Physics Informed Neural Networks for Reliable SOC Estimation in Lithium-Ion Battery Management

Physics Informed neural networks is gaining popularity by, providing more reliable SOC estimates as, traditional SOC estimation methods struggle with varying conditions and data quality. This paper presents SOC assessment of Lithium-ion cells integrating physical models and machine learning techniques. An equivalent circuit model has been created and validated through experiments, which is used to produce a substantial dataset, including voltage, current, and temperature measurements. This dataset is generated across a wide range of operating conditions, encompassing different load profiles and temperature variations, and is incorporated into the loss function as part of the NN framework. This innovation is essential for advancing EV technology and meeting the growing demands for efficiency and sustainability. SOC estimation is crucial for optimizing the performance, safety, and lifespan of lithium-ion cells in EVs.