State of Health Estimation of Lithium-Ion Batteries from Charging Data: A Machine Learning Method

Abstract

Accurate state of health (SOH) estimation of the lithium-ion battery plays an important role in ensuring the reliability and safety of the battery management system (BMS). The data-driven method based on the selection of degradation features can be effectively applied to SOH estimation. In practice, lithium batteries often work in complex discharge conditions, but they are charged under constant current (CC) conditions. Therefore, the suitable degradation features of the battery are extracted in this work for accurate SOH estimation. First, the degradation features are summarized and extracted from the CC charging data. Second, the Pearson correlation coefficient is utilized to quantify the relationship between the extracted degradation features and the battery SOH, thus determining the most influential degradation feature. Finally, the long short term memory (LSTM) is used for model training and SOH estimation based on the selected feature. The results show that LSTM model can give reliable and accurate SOH estimation with R2 of 1 and lower mean absolute error (MAE) and maximum error (MAX).