Forecasting Methods of Battery Charge and Discharge Current Profile for LEO Satellites

Abstract

The orbital characteristics of low Earth orbit (LEO) satellite systems prevent continuous monitoring because ground access time is limited. For this reason, the development of simulators for predicting satellite states for the entire orbit is required. Power-related prediction is one of the important LEO satellite simulations because it is directly related to the lifespan and mission of the satellite. Accurate predictions of the charge and discharge current of a power system’s battery are essential for fault management design, mission design, and expansion of LEO satellites. However, it is difficult to accurately predict the battery power demand and charging of LEO satellites because they have nonlinear characteristics that depend on the satellite’s attitude, season, orbit, mission, and operating period. Therefore, this paper proposes a novel battery charge and discharge current prediction technique using the bidirectional long short-term memory (Bi-LSTM) model for the development of a LEO satellite power simulator. The prediction performance is demonstrated by applying the proposed technique to the KOM-SAT-3A and KOMSAT-5 satellites operating in real orbits. As a result, the prediction accuracy of the proposed Bi-LSTM shows root mean square error (RMSE) within 2.3 A, and the prediction error well outperforms the most recent the probability-based SARIMA model.