State of Health Estimation for Lithium-Ion Battery Based on Long Short Term Memory Networks

Abstract

In this paper, a state of health (SOH) estimation method using long short term memory (LSTM) networks is applied to predict battery life for electric vehicles (EVs). During the discharging process, the battery shows external features that characterize its attenuation degree and current performance. The discharging time under a constant current, the number of charging and discharging cycles, and the charging capacity are employed to build the prediction model with LSTM networks. The internal modeling parameters are trained by public battery datasets, in which discharging process are introduced for battery SOH prediction. Experimental results indicate that the LSTM networks can accurately predict battery SOH, and estimate battery degradation and internal parameter variations.