AN LSTM Based Detection Algorithm for Space Target Maneuvering Using TLE Data

With the rapid increase in the number of near-Earth orbit satellites, especially the large-scale deployment of low-orbit mega-constellations, the detection of orbital maneuvers has become a critical technology for space traffic safety management, primarily satellites. Among traditional algorithms for space target maneuver detection, threshold-based methods are highly sensitive to complex spatial disturbances and prone to high false alarm rates. Detection methods based on fixed dynamic models heavily rely on model accuracy, model mismatch or unknown maneuver patterns easily lead to missed detections, which making them ill-suited for the variable real space environment. The physical meaning and calculation process of orbital elements are introduced. Then main theories and algorithms for maneuver detection are analyzed and advantages and disadvantages of different methods are compared. After the algorithmic principles of Long Short-Term Memory (LSTM) networks and their advantages in maneuver detection are analyzed, a maneuver detection method for space objects based on LSTM networks are processed. By analyzing changes in orbital elements between adjacent periods in Two-Line Element (TLE) data as features of space object maneuvers, the LSTM model is trained using TLE data to develop an LSTM-based algorithm for detecting orbital maneuvers. Tests on TLE data demonstrate that the proposed LSTM algorithm achieves higher detection accuracy compared to traditional methods, offering a new solution for orbit anomaly detection in large-scale satellite constellations.