Accurate State of Charge (SOC) estimation is a critical function for ensuring the safety, performance, and longevity of lithium-ion batteries used in electric vehicles. Traditional model-based estimation methods—such as coulomb counting and Kalman filtering—often suffer from cumulative errors, parameter drift, and poor adaptability under varying temperature and dynamic load conditions. In contrast, data-driven approaches provide a promising alternative by learning complex nonlinear relationships directly from experimental data without relying on predefined equivalent circuit or electrochemical models. This study introduces an HPPC-derived feature-augmented deep neural network (DNN) framework that leverages the strengths of data-driven learning for accurate and robust SOC estimation. The proposed DNN is trained exclusively on Hybrid Pulse Power Characterization (HPPC) data and enhanced through feature engineering that captures voltage recovery dynamics, internal resistance, and temperature effects. Unlike conventional HPPC-based approaches, the model demonstrates strong generalization by accurately predicting SOC during EPA Urban Dynamometer Driving Schedule (UDDS) cycles—despite no exposure to driving-cycle data during training. Experimental validation on a 14-cell lithium-ion battery pack, managed by an L9963E-based battery management system, shows that the proposed model achieves a root mean square error (RMSE) reduction of 0.72% compared to traditional estimation techniques. These results confirm that data-driven DNN models trained on HPPC-derived features can offer high-accuracy, scalable, and generalizable SOC estimation solutions suitable for real-time deployment in next-generation automotive battery management system (BMS) platforms.
Mühendis et al. (Sun,) studied this question.