Improved deep learning based telemetry data anomaly detection to enhance spacecraft operation reliability

Abstract Spacecraft is a complex system integrating a large number of electronic components and payloads. During the in-orbit operation, abnormal events often occur due to the influences of space environment, performance degradation and other factors. These anomalies affect the operational reliability of spacecraft system in orbit. The telemetry data of spacecraft is the main basis to determine its in-orbit state. Data-driven telemetry data anomaly detection method can timely detect the abnormal state of spacecraft system, which provide reference for ground maintenance and ensure the safety and reliability of operation as well as the spacecraft itself. This paper proposes an improved deep learning based anomaly detection method for the anomaly detection of spacecraft telemetry data. Especially, the highly nonlinear modeling and predicting ability of Long Short-Term Memory (LSTM) networks are combined with multi-scale anomaly detection strategy to increase the detection performance. The effectiveness of the proposed method is verified using the NASA benchmark spacecraft data and the hydrogen clock data of the Beidou Navigation Satellite.