Anomaly Detection in Satellite Telemetry Data Using a Sparse Feature-Based Method.

Area of Science:

• Spacecraft engineering
• Data science
• Artificial intelligence

Background:

• Satellite health monitoring relies heavily on anomaly detection in telemetry data to prevent failures.
• Sparse representation techniques offer promising avenues for anomaly detection, but their application in satellite systems is nascent.

Purpose of the Study:

• To propose a novel sparse feature-based anomaly detection (SFAD) method for identifying hybrid anomalies in satellite telemetry data.
• To enhance the reliability and safety of satellite operations through improved anomaly detection.

Main Methods:

• Utilized K-means Singular Value Decomposition (K-SVD) to generate a telemetry data dictionary and sparse matrix.
• Defined sparse features capturing local dynamics and co-occurrence relations in multivariate time series.
• Employed a one-class support vector machine (OCSVM) with lower-dimensional sparse features for anomaly classification.

Main Results:

• The proposed SFAD method demonstrated improved detection precision and F1-score compared to existing multivariate anomaly detection techniques.
• The False Positive Rate (FPR) was reduced, indicating higher accuracy in anomaly identification.
• Case analysis on satellite antenna telemetry data validated the method's effectiveness.

Conclusions:

• The SFAD method offers a robust and effective approach for detecting hybrid anomalies in satellite telemetry.
• This technique contributes to advancing anomaly detection capabilities in critical space missions.
• The findings suggest significant potential for sparse representation methods in ensuring satellite operational integrity.