(170e) State of Charge Estimation of Lithium-Ion Battery Using Surrogate Model Based on Electrochemical-Thermal Model

To safely and efficiently use a battery, it is essential to accurately estimate the battery's State of Charge(SoC). It is not easy to directly measure the SoC, so it is indirectly estimated through the coulomb counting method or open-circuit voltage(OCV) based method. The coulomb counting method has a limitation in that a difference from an actual value increases when errors accumulate, and the OCV method has a limitation in that the battery must be in an equilibrium state to measure the OCV. Model-based methods and data-driven methods are being studied to overcome the previous methods' limitations and estimate SoC with high accuracy. The models mainly used in the model-based method are the Equivalent Circuit Model and the Electrochemical model. The electrochemical model is not suitable for real-time use due to its high computational complexity. The data-driven method is difficult to guarantee accuracy in cases where data is not enough.
In this study, a surrogate model based on electrochemical-thermal models for SoC estimation is developed to solve computational complexity issues of electrochemical models and data acquisition issues of data-driven models due to data dependence. The electrochemical-thermal model parameter identification is performed using a Genetic Algorithm based on the discharge data. Output variables such as a current, voltage, and temperature for various drive cycle loads are derived from this model to create sufficient data for model training. By using these generated data, the surrogate model is trained based on artificial neural network algorithms Multi-Layer Perceptron(MLP), Long Short-Term Memory(LSTM), and Gated Recurrent Units(GRU). The performance of the surrogate model is compared with data-driven models in terms of accuracy.