Capacity and SOH Prediction of Lithium-Ion Batteries Using Various Mathematical Forms of EIS with Neural Networks

Increasing knowledge of lithium-ion battery degradation continues to play a vital role in fields such as consumer and transportation applications for optimal efficiency in power delivery to devices. To accomplish this, recent avenues in battery management systems employ various models that use data to predict a battery’s capacity and its resulting state of health (SOH). A rising data input for this field is electrochemical impedance spectroscopy (EIS) for health performance prediction since it provides information-rich complex data while being non-destructive to the cell. However, there is a lack of research into the different mathematical forms that characterize EIS and its usefulness in simultaneously predicting capacity and SOH. Neural networks are increasingly utilized for predictive modeling of battery degradation with EIS due to their high accuracy when working with non-linear and high-feature data. This research looks at the efficiency of a neural network to predict capacity and SOH simultaneously using EIS through rectangular and polar forms of its complex data structure. Using the most detailed and expansive EIS dataset available, our experimental results show that when using mean absolute percentage error, rectangular coordinates predict capacity better by 2.8% and polar coordinates predict SOH better by 11.3%. Capacity prediction showed negligible change between mathematical formats, with rectangular coordinates having a 0.8% reduction compared to polar coordinates. For SOH prediction, polar coordinates provided more robust predictions than rectangular coordinates by 13.1%. These results showcase the usefulness of the EIS input format for applications that utilize either capacity or SOH as their main performance metric.