Federated Learning for Satellite Telemetry

Satellite telemetry data is collected space operations and deep learning techniques are applied on these data for anomaly detection. However, different satellite missions operate under different environment conditions but data from different missions or companies cannot share their raw data. So, federated learning comes as a tool for utilization to produce a global model that is able to detect anomaly for various operating conditions. But satellite telemetry data is multivariate(voltage, temperature etc.) and onboard satellite units have limited power. So, in this work, we try to apply and benchmark different green federated learning techniques and determine which combination of techniques works best for this particular scenario. For telemetry dataset, we use the ESA-ADB dataset from European Space Agency.