A cloud-edge collaborative computing-based method for governing anomalous and missing satellite telemetry data

Satellite telemetry data serves as a critical foundation for diagnosing the operational status of satellites in orbit and predicting their health trends. However, constrained by multiple factors such as disturbances in the complex space environment and attenuation of the satellite-to-ground link, telemetry data frequently suffers from significant anomalies and missing values. Consequently, despite the large volume of telemetry data, its overall quality remains suboptimal. To address this issue, this paper proposes a cloud-edge collaborative computing-based method for governing anomalous and missing telemetry data. At the onboard edge layer, sporadic anomalies and missing data points are handled as follows: the t-SNE data dimensionality reduction visualization algorithm is employed to identify and clean outliers; the spectral clustering algorithm is used to partition parameter thresholds and define exceed identification zones; and a hybrid grey neural network is utilized to achieve data prediction and imputation at the edge. At the ground-based cloud server layer, addressing the frequent occurrence of anomalies and missing data during the satellite's long-term service, a K-Means-GSKNN-based data imputation method is proposed. This method leverages historical telemetry data and fault libraries to train the model, achieving high-precision governance of anomalous and missing data.