A Data-Driven Battery Degradation Estimation Method for Low-Earth-Orbit (LEO) Satellites

Abstract

Battery degradation is a critical challenge in the operation and longevity of low-Earth-orbit (LEO) satellites because of its direct impact on mission reliability and power system performance. This study proposes a data-driven approach to accurately estimating the degradation of satellite batteries by integrating a transformer network model for voltage prediction and unscented Kalman filter (UKF) techniques for online state estimation. By utilizing on-orbit telemetry data and machine-learning-based modeling, the proposed method provides processing-time improvements by addressing the limitations of traditional methods imposed by their reliance on predefined conditions and user expertise. The proposed framework is validated using real satellite telemetry data from KOMPSAT-5, demonstrating its ability to predict battery degradation trends over time and under varying operational conditions. This approach minimizes manual data processing requirements and enables the consistent and precise monitoring of battery health.