基于LSTM网络模型的航天器热变形预测

Thermal Deformation Prediction of Spacecraft based on LSTM Network

针对在轨航天器结构热变形问题,提出了一种基于贝叶斯优化的长短期记忆(LSTM)网络预测方法。结合航天器在轨温度测量数据和有限元模型仿真数据,完成了温度测点选取、LSTM网络模型结构设计、模型训练及热变形预测过程算法实现。实现了较少温度测点为输入下的航天器结构热变形预测。通过多组实验算例结果对比,验证了所提出的LSTM网络模型及其超参数优化方法适用于在轨航天器热变形预测,可以达到的最小均方误差为0.0634。

Aiming at the thermal deformation of the spacecraft in orbit,a long short-term memory network prediction method based on Bayesian optimization is proposed.Combining the spacecraft in-orbit temperature measurement data and the finite element model simulation analysis data,the temperature measurement point selection,LSTM network structure design,model training and thermal deformation prediction process algorithm implementation are completed.The thermal deformation prediction of the spacecraft with fewer temperature measurement points as input data has been completed.Through the comparison of the results of multiple sets of experiments examples,it is verified that the proposed LSTM network model and its hyperparameter optimization method are suitable for spacecraft thermal deformation prediction.The minimum mean square error of the test set is 0.0634.