A Comparative Study and Unified Framework for SOC and SOH Estimation of Lithium-Ion Batteries Using Machine Learning Under Thermal and Aging Variations

Precise assessment of State of Charge and State of Health is essential for the safe and effective functioning of lithium-ion batteries, especially under fluctuating temperature circumstances.This study proposes a temperature-dependent machine learning framework to estimate SOC and SOH using multiple ensemble and baseline models. Experimental data comprising voltage profiles, current, time, and SOC values estimated by the Extended Kalman Filter at five different surrounding temperatures (30°C, 25°C, 10°C, 0°C, and -20°C) were used to train and evaluate several machine learning models.Eight regression-based models were employed to estimate SOC and SOH. Results indicate that GPR outperformed all other models in SOC estimation, achieving the lowest error rates (MSE: 0.6118%, MAE: 0.5311%), while XGB and DT demonstrated superior accuracy in SOH estimation (MAE: 0.1355% and 0.1331%, respectively). Conversely, SVR and LR consistently yielded the highest error values across both tasks. The findings suggest that ensemble and probabilistic models provide more reliable performance for battery health diagnostics compared to traditional regression methods. This work contributes to the development of robust battery management systems by identifying optimal machine learning approaches for SOC and SOH estimation.The proposed approach not only enhances predictive accuracy but also offers a scalable framework for real-time battery management systems.