SOH estimation of lithium-ion batteries with local health indicators in multi-stage fast charging protocols

The multi-stage fast charging protocols, with diverse charging rates, induce irregular degradation patterns in lithium-ion batteries, posing formidable challenges to the precise monitoring of state of health. To achieve the accurate state of health estimation in fast-charging protocols, a hybrid deep learning architecture with partial health features is developed. Initially, the data segment is extracted as latent features from charging data without any computational processing, named state of charge window. The latent features are composed of charging data in different constant current modes, which considers the ageing behavior at diverse charging rates. Furthermore, the most correlated health indicator with state of health is obtained with the maximum mutual information coefficient, using only 40 % of charging data. Subsequently, a hybrid data-driven model is constructed, which is employed for the extraction of spatial features, enhancing the weights of critical information, and capturing the relationship between features and state of health. Finally, the datasets of 45 cells are applied to validate the superiority of the proposed method under 9 distinct fast charging protocols. Validation results demonstrate that the developed approach enables precise state of health estimation with root mean square errors and mean absolute errors within 1 %, achieving 53.7 % and 51.5 % improvements respectively compared to the benchmark.