Similarity based telemetry data recovery for enhancing operating reliability of satellite

Recently, the satellite launching grows rapidly, especially in the commercial aerospace, which produces thousands of in-orbit spacecrafts. The operation and ground management of these satellites is a hotspot issue to ensure the satellites' in-orbit operation reliability. The telemetry data transmission may be delayed by the unstable telemetry link, insufficient ground station sources, interface failures and so on, which exerts a negative influence on the ground management capability and increases the failure risk. Meanwhile, the deep space spacecrafts, space station and Mars exploration expand this transmission delay. Thus, management and intelligent analysis of the telemetry data from Telemetry Ground Station (TGS) gradually become critical for in-orbit satellite operation. To solve above issues, a similarity-based deep learning time series prediction method is proposed to achieve telemetry data recovery. Considering the pseudo-periodic characteristics in telemetry data, this approach extracts similar segments to construct training samples instead of feeding all the observation samples to train the prediction model. As the predicted data is obtained ahead of the actual time-delayed telemetry data, which can provide sufficient analysis time for ground operator to optimize the maintenance strategy. Furthermore, the telemetry prediction data can also guarantee high-complexity digital twin modelling, virtual maintenance and so on. Experimental results with actual satellite telemetry data verify the effectiveness of the proposed method.