Research on spacecraft anomaly detection method based on neural networks

The monitoring of telemetry data of spacecraft in orbit is of great significance for fault detection and early warning. The existing spacecraft monitoring methods are mainly based on rule knowledge and do not make good use of historical data, especially for multi parameter system anomalies. In this paper, a method of anomaly detection based on long-short term memory network(LSTM) which has memory effect is proposed for spacecraft, especially for dynamic parameters. Firstly, a set of telemetry data preprocessing methods are proposed. Secondly, for the modeling of complex system, Pearson correction coefficient method is proposed to reduce the dimension of telemetry parameters. Then, for dynamic parameters, the modeling methods of LSTM neural network are proposed, and the process of anomaly monitoring is given. Finally, using the historical telemetry data of a system, the modeling effect of neural network is analyzed. The results show that the estimated value of the model is consistent with the measured value, and the abnormal changes of the system can be found effectively.