| Title |
Advanced Battery Equivalent Circuit Model Parameter Optimization and SOC Estimation Based on Levenberg-Marquardt Algorithm Using Parameter Transformation |
| Authors |
홍기준(Gi-Jun Hong) ; 공태현(Tae-Hyeon Gong) ; 이상승(Sang-Seung Lee) ; 김태윤(Tae-Yoon Kim) ; 김종훈(Jong-Hoon Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.8.1801 |
| Keywords |
Lithium-ion battery; State-of-Charge(SOC); Levenberg-Marquardt(LM) algorithm; Parameter transformation; Extended Kalman filter(EKF); Equivalent circuit model(ECM) |
| Abstract |
This paper proposes an offline parameter identification framework based on the Levenberg-Marquardt (LM) algorithm with parameter transformation techniques to improve the accuracy and numerical stability of one-RC equivalent circuit model (ECM) parameter optimization for lithium-ion batteries. Since ECM-based state-of-charge (SOC) estimation accuracy is strongly dependent on parameter quality, it is important to mitigate the numerical instability caused by the large scale disparity between polarization resistance ? ??? and capacitance ? ??? . To address this issue, this study compares two transformation-based methods: log-scale transformation (LM-Log) and sigmoid function transformation (LM-Sig). LM-Log mitigates scale imbalance through relative rate-of-change updates in logarithmic space, while LM-Sig provides both scale normalization and absolute parameter bounds through sigmoid mapping. An extended Kalman filter (EKF) is constructed using the identified parameters, and terminal voltage and SOC estimation performance is evaluated under two scenarios: an OCV-based pulse discharge profile for a Samsung SDI INR21700-50E cylindrical cell and a photovoltaic energy storage system (PVESS) dynamic charge-discharge profile for an SK pouch cell. In the PVESS validation, the parameter set identified and optimized from the OCV test is applied without additional re-optimization. The results show that LM-Sig achieves the best overall performance in both scenarios. For the INR21700-50E cell, LM-Sig reduces the maximum terminal voltage and SOC errors by 63.4% and 68.4%, respectively, compared with non-optimized parameters. For the SK pouch cell under the PVESS profile, LM-Sig reduces the maximum terminal voltage and SOC errors by 14.0% and 21.5%, respectively. These results indicate that the proposed framework can provide physically reasonable and numerically stable initial parameters for ECM-based SOC estimation, and may serve as a useful basis for subsequent online adaptive algorithms in applications where post-deployment maintenance is structurally constrained. |