Capacity estimation of lithium-ion batteries using convolutional neural network and impedance spectra

Abstract

Battery capacity is a parameter that has a very close association with the state of health (SoH) of a Li-ion battery. Due to the complex electrochemical mechanisms behind the degradation of battery life, the estimation of SoH encounters many difficulties. To date, experiment-based methods, model-based methods, and data-driven models have been developed to estimate capacity using data from charging curves. In the case of EVs and HEVs, due to the unpredictable charge patterns employed by users, it is not known how much charging data will be available to predict the capacity at a given point in time. This paper presents a method to accurately estimate capacity using impedance curves obtained from an electrochemical impedance spectroscopy (EIS) test and a convolutional neural network (CNN). The CNN model was trained using the impedance data of a single cell collected at different SoH values, and the trained model was verified using the impedance data of eight other cells. The maximum error in the prediction was found to be 0.57 (% capacity) and the RMS error was found to be 0.233 (% capacity). These results illustrate the accuracy and robustness of the proposed model.