A Physics-Informed Neural Network for SoC and SoH Estimation in Solar-Charged Li-ion Batteries Under Partial Cycling

Abstract

Accurate estimation of State of Charge (SoC) and State of Health (SoH) is essential for lithium-ion batteries in solar energy storage systems where partial and irregular cycles are common. Traditional methods such as Coulomb counting and Extended Kalman Filters often struggle under dynamic operating conditions, while purely data-driven approaches lack robustness when usage patterns deviate from training data. This paper proposes a Physics-Informed Neural Network (PiNN) framework that embeds electrochemical constraints within a neural architecture to improve SoC and SoH estimation for solar-charged lithium-ion batteries undergoing partial cycling.

A synthetic solar charging scenario is constructed by overlaying real solar irradiance profiles onto publicly available battery datasets, thereby generating incomplete charge–discharge patterns. The proposed PiNN is evaluated against classical filtering methods and purely data-driven neural models. Results demonstrate improved estimation accuracy under fluctuating charging conditions, indicating that the proposed framework offers a practical pathway toward reliable real-time battery management in renewable energy storage systems.