RUL and SOH Prediction for Li-ion Batteries With a Hybrid Model Using Multi-source Feature Selection and Mixed Quadratic Modal Decomposition-DBN

Lithium-ion batteries (LIBs) encounter multiple challenges, including difficulty in online monitoring of internal parameters and intricate capacity degradation mechanisms. Therefore, accurately predicting the state of health (SOH) and remaining useful life (RUL) of LIBs can reduce the risk of battery failure, extend their cycle life, and enable data-driven prognostics and health management (PHM). This study proposes a hybrid collaborative prediction model based on multi-source feature selection and mixed quadratic modal decomposition-deep belief network (DBN). Specifically, the indirect health indicators (IHIs) of LIBs are extracted from the constant current (CC), constant voltage (CV), impedance cycle curves, and their latent information. The multi-source feature selection is conducted on 20 different types of IHIs using Pearson and Kendall correlation analyses. To characterize capacity degradation and nonstationarity, a mixed quadratic modal decomposition model is developed. This model employs multivariate variational mode decomposition (MVMD) and improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) for the optimal IHIs. Finally, a DBN model is employed to perform collaborative indirect prediction of the SOH and RUL for batteries B5, B6, B7, and B18. Experimental results indicate that the proposed model has a good predictive accuracy and generalization ability, effectively predicting the RUL and SOH of LIBs.