Physics Informed Deep Neural Network for SOC Prediction for Lithium-Ion Batteries

In this work, we discuss a better method for State of Charge (SOC) estimation in lithium-ion batteries through the use of Physics-Informed Neural Networks (PINNs). SOC estimation is critical for battery longevity, efficiency, and safety, but conventional methods such as Coulomb Counting and Equivalent Circuit Models (ECMs) tend to encounter inaccuracies due to aging and environmental variations. Although machine learning (ML) models provide enhanced accuracy, they need extensive data and are non-physical in nature. Overcoming these, PINNs use battery electrochemistry principles and integrate them with deep learning to achieve more robust and flexible SOC estimates. By embedding physical equations in the learning, PINNs eliminate data dependency along with enhancing the generalization to varying conditions. The findings indicate that PINNs surpass both conventional and ML-based methods and are a potential tool for real-time Battery Management Systems (BMS). The computational efficiency needs to be optimized in future work to facilitate practical application in electric vehicles and energy storage systems.