Research On Modeling Method Of Data-driven Spacecraft Battery Performance Degradation

The complex space environment is “dangerous”.Sudden factors such as space junk and temperature changes may cause abnormal spacecraft faults at any time.Spacecraft system functions are becoming more and more abundant and perfect,and the degree of integration,intelligence and integration is increasing,resulting in aerospace.The structure of the device is more complicated,and the on-orbit failure rate is significantly increased.The technology of the spacecraft system fault diagnosis and health management(prognostics and health management)with the prediction technology as the core has attracted the attention of many scholars at home and abroad.The spacecraft power system is the most important component of the spacecraft,and battery life is the most important constraint for the spacecraft power system.Therefore,it is of great practical significance and application value to extract the characteristic indicators for measuring the degradation degree of spacecraft battery performance and construct an effective battery performance degradation prediction model.This paper describes the development,features and performance of spacecraft batteries.On this basis,a new characteristic index of performance degradation of spacecraft battery-"end-of-discharge voltage value" is proposed.The effectiveness of this index is verified by NASA data set.This index is applied to the performance degradation modeling research of real satellite batteries.The discharge current,discharge duration,temperature and time are used as the input of the optimized neural network,and the "discharge final voltage value" is used as the output to train the network.In order to study satellite battery performance degradation in the same working conditions and environment,this paper sets the discharge current,discharge capacity,temperature,and discharge duration to constant(ie,does not consider the influence of the above parameters),separate evolution time,and utilizes optimized training.The neural network model outputs "the final discharge voltage value evolved over time",which in turn reflects the deterioration trend of the spacecraft battery performance.The paper studies the performance degradation of lithium-ion batteries in spacecraft power systems.The "capacity regeneration" and fluctuations of lithium batteries in performance degradation make their performance degradation curves very complicated.Traditional time series models(such as autoregressive integrated moving average models)and regression models(such as Gaussian process regression)can not accurately predict the performance degradation trend of lithium batteries,and are not sensitive to "capacity regeneration" and fluctuation phenomena,and long-term prediction is accurate.Therefore,this paper proposes a multi-scale ARIMA and GPR fusion model,using the empirical mode decomposition method to effectively extract the global degradation trend and local "capacity regeneration" and fluctuation phenomena of the battery SOH time series.The fusion model uses the ARIMA model to fit the global trend of the SOH time series,using the improved GPR model to fit the local “capacity regeneration” and fluctuations of the SOH time series,and finally integrates the fitting results of the model.Through case analysis,the fusion model can capture the true performance degradation trend of the battery.The accuracy of long-term prediction is obviously improved,and the 95% confidence interval of the predicted results is more important for the decision-making judgment of managers.