As space traffic increases, Space Situational Awareness (SSA) is becoming fundamental for safe spaceflight operations. Cost-driven missions based on small satellite platforms would benefit from the availability of alternative tools providing preliminary SSA from publicly available information, such as two-line elements. In this work, we propose an orbit prediction and uncertainty evaluation method based on the well-established TLE differencing technique aided by a machine learning corrector. By designing a Recurrent Neural Network with carefully chosen input parameters, the TLE prediction accuracy is significantly improved, when tested against precise orbital data of real satellites. The prediction error is reduced, on average, by 45% across a prediction window of 16 days which may include manoeuvres. We further show that in combination with a statistical test for equality between error distributions, the differencing technique applied to the corrected TLE allows a reliable variance estimate in most situations. Limitation of the work is the training of a dedicated neural network corrector for each specific space object, which will be deposed as part of our ongoing efforts.

Two-line-element propagation improvement and uncertainty estimation using recurrent neural networks

Curzi, Giacomo;Modenini, Dario;Tortora, Paolo
2022

Abstract

As space traffic increases, Space Situational Awareness (SSA) is becoming fundamental for safe spaceflight operations. Cost-driven missions based on small satellite platforms would benefit from the availability of alternative tools providing preliminary SSA from publicly available information, such as two-line elements. In this work, we propose an orbit prediction and uncertainty evaluation method based on the well-established TLE differencing technique aided by a machine learning corrector. By designing a Recurrent Neural Network with carefully chosen input parameters, the TLE prediction accuracy is significantly improved, when tested against precise orbital data of real satellites. The prediction error is reduced, on average, by 45% across a prediction window of 16 days which may include manoeuvres. We further show that in combination with a statistical test for equality between error distributions, the differencing technique applied to the corrected TLE allows a reliable variance estimate in most situations. Limitation of the work is the training of a dedicated neural network corrector for each specific space object, which will be deposed as part of our ongoing efforts.
Curzi, Giacomo; Modenini, Dario; Tortora, Paolo
File in questo prodotto:
Eventuali allegati, non sono esposti

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11585/858672
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? 0
social impact