con una rete neurale a Memoria a Breve e Lungo Termine (LSTM). La metodologia ibrida proposta è progettata per prevedere e correggere gli errori sistematici dello stato orbitale approssimato da SGP4, apprendendo schemi di correzione dai dati derivanti dalle efemeridi. L’efficacia del modello ibrido SGP4-LSTM viene valutata su molteplici oggetti. I risultati sperimentali dimostrano una significativa riduzione degli errori di previsione della posizione e della velocità, con un miglioramento fino al 98% rispetto alla propagazione standard di SGP4. Questo approccio è non invasivo ed efficiente dal punto di vista computazionale, poiché si integra perfettamente nei flussi di lavoro esistenti, migliorando la stima dello stato orbitale senza necessità di modificare la pipeline tradizionale SGP4-TLE. Raffinando le previsioni orbitali e quantificando l’incertezza, questa metodologia rappresenta un passo avanti nell’applicazione di tecniche data-driven nell’ambito della Space Situational Awareness (SSA).
An hybrid LSTM-based SGP4 propagator
BERTOLINI, LUDOVICA
2023/2024
Abstract
The Near-Earth environment is becoming increasingly congested due to the rapid growth of commercial satellite constellations and service-oriented missions. As a consequence, an efficient Space Traffic Management (STM) framework becomes necessary to ensure sustainable and safe operations. Within this context, the Space Surveillance and Tracking (SST) programs play a crucial role by continuously processing tracking measurements and maintaining up-to-date catalogs of Resident Space Objects (RSOs). However, the inherent limitations of the widely used Simplified General Perturbation Model (SGP4) in orbital prediction, particularly due to its reliance on mean orbital elements and simplified force modeling, lead to significant inaccuracies over extended propagation periods. This work leverages a data-driven approach to enhance orbit prediction by integrating the SGP4-based propagation model with a Long Short-Term Memory (LSTM) neural network. The proposed hybrid methodology is designed to correct the systematic errors of SGP4 approximated states by learning correction patterns from high-fidelity ephemeris data. The effectiveness of the hybrid SGP4-LSTM model is assessed on multiple objects. Experimental results demonstrate a substantial reduction in position and velocity prediction errors, achieving up to a 98% improvement over the standard SGP4 propagation. This approach is non-intrusive and computationally efficient by seamlessly integrating into the existing SST workflows, thus improving orbital state estimation without requiring modifications to the traditional SGP4-TLE pipeline. By refining orbit predictions and quantifying uncertainty, this approach represents a step forward in the advancement of data-driven techniques for space situational awareness (SSA).File | Dimensione | Formato | |
---|---|---|---|
Thesis_Bertolini_Ludovica.pdf
accessibile in internet per tutti
Descrizione: Thesis in An Hybrid LSTM-based SGP4 propagator
Dimensione
21.44 MB
Formato
Adobe PDF
|
21.44 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/235960