Abstract
Lithium-ion battery energy storage systems (BESS) are essential components for stabilizing microgrids integrated with intermittent renewable energy resources. Accurate, concurrent estimation of State of Charge (SOC) and State of Health (SOH) is crucial for real-time energy management, operational safety, and optimal asset degradation control. Traditional data-driven estimation techniques often suffer from poor generalization under unseen load profiles, whereas physics-based electrochemical or equivalent circuit models face severe parameter drift and computational overhead over long operational horizons. To resolve these challenges, this study presents a novel Physics-Informed Neural Network (PINN) framework for simultaneous SOC and SOH estimation tailored to microgrid applications. By embedding governing electro-thermal dynamic equations and capacity decay formulations directly into the loss function of a Recurrent Neural Network, the proposed model maintains strict physical consistency while benefiting from high-dimensional data fitting. The framework was validated using realistic microgrid load duty cycles across diverse thermal conditions (-10°C to 45°C). Experimental results demonstrate that the PINN model achieves a Root Mean Square Error (RMSE) below 1.08% for SOC and below 0.76% for SOH co-estimation. Furthermore, the embedding of physical laws reduces data reliance by 60% compared to unconstrained deep learning models and provides superior noise immunity under severe measurement disturbances.
