Estimation of Lithium-Ion Battery Health State by Evae and Bigru Based on Eis

Abstract

Lithium-ion batteries (LIBs) are commonly utilized in energy storage and electric vehicles due to its excellent performance. However, after the battery goes through repeated charge and discharge cycles, its performance will inevitably decay, which may trigger safety risks and lead to serious safety accidents. Therefore, it is critical to precisely assess the health of LIBs. This paper presents an estimation model combining Enhanced Variational Autoencoder (EVAE) and the bidirectional gated recurrent unit (BiGRU). First, Variational autoencoders (VAE) learns the potential spatial distribution of Electrochemical Impedance Spectroscopy (EIS) data, obtains the distribution parameters of potential space, and forms the preliminary characteristics. Then, variational mode decomposition (VMD) technology is used to decompose the characteristic signals extracted by VAE, and the data highly related to the capacity trend is separated. Finally, the features after VMD processing are input into BiGRU model for mapping processing to achieve accurate estimation of battery capacity. Through a series of experiments under different temperature conditions, the EVAE-BiGRU model shows excellent performance, with MAE and RMSE below 0.8413mAh and 1.0711 mAh, respectively. The study's conclusions aid in the creation of data-driven SOH estimation methods to provide a reliable framework for improving battery performance and life.

Keywords: LIBS, EIS, Deep Learning, Noise processing, Health state estimation