A hybrid LSTM-transformer model for accurate
With the widespread application of lithium-ion batteries in electric vehicles and energy storage systems, health monitoring and remaining useful life prediction have become critical components of battery management systems. To address the challenges posed by the high nonlinearity and long-term depen...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-08-01
|
| Series: | Frontiers in Electronics |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/felec.2025.1654344/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | With the widespread application of lithium-ion batteries in electric vehicles and energy storage systems, health monitoring and remaining useful life prediction have become critical components of battery management systems. To address the challenges posed by the high nonlinearity and long-term dependency in battery degradation modeling, this paper proposes a deep hybrid architecture that integrates Long Short-Term Memory networks with Transformer mechanisms, aiming to improve the accuracy and robustness of RUL prediction. Firstly, time-series samples are constructed from raw battery data, and physically consistent temperature-derived features—including average temperature, temperature range, and temperature fluctuation—are engineered. Data preprocessing is performed using standardization and Yeo-Johnson transformation. The model employs LSTM modules to capture local temporal patterns, while the Transformer modules extract global dependencies through multi-head self-attention mechanisms. These complementary features are fused to enable joint modeling of battery health states. The regression task is optimized using the Mean Squared Error loss function and trained with the Adam optimizer. Experimental results on the MIT battery dataset demonstrate the proposed model achieves excellent performance in a 7-step multi-point prediction task, with a Root Mean Square Error of 0.0085, Mean Absolute Percentage Error of 0.0200, and a coefficient of determination of 0.9902. Compared with alternative models such as MC-LSTM and XGBoost-LSTM, the proposed model exhibits superior accuracy and stability. Residual analysis and visualization further confirm the model’s unbiased and stable predictive capability. This study shows that the LSTM-Transformer hybrid architecture offers significant potential in modeling complex battery degradation processes and enhancing RUL prediction accuracy, providing effective technical support for the development of intelligent battery health management systems. |
|---|---|
| ISSN: | 2673-5857 |