Anomaly Detection in Satellite Attitude Control System Based on GAT-TCN

The satellite attitude control system (ACS) is a critical system for satellite mission execution, and its operational status significantly impacts the overall performance of the satellite. Addressing the challenges associated with the ACS, including high coupling, complex causality, and the difficulty of obtaining anomaly data, this paper proposes an unsupervised anomaly detection method based on Graph Attention Network (GAT) and Temporal Convolutional Network (TCN) for monitoring the telemetry data of the ACS. Given the complexity of the ACS and the high-dimensional nature of the data, the proposed method first utilizes GAT to capture the spatial dependencies among multidimensional parameters and reconstruct the monitoring data. Subsequently, TCN is employed to learn the temporal dependencies in the time series of the ACS, integrating partial mutual information to represent causal relationships for data prediction. Finally, based on the reconstruction and prediction residuals of the multidimensional parameters, the method assesses whether anomalies are present in the multidimensional sequences. The proposed method is validated through simulation data of the ACS, with the results demonstrating the effectiveness of the anomaly detection algorithm presented in this study.