Anomaly Detection in Time Series Satellite Data Using a Deep Learning Method
Xiaodong Han, Yakun Wang, Xiao Zhang, Nan Xu · 2023
Minimizing satellite failures and maintaining orbital health is crucial. Current systems spot certain anomalies, relying on expert insight. This paper suggests a data-driven anomaly detection framework, using Deviation Divide Mean over Neighbors (DDMN) to counter fake anomalies from data errors. Long Short-Term Memory (LSTM) models multifaceted time-series data, with a Gaussian model for anomalies. We applied our approach to a two year’s telemetry data from on in-orbit satellites and demonstrate its superiority. This approach is operational in ground stations for real-time monitoring.