Real-Time Lithium-Ion Battery Capacity Curve Recovery and High-Precision RUL Prediction in the Presence of Capacity Data Loss

Lithium batteries have been widely used in industrial fields. Accurate evaluation of the health status and remaining useful life of lithium-ion batteries is crucial for saving operational and maintenance costs and preventing unexpected accidents. Battery capacity is regarded as a valuable indicator for the degradation process of lithium-ion batteries over an extended period. However, in practical industrial scenarios, it is challenging to achieve high-sampling-rate real-time measurements of capacity indicators. Sensor failures and noise further contribute to the loss of capacity data, presenting significant obstacles to quantitative research of lithium-ion batteries, which hampers the accurate prediction of remaining useful life. In this paper, a high-precision method is proposed for recovering lost capacity indicators and providing real-time predictions of battery remaining useful life based on the recovered curves, by utilizing indicators that can readily be measured during the operation of lithium-ion batteries. Empirical Mode Decomposition (EMD) and Long Short-Term Memory (LSTM) neural networks are incorporated into our framework to achieve high-precision predictions of battery remaining useful life. Experiments demonstrate that our approach outperforms previous methods with reduced deviation from the true capacity curve. Especially, crucial battery physical mechanism information can be effectively restored, such as capacity regeneration points (CRPs), which are ignored by prior algorithms. Finally, we utilize the recovered capacity curve to predict the degradation process of lithium-ion batteries and achieve a high-level prediction accuracy.