Early and accurate prediction of their degradation behavior helps investigate the lifespan of lithium-ion batteries (LIBs). Charge-discharge cycling data can be used for this purpose, but simultaneous feature extraction regarding LIB’s intra- and inter-cycle behavior, despite its importance, has not been thoroughly investigated. In this study, machine learning methods are proposed to facilitate the simultaneous extraction of intra- and inter-cycle features from VIT cycling datasets for the knees prediction of LIBs. The length of each cycling data is made the same by realigning the data in terms of voltage or using zero-padding, depending on the type of dataset used. The cycling dataset is arranged as an array along the time and cycle axes before applying convolutional neural networks (CNNs) and/or recurrent neural networks (RNNs) for feature extraction. Three ways to use these nonlinear regression tools are explored, which are 2-dimensional CNN (2D CNN), RNN + 1-dimensional CNN (1D CNN), and RNN + 2D CNN. The performances of the resulting knees prediction models are evaluated using the benchmark dataset. Our results confirm the benefit of explicitly considering cycle-to-cycle behavior in conjunction with intra-cycle temporal behavior in building the data-driven prediction model. The possibility of reducing the input data requirement is examined to facilitate the early prediction of knees. It is shown that the input size can be reduced down to the first 60 cycles from 100 cycles with comparable prediction performance.
Supplementary figures, tables, and notes