Data-driven Lithium-ion Battery Degradation Estimation using Electrochemical Impedance Spectroscopy and Throughput

Accurately estimating the state of health (SOH) of Lithium-ion batteries not only can reflect the degradation level of batteries but also improve the reliability and efficiency of electric vehicles (EVs). Due to the dynamic operation scenarios of the EVs, accurately estimating the degradation of Li-ion batteries remains a challenge in the battery management system. This work incorporates Electrochemical Impedance Spectroscopy (EIS), battery cycle number, and battery throughput, which provide detailed insights into battery degradation mechanisms, with machine learning models for estimating the battery state of health (SOH). The effectiveness of each combination of EIS measurement data, battery cycle number, and battery throughout has been examined in order to determine the efficacy of each feature in predicting battery degradation. The results on one of the largest EIS measurement datasets indicate that using the proposed battery cycle number and throughput as additional features in a Gaussian process regression model (GPR) to estimate the degradation of lithium-ion batteries improves prediction accuracy up to 0.0105 for mean squared error (MSE) and 0.0655 for mean absolute error (MAE), compared with relying solely on EIS measurement data. Moreover, in order to demonstrate the effectiveness of the GPR model, the deep neural network model (DNN) and support vector machine (SVM) have been used to compare the performance with the GPR model for the estimation of SOH. Results show that the GPR model outperformed the DNN and SVM models when the EIS measurement data were combined with cycle number and battery throughput to describe the battery degradation.