State of Health Estimation of Lithium Batteries Based on a Transformer with Physical Constraints and Multidimensional Feature Fusion
Hao Guo , Jinye Lu , Kun Wang , Puming Yu , Huajun Dong , Kai Wang
To address the key bottlenecks in the State of Health prediction of lithium-ion batteries under random variable load conditions, such as poor physical consistency, insufficient feature representation, and non-physical fluctuations in the prediction curves, this paper proposes a high-precision State of Health prediction model that combines physical information constraints and multi-dimensional features, aiming to enhance the safety and long-term reliability of the battery management system. A neural network-transformer framework based on physical information was constructed, converting the consistency laws of irreversible battery aging into regularization loss terms in the training process to ensure physical rationality. Multi-dimensional coupled features including voltage, current, and cycle count were extracted to characterize the aging characteristics of the battery, and Bayesian optimization was applied to adaptively adjust key hyperparameters. Experimental results show that this model outperforms the comparison models in terms of MAE, RMSE, and R² metrics, effectively suppressing non-physical fluctuations, balancing high prediction accuracy and physical rationality, and providing technical support for the full life cycle health management and safe operation of lithium-ion batteries.
Lithium-ion battery / State of health / Physics-informed neural network / Multi-dimensional feature fusion / Bayesian optimization
/
| 〈 |
|
〉 |
PDF
(1192KB)
