A Domain-Adversarial Neural Network for Transferable Lithium-Ion Battery State-of-Health Estimation
Jinhao Meng, Die Hu, Mingqiang Lin, Jichang Peng, Ji Wu, Daniel‐Ioan Stroe
Abstract
As a key indicator of the lithium-ion (Li-ion) battery performance, state-of-health (SOH) estimation still faces challenges in model generalization between datasets. Electrochemical impedance spectroscopy (EIS) provides a nondestructive solution to capture the electrochemical dynamic processes inside the Li-ion battery, while it is highly sensitive to measurement conditions and battery status. To alleviate these issues, this work proposes a transferable data-driven framework for Li-ion battery SOH estimation with the domain-adversarial neural network (DANN) and EIS. The health feature is obtained from battery EIS through the latent representation of an autoencoder (AE) where valuable information can be automatically extracted from the original EIS measurement. The DANN can connect the feature distributions for the samples in both the source and target domains. Two datasets from both cycling and calendar aging tests are used to verify the superior performance of the proposed method in SOH estimation for different cases where the average root mean square error is only 1.28%. This method demonstrates significant potential in practical applications such as electric vehicles (EVs) and battery energy storage systems (BESSs), where accurate and reliable SOH estimation is critical for enhancing safety and prolonging battery lifespan.