Enhancing interpretability in data-driven battery capacity estimation through degradation mode analysis

Abstract

To explore the relationship between aging features and battery capacity, this study utilizes degradation modes (DMs) as bridging features to construct a simple yet robust capacity estimation model. DMs explain the impact of aging mechanisms on capacity loss, leading to the DM–capacity model (Model 2). Moreover, the aging information reflected by the features is validated through DMs, resulting in the feature–DM model (Model 1). The single-cell model (SCM), composed of these two submodels, is validated on a dataset of 12 cells, which are paired into 6 parallel packs undergoing cycling experiments under various thermal gradient conditions. The validation errors are less than 0.06 Ah in terms of the root mean square error (RMSE), confirming the model's robustness and generalizability. Owing to Model 2 being suitable for packs, the SCM is adapted for estimating battery pack capacity by retraining Model 1 on the basis of pack data, with RMSEs below 0.06 Ah. This finding indicates that the proposed method reduces the modeling effort for pack capacity estimation when considering interpretability. When thermal gradients exist in the battery pack, adding temperature features as inputs in addition to the aging features compensates for the missing condition information, achieving 0.008 Ah RMSE at capacity estimation. This work offers a valuable reference for practical battery pack capacity estimation.