Abstract:
As more countries launch and operate spacecraft, the potential for misunderstandings or uncertainties regarding the intent of a spacecraft’s actions increase. The LEO region is increasingly populated, and most manoeuvres are conducted for stationkeeping purposes. While the GEO region is not as heavily populated, GEO missions are longer-lived, conduct frequent manoeuvres, and depend upon the critical GEO orbital regime and its “thin altitude slice” that we must preserve and protect. This paper presents a new Machine Learning (ML) manoeuvre intent classification tool for GEO spacecraft to help characterize the intent of GEO manoeuvres and detect abnormal or potentially hostile behaviours. Manoeuvre intent for GEO satellites can include inclination station-keeping, longitudinal stationkeeping, longitudinal translation, and collision avoidance. To reach this goal, AIKO worked closely with COMSPOC to explore the advantages of an AI approach to this problem. Literature frequently addresses the problem of manoeuvre detection using both machine and deep learning techniques. The present research instead aims to employ AI algorithms for manoeuvre intent classification drawing upon a rich dataset of manoeuvres for an initial group of 18 satellites provided by COMSPOC. The specific core technology utilized for this study is the Decision Tree algorithm. This, together with time-series prediction techniques for manoeuvre impact assessment, can classify the intent of one or more manoeuvres (in sequence). The intent classes were mutually defined by AIKO and COMSPOC and additionally informed by international standards definitions for manoeuvre intent (i.e., MAN_PURPOSE in ISO 26900: Space Systems – Orbit Data Message). The result of this study is a prototypical tool that seeks to analyse and classify with high confidence sequences of manoeuvres to support Space Situational Awareness (SSA) and Flight Operations control centres. An accompanying live demo has also been developed. The prototype’s performance is promising; the tool was able to classify correctly 98% of the manoeuvre sequences extracted from the dataset, as identified in the dataset.
Keywords:
Click a keyword to filter the list of related assets below.
T. Valentinova Avramova, P. De Marchi, D.L. Oltrogge, J.J. Cornelius, D.A. Vallado, F. Caronte, N. Casciola, “Watch out GEO satellites, here’s a new ML-method for manoeuvre detection and intent classification,” 4THInternational Conference on Space Situational Awareness (ICSSA), IAA-ICSSA-24, Daytona Beach, FL USA 8 May 2024.
