Forecasting and Uncertainty Quantification of Simulated Solar Wing Deployment Process under Microgravity Conditions
In the simulated deployment test of the space solar wing mechanism within a ground micro-gravity environment,a methodology is proposed to employ a high-fidelity rigid-flexible coupled model for generating datasets for training neural networks.The objective is to predict the position and attitude of the mechanism to achieve more precise gravity unloading control and identify potential deviations or problems for risk mitigation.The approach entails characterizing the hybrid uncertainty of the observation data by using the probability box(p-box)method and leveraging the variational Bayesian Monte Carlo(VBMC)to establish the high-fidelity rigid-flexible coupling dynamic model through updating the uncertainty parameters.Subsequently,the updated model is utilized to generate the datasets for training and testing the nonlinear auto-regressive(NAR)neural network.Then,the forecasting error set is produced and utilized to train the nonlinear auto-regressive with exogeneous inputs(NARX)neural network.The NAR network is used to predict the attitude parameters of the solar wing deployment process.Then,the predicted values are input into the NARX network to obtain the uncertainty quantification(UQ)interval of the predicted values.Numerical and experimental examples demonstrate that the proposed method possesses high prediction accuracy,reasonable UQ,and fast solution speed,thereby verifying the effectiveness and robustness of the proposed method.
Solar wingBayesian updatingUncertainty quantificationHigh-fidelity modelNeural network
NETL
NSTL
万方数据
