Transfer learning applying electrochemical degradation indicator combined with long short-term memory network for flexible battery state-of-health estimation

Abstract

The battery mounted in electric vehicle (EV) has various degradation patterns influenced by operating environment (OE), including road conditions and temperature. The diagnostic performance errors in the existing health monitoring model stem from changes in the internal electrochemical characteristics of the battery. Consequently, a state-of-health (SOH) estimation system capable of simulating battery degradation characteristics based on different OEs is deemed necessary. This paper introduces a transfer learning (TL)-based SOH estimation system that can be flexibly updated in response to OE changes in EVs. We also propose a method for deriving electrochemical characteristic indicator (ECI) during operation to simulate the internal chemical characteristics of the battery. An electrochemical parameter is extracted from the battery's discharging current-voltage profile, and its reliability is verified through comparison with parameters obtained from the electrochemical impedance spectroscopy-based Randles circuit model. Furthermore, the SOH estimation performance under various OEs is assessed using both the base-model long short-term memory (LSTM) and TL. Subsequently, the model is validated using degradation data collected in an operating environment different from the one used for training the pre-training model. The TL strategies for each environment are discussed and the SOH prediction performance of the proposed model surpasses that of LSTM without TL, with mean absolute error and root mean square error measuring less than 1 %.