Hybrid Physical and Data-Driven State of Charge Estimation for Lithium-Ion Batteries Under Low-Temperature and Noisy Conditions
- 1School of Automation, Guangxi University of Science and Technology, 2 Wenchang Road, Chengzhong District, Liuzhou 545006, China
- 2School of Electronic Engineering, Guangxi University of Science and Technology, 2 Wenchang Road, Chengzhong District, Liuzhou 545006, China
Abstract
Accurate state-of-charge (SOC) estimation is essential for lithium-ion battery management, but low temperatures, dynamic loads, and measurement disturbances challenge both model-based and data-driven estimators. This study proposes a physics-gated multi-scale temporal convolutional network (PG-MS-TCN) for SOC estimation from −20 to 40 °C. Within an established second-order RC modeling framework, an adaptive forgetting-factor recursive least-squares scheme adjusts parameter updates according to voltage innovation and current increments. The resulting circuit-derived features are smoothed using a dual-threshold residual-adaptive exponentially weighted moving average filter, and a temperature-aware two-dimensional open-circuit-voltage-to-SOC mapping provides the physical baseline. To learn the nonlinear residual between this baseline and the reference SOC, PG-MS-TCN employs a cross-domain gating mechanism in which dynamic physical states modulate external measurement features. This mechanism is designed to attenuate transient voltage anomalies during severe polarization, while multi-scale causal convolutions capture temporal signatures of multi-frequency electrochemical dynamics. Experiments on the LG HG2 battery dataset yield an overall root mean square error (RMSE) of 1.18% under interference-free conditions and 1.38% under the specified current disturbances. The average processing time is 0.021 ms per sample. These results demonstrate SOC estimation accuracy and computational efficiency under the evaluated conditions.
Related
- State of Charge Estimation for Lithium-Ion Batteries Under Wide Temperature Ranges Using Error Feedforward of Sparrow-Optimized Back Propagation Neural Network
- Generative Data Augmentation and Sparse-Attention Informer for Cross-Temperature SOH Estimation of Lithium-Ion Batteries
- A Physics-Informed TCN-LSTM Approach for Interpretable Internal Temperature Estimation of NMC811 Lithium-Ion Batteries
- Multi-Stage State-of-Health Estimation for Lithium-Ion Batteries Based on an Adaptive SHAP-Guided Dual-Attention LSTM
- Physics-Informed Explainable Machine Learning for Hybrid State of Health and Remaining Useful Life Estimation of Lithium-Ion Batteries
BibTeX
@article{duan2026hybrid,
title = {{Hybrid Physical and Data-Driven State of Charge Estimation for Lithium-Ion Batteries Under Low-Temperature and Noisy Conditions}},
author = {Yafan Duan and Shengyong Liu and Anqiong Tang and Shenglong Yang and Qipei Gan},
journal = {Batteries},
year = {2026},
month = nov,
doi = {10.3390/batteries12110476},
url = {https://www.mdpi.com/2313-0105/12/11/476},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Hybrid Physical and Data-Driven State of Charge Estimation for Lithium-Ion Batteries Under Low-Temperature and Noisy Conditions
AU - Yafan Duan
AU - Shengyong Liu
AU - Anqiong Tang
AU - Shenglong Yang
AU - Qipei Gan
PY - 2026
DA - 2026/11/06
JO - Batteries
DO - 10.3390/batteries12110476
UR - https://www.mdpi.com/2313-0105/12/11/476
AB - Accurate state-of-charge (SOC) estimation is essential for lithium-ion battery management, but low temperatures, dynamic loads, and measurement disturbances challenge both model-based and data-driven estimators. This study proposes a physics-gated multi-scale temporal convolutional network (PG-MS-TCN) for SOC estimation from −20 to 40 °C. Within an established second-order RC modeling framework, an adaptive forgetting-factor recursive least-squares scheme adjusts parameter updates according to voltage innovation and current increments. The resulting circuit-derived features are smoothed using a dual-threshold residual-adaptive exponentially weighted moving average filter, and a temperature-aware two-dimensional open-circuit-voltage-to-SOC mapping provides the physical baseline. To learn the nonlinear residual between this baseline and the reference SOC, PG-MS-TCN employs a cross-domain gating mechanism in which dynamic physical states modulate external measurement features. This mechanism is designed to attenuate transient voltage anomalies during severe polarization, while multi-scale causal convolutions capture temporal signatures of multi-frequency electrochemical dynamics. Experiments on the LG HG2 battery dataset yield an overall root mean square error (RMSE) of 1.18% under interference-free conditions and 1.38% under the specified current disturbances. The average processing time is 0.021 ms per sample. These results demonstrate SOC estimation accuracy and computational efficiency under the evaluated conditions.
SN - 2313-0105
ER -