Forecasting Sudden Changes in Telemetry Data from Satellite Control System

Classical multi-step telemetry data prediction methods have low accuracy when predicting data patterns with sudden changes, and can be hardly applied to fault detection. We propose a Composite LSTM Autoencoder Model with Attention Mechanism (CLAMA) to realize multi-step prediction of satellite telemetry data patterns with sudden changes. The proposed method learns correlated multivariate parameters, and automatically extracts features and predicts telemetry sequence. The result of experiments based on data sampled from quantum satellite shows that the proposed method has smaller deviation in the root mean square error (RMSE) comparing with the vector autoregression (VAR) model and support vector regression (SVR) model, it can effectively predicts sudden changes in telemetry data, and has higher accuracy for multi-step prediction, thus support ground operators' decision making on fault detection.