Life Prediction of Small Satellite Storage Battery Based on Dynamic Bi-step MPNN Method

Abstract

Aiming at the characteristic that small satellite storage battery usually has no enough ground experiment & in-orbit telemetry data, a new Dynamic Bi-step MPNN based method is proposed to solve the problem of life prediction of small satellite storage battery. The method can mainly be divided into two steps: The first named 1st-MPNN process is to establish a modified probability neural network (MPNN) model using the time series data of major life-dependant factors; the second called 2nd-MPNN is to construct another MPNN based on the predicted data from 1st-MPNN. The life prediction can be realized based on current storage battery data and bi-step MPNN, moreover, the bi-step MPNN can also be re-trained and updated for a higher life prediction precision once new in-orbit data is available. Finally, the proposed bi-step MPNN method was applied to the life prediction of HY-1B small satellite storage battery using its ground experiment & in-orbit data. It was found that the method is valuable and pragmatic for life prediction of small satellite storage battery with small sample size.