About
Experience & Education
Licenses & Certifications
-
Artificial Intelligence in Health Care
MIT Sloan School of Management
Publications
-
A Novel Method for Satellite Maneuver Prediction
Advanced Maui Optical and Space Surveillance Technology Conference publication description
A space operations tradecraft consisting of detect-track-characterize-catalog is insufficient for maintaining Space Situational Awareness (SSA) as space becomes increasingly congested and contested. In this paper, we apply analytical methodology from the Geospatial-Intelligence (GEOINT) community to a key challenge in SSA: predicting where and when a satellite may maneuver in the future. We developed a machine learning approach to probabilistically characterize Patterns of Life (PoL) for…
A space operations tradecraft consisting of detect-track-characterize-catalog is insufficient for maintaining Space Situational Awareness (SSA) as space becomes increasingly congested and contested. In this paper, we apply analytical methodology from the Geospatial-Intelligence (GEOINT) community to a key challenge in SSA: predicting where and when a satellite may maneuver in the future. We developed a machine learning approach to probabilistically characterize Patterns of Life (PoL) for geostationary (GEO) satellites. An example of PoL are station-keeping maneuvers in GEO which become generally predictable as the satellite re-positions itself to account for orbital perturbations.
In an earlier publication, we demonstrated the ability to probabilistically predict maneuvers of the Galaxy 15 (NORAD ID: 28884) satellite with high confidence eight days in advance of the actual maneuver (Shabarekh et al, 2016). This was done with a custom unsupervised machine learning algorithm, the Interval Similarity Model (ISM), which learns repeating intervals of maneuver patterns from unlabeled historical observations and then predicts future maneuvers. In this paper, we introduce a supervised machine learning algorithm that works in conjunction with the ISM to produce a probabilistic distribution of when future maneuvers will occur. The supervised approach uses a Support Vector Machine (SVM) to process the orbit state whereas the ISM processes the temporal intervals between maneuvers and the physics-based characteristics of the maneuvers. This multiple model approach capitalizes on the mathematical strengths of each respective algorithm while incorporating multiple features and inputs. Initial findings indicate that the incorporation of the SVM and orbit state information can increase accuracy and timeliness of predicted maneuvers over the ISM and astrometric observations only.Other authorsSee publication -
Efficient Target Characterization to Support Space Situational Awareness
2016 Space Symposium
As the number of objects in space increase exponentially, the need for Space Situational Awareness (SSA) to protect assets from environmental dangers (such as collisions) increases as well. SSA may become a “big data” problem due to the prevalence of low-cost small satellites and proliferation of debris. Additionally, there are multiple, uncoordinated, space observation systems collecting data at varying cadences to create datasets that grow in proportion to the number of telescopes and other…
As the number of objects in space increase exponentially, the need for Space Situational Awareness (SSA) to protect assets from environmental dangers (such as collisions) increases as well. SSA may become a “big data” problem due to the prevalence of low-cost small satellites and proliferation of debris. Additionally, there are multiple, uncoordinated, space observation systems collecting data at varying cadences to create datasets that grow in proportion to the number of telescopes and other sensors. While these datasets are large, they are not persistent or conditioned and are frequently noisy, which makes it challenging to maintain a satellite’s chain of custody and detect out-of-class maneuvers in a timely manner.
Although many SSA operations remain a manual process, Aptima, in partnership with the Air Force Research Labs Space Vehicles Directorate (AFRL/RV), has developed automated satellite maneuver prediction algorithms that learn a satellite’s pattern of life (PoL) and predict when and where future maneuvers will occur. The objective is to incorporate spatio-temporal and relational context to identify maneuvers that are inconsistent with expected, nominal operations. In turn, this approach enables accurate prediction of future states and the rapid identification of deviations from expected behaviors, even in non-persistent environments.
To achieve this, we adapted computationally efficient machine learning algorithms that we originally developed for Activity Based Intelligence (ABI) capabilities in the land, sea and air domains. We have demonstrated high accuracy of probabilistically predicting maneuvers of the Galaxy 15 (NORAD ID: 28884) satellite on noisy, intermittent synthetic datasets. Early results indicate accurate prediction of future maneuvers from a short time history of past observations and lay the groundwork for applications in UCT association, dynamic sensor tasking and patterns of life analysis.Other authorsSee publication
Honors & Awards
-
Air & Space Campaign Medal
Department of Defense
-
Air Force Achievement Medal
Department of Defense
-
Air Force Commendation Medal
Department of Defense
-
Combat Readiness Medal
Department of Defense
-
Defense Meritorious Service Medal
Department of Defense
-
Global War on Terrorism Medal
Department of Defense
-
Joint Service Achievement Medal
Department of Defense
-
Joint Service Commendation Medal
Department of Defense
-
Meritorious Service Medal
USAF
-
NOAA Corps Commendation Medal
NOAA
-
National Defense Service Medal
Department of Defense
Languages
-
Russian
Native or bilingual proficiency
-
Ukrainian
Limited working proficiency
People also viewed
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Gene Keselman
3 others named Gene Keselman are on LinkedIn
See others named Gene Keselman