Year: 2025
Type: article
Abstract: 航天器电源系统是保障航天器在轨长期稳定运行的关键分系统。然而，在极端空间环境下，电源系统性能易受影响，传统建模方法难以准确捕捉其复杂非线性、时变及退化特性。针对这一挑战，本文提出一种基于改进即时学习(JITL)与长短期记忆网络(LSTM)的航天器电源模型构建方法。该方法通过引入加权相似度策略，动态选择与当前工况最相关的历史数据，构建局部LSTM模型，并设计自适应更新机制。实验结果表明，所提出的改进JITL-LSTM模型能够有效提升航天器电源系统建模的精度和鲁棒性，为航天器健康管理与故障预测提供有力支持。The spacecraft power system is a critical subsystem that ensures the long-term stable operation of spacecraft in orbit. However, under extreme space environments, the performance of the power system is susceptible to influence, and traditional modeling methods struggle to accurately capture its complex nonlinear, time-varying, and degradation characteristics. To address this challenge, this paper proposes a spacecraft power model construction method based on improved Just-In-Time Learning (JITL) and Long Short-Term Memory (LSTM) networks. This method introduces a weighted similarity strategy to dynamically select historical data most relevant to the current operating conditions, constructs local LSTM models, and designs an adaptive update mechanism to cope with the continuously changing operating modes and degradation trends of the spacecraft power system during in-orbit operation. Experimental results demonstrate that the proposed improved JITL-LSTM model can effectively enhance the accuracy and robustness of spacecraft power system modeling, providing strong support for spacecraft health management and fault prediction.
Source: Modeling and Simulation
Author: 庚 李
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4412773118