Simultaneous prediction of impedance spectra and state for lithium-ion batteries from short-term pulses

Electrochemical impedance spectroscopy (EIS) is a versatile tool to characterise lithium-ion batteries. However, EIS measurement is challenging in practice as it needs costly hardware and stringent test requirements. In this study, we propose a data-driven solution to predict battery impedance spectra at different states. An encoder-decoder deep neural network is developed to achieve simultaneous predictions of both impedance spectra and state of charge (SOC) only using short-term pulse data sampled at 1 Hz, thereby precluding the need for specific hardware and alleviating test requirements. A large dataset covering over 2700 impedance spectra over the frequency range of 100 mHz to 10 kHz is established to validate the proposed method at different SOCs, tem-peratures and ageing states. From the validation results, the proposed method enables accurate predictions at different temperatures and ageing levels while the associated errors of impedance spectra and SOC can be restricted within 1.5 m omega and 1.26%, respectively. We further demonstrate that the predicted impedance spectra can provide detailed physical insight into battery kinetics as it offers accurate extractions of critical parameters of an impedance model. Our method makes EIS measurement more accessible to evaluate battery characteristics and highlights the potential of deep learning in battery research.