SOC estimation of lithium-ion batteries using equivalent circuit model and Nyquist plots from EIS data: A machine learning approach

Abstract

Lithium-ion batteries (LIBs) are currently the most widely used new energy storage devices, whose state of charge (SOC) estimation is critical for their safe operation. Electrochemical impedance spectroscopy (EIS) reveals detailed characteristics of the LIB's electrochemical state, making it useful for SOC estimation. This paper proposes a SOC estimation method based on random forest (RF) combined with a convolutional neural network (CNN) (RF-CNN algorithm) using an equivalent circuit model (ECM) and Nyquist plots from EIS data. Firstly, the ECM parameters are fitted from the 1D EIS data. Then, CNNs are employed to extract the image features (shapes and edges) from the 2D Nyquist plot of EIS data. Finally, the fitted ECM parameters, along with the extracted image features, serve as inputs for the RF algorithm, in which Optuna is utilized for hyperparameter tuning to refine SOC estimation. Experiments on open-access EIS datasets of LIBs demonstrate that the proposed SOC estimation method achieves the best performance in terms of accuracy and speed with a determination coefficient of 0.9926 in 5-fold cross-validation. By integrating 1D ECM parameters with 2D Nyquist plot features, this paper establishes an effective SOC estimation method for LIBs based on the RF-CNN machine learning approach and has important reference values for battery SOC estimation based on small-sample EIS datasets.