Remaining Useful Life Prediction of Lithium-Ion Batteries Based on a Combination of Ensemble Empirical Mode Decomposition and Deep Belief Network-Long Short-Term Memory

The prediction of remaining useful life (RUL) for lithium-ion batteries is a critical component of electric vehicle battery management systems. However, during the aging process, batteries exhibit an overall declining trend in capacity curves, coupled with capacity regeneration and localized fluctuations. Directly modeling this degradation trend based on the original capacity curve proves challenging, leading to reduced accuracy in RUL prediction. This article introduces a hybrid method to enhance the precision of battery RUL prediction. Utilizing the ensemble empirical mode decomposition technique, the battery's capacity degradation sequence is decomposed into intrinsic mode functions (IMFs) with varying degrees of fluctuations, along with a residue that characterizes the battery's overall declining trend. Subsequently, deep belief networks and long short-term memory networks are established to predict the residue and IMFs separately. The combined results from these models yield the final battery RUL prediction. Finally, the effectiveness of this approach is validated on the NASA battery dataset, with diverse training periods and prediction time steps. Experimental results demonstrate that the root mean square error of predictions for all four batteries remains below 2%. Utilizing the ensemble empirical mode decomposition technique, the battery's capacity degradation sequence is decomposed into intrinsic mode functions (IMFs) and a residue (RES). Subsequently, deep belief networks and long short-term memory networks are established to predict the RES and IMFs separately. The remaining useful life prediction is obtained by accumulating the above predicted values.image (c) 2024 WILEY-VCH GmbH