Abstract
Perceiving the potential uneven cell aging within battery packs and estimating the pack-level state of health (SOH) remains an unresolved challenge. We propose a data-driven pack-level SOH estimator, PackFormer, to reliably observe performance degradation in battery packs while considering the differences in intrinsic cell aging. PackFormer introduces feature attention, temporal attention, and cell attention mechanisms to capture critical degradation characteristics and uneven aging of individual cells within the battery pack, providing more accurate pack-level SOH estimation for battery management systems. The results show that PackFormer significantly outperforms other popular data-driven baselines in observing the pack-level SOH; the root-mean-square error of pack-level SOH predictions from 4 manufacturers, applying PackFormer, decreases by an average of 61.6%–0.59%. This work highlights the potential of modeling and analyzing the complex dynamics of cell interactions within battery packs using advanced deep learning solutions, offering a new avenue to estimate pack-level SOH.
Original language | English |
---|---|
Article number | 102363 |
Journal | Cell Reports Physical Science |
Volume | 6 |
Issue number | 1 |
DOIs | |
Publication status | Published - 15 Jan 2025 |
Keywords
- artificial intelligence
- batteries
- battery pack
- cell degradation
- deep learning
- state of health
- transformer