Data-driven In-orbit Current and Voltage Prediction using Bi-LSTM for LEO Satellite Lithium-ion Battery SOC Estimation

Abstract

Accurate estimation of the battery system state of charge (SOC) is essential to the satellite mission design and fault management. However, it is difficult for low Earth orbit (LEO) satellites to continuously monitor the battery SOC on the ground due to the non-contact duration. To estimate the battery SOC for the entire orbit, it is necessary to predict or monitor the battery data for all times. Therefore, existing studies use SOC estimation that relies on real-time onboard battery information or utilizes probability-based technique and power budget-based technique. The real-time onboard-based technique is unsuitable for mission design because the status information is not available to the ground during the non-contact duration. Probability-based and power budget-based techniques are not reliable during the non-contact duration. In this study, we propose the ground-based battery SOC estimation technique that predicts the current and voltage by using bidirectional long short-term memory (Bi-LSTM) network for the non-contact duration and estimates the SOC by unscented Kalman filter (UKF) for all operating conditions. The proposed technique is tested with in-orbit data of the KOMPSAT-3A satellite, and we demonstrate its superior performance than other conventional ground-based SOC estimation techniques.