Contrastive Learning with Prototype-Based Negative Mixing for Satellite Telemetry Anomaly Detection.

Area of Science:

• Spacecraft engineering
• Artificial intelligence
• Data science

Background:

• Telemetry data is crucial for monitoring satellite health and safety.
• Current deep learning methods struggle to capture complex correlations in telemetry data, limiting anomaly detection accuracy.

Purpose of the Study:

• To develop an advanced anomaly detection framework for satellite telemetry data.
• To improve the accuracy and robustness of identifying anomalies in spacecraft operations.

Main Methods:

• CLPNM-AD (Contrastive Learning with Prototype-based Negative Mixing for Correlation Anomaly Detection) framework.
• Utilizes data augmentation with random feature corruption.
• Employs a consistency strategy for sample prototyping and prototype-based negative mixing contrastive learning.
• Proposes a prototype-based anomaly score function for decision-making.

Main Results:

• CLPNM-AD significantly outperforms baseline methods in anomaly detection.
• Achieved up to an 11.5% improvement in F1 score on public and real-world satellite mission datasets.
• Demonstrated increased robustness against noisy telemetry data.

Conclusions:

• CLPNM-AD effectively models complex correlations in telemetry data for accurate anomaly detection.
• The proposed method enhances satellite reliability and safety through superior anomaly identification.
• CLPNM-AD offers a robust solution for real-time spacecraft monitoring.