SOH Estimation of Batteries using Lithium-Ion Internal Parameters with Convolution Neural Network and Gated Recurrent

Abstract

This paper proposes an estimation method for Lithium-Ion Batteries SOH by learning the batteries’ internal parameters using the Convolution Neural Network and the Gated Recurrent Unit. Various equivalent circuit models exist to represent the batteries’ internal parameters. Among these equivalent circuit models, the most representative model is the Randles model, and the data measured based on the Randles model is used as learning input data. The internal parameters of batteries change non-linearly depending on the operation condition and use time. So, nonlinear features are extracted using the CNN input as the batteries' parameters. The extracted features are used as an input of the GRU to learn the characteristics of change over time, and SOH is predicted through this. The learning dataset utilizes 17IND10 LibForSecUse of EMPIR, which validates the performance of the proposed model.