Anomaly Detection for Slowly Varying Analog Telemetry Data of Spacecraft Based on Autoencoder

Spacecraft test stage faces a large number of slowly varying analog telemetry data, such as temperature telemetry value and so on. The traditional method based on the upper and lower threshold is difficult to effectively detect the slow changing trend of the data, but the manual labeling method for anomaly detection has high labor cost and poor consistency. In view of the fact that there are few or no labels for abnormal data of slowly varying analog telemetry data and its structural characteristics,an anomaly detection method of telemetry data using auto-encoder is proposed in this paper. Firstly, an autoencoder is trained by historical data, and then the trained auto-encoder encodes and decodes the verification data to realize restoration and reconstruction. Finally, through adaptive threshold discrimination method, the deviation between the reconstructed data and the original data is detected. Experiments are carried out on the telemetry data of the temperature sensor of a spacecraft, and the results show that the proposed method has the ability of anomaly detection.