| 摘要: |
| 针对航天器在轨运行时表面温度受太空环境辐射影响而产生剧烈波动的问题，开展了用于快速实时预测航天器表面温度场的代理模型构建研究。通过数值计算方法提供训练数据，输入全连接神经网络进行代理建模，建立工况数据到航天器表面温度场的高维快速预测，对深度学习方法、混沌多项式展开和高斯过程进行了对比分析。结果显示，基于深度学习的代理模型在稳态条件下训练效率和预测精度均表现优异，平均绝对误差可降低至0.98 K，并且可在0.1 s 内完成预测。然而，单一的网络结构对复杂多变的环境特征捕捉能力较差，因此对航天器表面辐射热通量变化率较大部位预测误差较大。研究为航天器表面温度提供了实时预测的新方法，促进了深度学习技术在航天领域的应用验证，对保障航天器稳定运行及延长寿命具有重要意义。 |
| 关键词: 航天器 温度场 深度学习 快速预测 代理模型 |
| DOI： |
|
| 基金项目:中国科协青年人才托举工程（2021QNRC001） |
|
| Research on Fast Prediction Method of Spacecraft Surface Temperature Field Based on Surrogate Model |
| LI Xingchen,LI Qiao,ZHOU Weien,WANG Ning,YAO Wen |
| (1.Academy of Military Sciences, Beijing 100071, China;2. Intelligent Game and Decision Laboratory, Beijing 100071, China) |
| Abstract: |
| Aiming at the problem of drastic fluctuation of spacecraft surface temperature caused by radiation from the space environment when the spacecraft is in orbit, a research was conducted on the construction of a surrogate model for rapid and real-time prediction of the spacecraft surface temperature field. The training data are provided by numerical computation method and input into fully connected neural network for surrogate modelling to establish the high-dimensional fast prediction of spacecraft surface temperature field from operating condition parameters to spacecraft surface temperature field, and a comparative analysis among the deep learning-based method, chaotic polynomial expansion and Gaussian process is carried out. The results show that the deep learning-based surrogate model performs well in both training efficiency and prediction accuracy under the condition of steady state, with mean absolute error reduced to 0.98 K, and complete the prediction within 0.1 s. However, the single network structure is less capable of capturing complex and variable environmental features, resulting in larger prediction errors in areas with high variability of spacecraft surface radiation heat flux. The research provides a new method for real-time prediction of spacecraft surface temperature field, promotes the application of deep learning technology in the aerospace field, and is of great significance for ensuring the stable operation and extending the life of spacecraft. |
| Key words: Spacecraft Temperature field Deep learning Rapid prediction Surrogate model |