The rapid expansion of electric vehicles (EVs) and energy storage systems (ESSs) has led to a significant increase in the demand for lithium-ion batteries as well as their technological importance. In particular, in EV and ESS environments where hundr...
The rapid expansion of electric vehicles (EVs) and energy storage systems (ESSs) has led to a significant increase in the demand for lithium-ion batteries as well as their technological importance. In particular, in EV and ESS environments where hundreds to thousands of high-energy-density cells are combined, variations in the characteristics of individual cells can directly affect the safety and efficiency of the entire system, thereby requiring accurate state diagnosis. In such systems, a battery management system (BMS), which interprets battery states and performs protection strategies based on voltage, current, and temperature information, is essential for the stable operation of batteries at the module and pack levels. The BMS processes these measured data in real time to estimate key state variables such as the state of charge (SOC) and the state of health (SOH), and plays a central role in safety control functions including thermal management, power limitation, and fault detection.
However, SOC and SOH are difficult to directly measure, and in battery systems, the measured data are affected by variability depending on operating conditions such as applied current and ambient temperature. As a result, there may be limitations in achieving reliable state estimation of SOC and SOH under diverse operating environments. In addition, BMSs used in vehicles or ESSs must consider constraints related to hardware and software design, including limitations on real-time data storage and logging as well as real-time computation and complex algorithm processing. Therefore, it is difficult to directly apply complex physics-based models or large-scale data-driven models in practical BMS implementations. For these reasons, equivalent circuit models (ECMs), which have simple computational structures, fast execution, and relatively high accuracy, are widely used in BMSs. ECMs approximate the current–voltage response of batteries through the open-circuit voltage (OCV)–SOC relationship, series internal resistance, and one or more RC networks, thereby enabling efficient state estimation and voltage prediction.
However, many practical and conventional ECM approaches assume fixed parameters or update them only in a limited manner, and although online parameter update techniques have been actively studied, limitations still exist in the estimation of states such as SOC and SOH required for precise monitoring within BMSs.
In this study, to address these limitations, an ECM-based state estimation model that reflects aging characteristics is proposed. To extract the parameters constituting the ECM, partial discharge data with durations of 6 min, 3 min, and 1 min were obtained under a 1C nominal current condition, and third-order RC ECM parameters were extracted based on the current and voltage responses obtained at each cycle. Subsequently, a gradient boosting regressor (GBR), a machine learning technique that has the advantage of effectively learning the nonlinear and complex parameter evolution trends observed during battery aging, was employed. By learning the cycle-dependent trends of the parameters, the model was constructed such that SOH estimation and voltage behavior prediction under aged conditions are possible using only the initially extracted parameters. Through this approach, it was confirmed that battery health estimation can be achieved based on short-duration discharge data without requiring large-scale datasets.
Experimental results show that the proposed model first estimates the SOH at the cell level, and the root mean square error (RMSE) between the predicted voltage and the experimentally measured voltage at the corresponding aging states remains below 0.03 V at all aging points. Using only the initial parameters and the learned aging trend model, it was confirmed that SOH estimation reflecting the battery aging state is possible, and by applying ECM parameters corresponding to the estimated SOH, the voltage behavior at the given aging state can be accurately reproduced. Although this study was validated at the cell level, the proposed methodology can be extended as a foundational technology for the development of real-time state diagnosis and lifetime prediction algorithms for future module- and pack-level battery systems.