Research on State of Charge Estimation for Battery Packs Based on Physics-Informed Neural Network

State of Charge (SoC) estimation for battery packs is a critical challenge in Battery Management Systems (BMS), particularly due to cell-to-cell variations in multi-cell configurations. To address this issue, this study proposes a Physics-Informed Neural Network (PINN)-based framework that integrates electrochemical principles with deep learning. By incorporating cell voltage, temperature, and operational parameters, the proposed method achieves high-precision SoC estimation. Experimental results on real-world electric vehicle datasets demonstrate a Mean Absolute Error (MAE) of 0.89% and a Root Mean Square Error (RMSE) of 1.12%, outperforming traditional voltage-based method (72.0% reduction), ablation models (53.3% reduction), Extended Kalman filter (87.7% reduction) and Recurrent Neural Network (86.1% reduction). These findings validate the effectiveness of hybrid physical-data-driven strategies in enhancing BMS reliability under dynamic operating conditions.