Long-term degradation prediction and assessment with heteroscedasticity telemetry data based on GRU-GARCH and MD hybrid method: An application for satellite

Degradation prediction and state-of-health assessment are core information support for on-orbit operation management of satellite. However, the multi-task and on-orbit operation with changeable and complex space environment bring great challenges to predicting and assessing the long-term degradation of satellite. The seasonal variation and heteroscedasticity of status-related telemetry data will lead to a large deviation in the prediction. While, traditional threshold baselines for satellites assessing are simple curves, which are not sensitive to catch the state information of degradation data. To solve these problems, we propose a prediction-assessment methodology which containing data-preprocessing, prediction strategy-model and assessment strategy-model. Data-preprocessing including the elimination of wild values, moving average and Seasonal-Trend decomposition based on Loess (STL), which can accurately dig out the degradation and fluctuation characteristics of satellite long-term telemetry data. The Gate Recurrent Unit (GRU)-Generalized Autoregressive Conditional Heteroscedasticity (GARCH) fusion model is proposed to cognize different fluctuation characteristics and further improve the prediction accuracy. Subsequently, an online updateable health baseline construction strategy is proposed, which is effective and sensitive than traditional monitoring means. Based on Mahalanobis distance (MD) and the constructed health baseline, a health assessment model GRU-GARCH-MD is proposed. To comprehensively verify the capabilities of the proposed prediction and health assessment methodology, real telemetry data is quoted from a certain satellite in Sun-synchronous orbit. The prediction in real application can get a result with MAPE of 2.16% and RMSE of 0.5636 indicate the accuracy and stability of the proposed model, meanwhile, the proposed health baseline construction strategy is superior to the other state-of-the-art methods. (C) 2021 Elsevier Masson SAS. All rights reserved.

Automated summary

A methodology for satellite prediction and health assessment is proposed, including data-preprocessing, prediction strategy-model and assessment strategy-model. The methodology accurately predicts degradation through elimination of wild values, seasonal-trend decomposition and GRU-GARCH fusion model. An online updateable health baseline construction strategy is introduced for sensitive health assessment.