A Novel Method Based on Hybridization of Generative Adversarial Imputation Nets and SDAE-Kriging for RUL Prediction of Lithium-Ion Battery in Scenarios of Missing and Incomplete Data

Lithium-ion batteries (LIBs) have received enormous attention as the core components of Electric vehicles (EVs). An unavoidable issue is that battery performance will continue to degrade as materials age and cycle time increases. Accurately predicting the Remaining useful life (RUL) of LIBs is an important prerequisite to ensure the safe driving of EVs. However, the actual battery management system may encounter sensor or communication system failures, resulting in missing or incomplete data, which will result in inaccurate battery RUL predictions. This article presents a novel method based on hybridization of Generative adversarial imputation nets (GAIN) and Stacked denoised autoencoder with Kriging (SDAE-Kriging) for the prediction of RUL of LIBs in scenarios of missing and incomplete data. In the proposed method, the GAIN is leveraged to realize the filling of missing and incomplete data. The SDAE-Kriging is used to predict the RUL of LIBs with filled data. Different missing rates (10%, 20%, 30%, and 40% ) are investigated to establish RUL prediction models. It is manifested that the GAIN has better data filling results on two datasets. After completing the data filling, the results show that SDAE-Kriging has high prediction accuracy (RMSE is less than 0.1) for both datasets, even when the missing rate is 40% . The proposed scheme can provide an effective solution for the RUL prediction of LIBs in the scenario of missing and incomplete data in practical industrial applications.