34 / 2025-03-28 09:51:57

Safety assessment of lithium battery health state based on multi-scale feature quantification and ELA-Mamba-GA model

State of Health, Mamba, Lithium Battery, Feature quantification, Genetic algorithm

Accurate estimation of the state of health (SOH) of lithium batteries is crucial for the performance optimisation and safe operation of battery management systems, however, the description of the capacity degradation process is often inaccurate, leading to errors in SOH analysis. There exists a strong connection between indirect characteristics and SOH, which can reflect the state of remaining life and thus provide an important basis for remaining life assessment. For this reason, this paper proposes a data-driven SOH estimation method, which analyses the relationship between feature curves in the full life cycle of lithium batteries through maximum information coefficient (MIC) analysis, and screens out the feature categories that are the most contributing to SOH assessment. Combining the multi-scale feature quantification technique and the hybrid model of efficient local attention (ELA) and Mamba Neural Network for training, the genetic algorithm (GA) is also used to optimise the model parameters to achieve high-precision prediction. The experiments are validated based on the NASA lithium battery degradation dataset, and the results show that the method outperforms the traditional method in mean absolute error (MAE) metric, and exhibits good robustness. This demonstrates the effectiveness and practicality of the proposed method in SOH estimation for lithium batteries.