Improving State of Charge Estimation Accuracy in Sodium‐Ion Batteries Using Optimized Neural Networks

ABSTRACT Accurate State of Charge (SOC) estimation is vital for the safe and reliable operation of sodium‐ion (Na‐ion) batteries. This paper proposes an Improved Feed‐Forward Neural Network (IFFNN) that incorporates physics‐motivated electrochemical features, including voltage, current, auxiliary temperature, and computed internal resistance, into a deeper multi‐layer architecture with dropout regularization for robust state‐of‐charge estimation of Na‐ion batteries. The model is evaluated on driving‐cycle datasets from 3.2 Ah and 10 Ah Na‐ion cells across temperatures ranging from −5°C to 45°C. Experimental results demonstrate that the IFFNN consistently outperforms a FFNN, achieving up to 47.3% reduction in MAE and 35.5% reduction in RMSE, while exhibiting higher R 2 values. Feature sensitivity analysis identifies internal resistance as the dominant contributor to improved accuracy. The proposed IFFNN provides a robust, computationally efficient framework for real‐time SOC estimation in Battery Management Systems, effectively capturing nonlinear electrochemical dynamics such as voltage–current coupling and temperature‐dependent resistance.