State-of-Health (SOH)–Based Diagnosis System for Lithium-Ion Batteries Using DNN With Residual Connection and Statistical Feature

Abstract

Lithium-ion batteries (LIBs) degrade through repeated charge and discharge, causing increased internal resistance and reduced maximum capacity. This affects their discharge performance, such as maximum power output and runtime, which in turn affects the safety and reliability of the system using the LIB. Therefore, identifying and predicting the state of the LIB is essential to ensure the safety and reliability of the system. This paper proposes a system for diagnosing the health state of LIBs using time-series discharge data. The system for diagnosing the health state of LIBs is constructed by utilizing a residual-deep neural network (R-DNN). DNN with residual connections can have a deeper and wider structure than conventional neural networks, which enables abundant feature extraction. The time-series discharge data are processed to form the input and output data for the proposed diagnostic system, upon which training is conducted. The output of the trained diagnostic system is then used to determine the health state of the LIB. Furthermore, to validate the proposed method, diagnosis was performed on data not used for model training, and the results were analyzed. Additionally, a comparison group model was trained to perform a comparative analysis with the proposed method.