Robust Gaussian Mixture Model for Maneuver Detection Using TLE Data

Objectives: The number of space objects has grown exponentially due to increased space activities, significantly increasing the collision risk in the Earth’s orbit. Proactive detection and accurate monitoring of changes in the orbits of space objects, including in-orbit collisions and orbital maneuvers of space objects, have become essential. The two-line element (TLE) set is a data format containing information on the movement of objects in the Earth’s orbit. It is the primary resource for monitoring space objects. However, frequent variations in the orbits can cause anomalies in the TLE data, potentially affecting the accuracy of maneuver detection. Therefore, we propose a new maneuver detection methodology that uses a robust Gaussian mixture model (RGMM) to perform probabilistic adjustment of the TLE. Methods: The method used the rate of change of TLE semi-major axis prediction error to detect maneuvers. The robustness of the model is improved by pruning the Gaussian mixture model (GMM) and constraining the a posteriori probability of suspicious data through the incorporation of a robust correction function in the parameter solving process. We compared the performance of the proposed approach for detecting the maneuver of a typical space object with the GMM. Results: The results show that: (1) The RGMM demonstrated greater stability and robustness to outliers in comparison to the GMM. It is effective in accurately modelling the probability distribution of the rate of change of TLE semi-major axis prediction error. (2) Maneuver detection experiments indicated that the RGMM outperformed the GMM. It had a 60.6% higher recall and 18% higher F1 score than the GMM. Conclusions: The appropriate processing of the anomalous data can improve the model’s performance for maneuver detection using TLE data with errors. RGMM can be used to analyze the movements of space objects and ensure greater safety in executing future complex space missions. We plan to improve the model’s performance in future research by incorporating more advanced algorithms.