Hybrid Anomaly Detection in Spacecraft Telemetry Data Using Sparse Feature-Based Methods and Spatial-Temporal Generative Adversarial Networks

Anomaly detection in spacecraft telemetry data is critical for ensuring mission success and operational reliability. However, the high dimensionality, complex temporal dynamics, and multivariate nature of telemetry data pose significant challenges for traditional anomaly detection methods. This paper proposes a hybrid anomaly detection system that
combines Sparse Feature-Based Anomaly Detection (SFAD) and Spatial-Temporal Generative Adversarial Networks (ST-GAN) to address these challenges. The SFAD module reduces dimensionality and extracts sparse features from telemetry data, while the ST-GAN module captures temporal dependencies and spatial correlations between parameters. Additionally, an adaptive thresholding mechanism is introduced to dynamically adjust the anomaly detection threshold, reducing false positives and improving robustness. The proposed system is evaluated on the SMAP and MSL datasets, demonstrating superior performance in terms of Precision, Recall, and F1-Score compared to state-of-the-art methods such as LSTM-GAN, GRU-VAE, and Isolation Forest. The results show that the hybrid approach is particularly effective at detecting multivariate and contextual anomalies, which are often missed by traditional methods. The system’s ability to perform near real-time anomaly detection makes it suitable for practical spacecraft monitoring applications. This work contributes to the field of telemetry analysis by providing a robust, scalable, and accurate solution for anomaly detection, with potential applications in other domains such as industrial monitoring and autonomous vehicles.