State Estimation Method for Lithium-ion Batteries Using Electrochemical Impedance Spectroscopy and Neural Network

Abstract

To estimate of lithium-ion battery state, a new method using Electrochemical Impedance Spectroscopy (EIS) and Machine Learning (ML) is proposed. In this method, massive impedance data obtained from the EIS is used as training data for ML model, which enables state estimation with a high accuracy. However, to create a highly accurate and versatile model, it is necessary to obtain a large amount of diverse data, which requires a lot of time. In this study, we investigated two approaches to obtain many data in a short time: the introduction of a weighted division method that generates many training data from a few numbers of data, and the development of an inexpensive impedance measurement system that can be operated in parallel. For weighted division, we estimated the temperature of lithium-ion batteries using neural network and found that the RMSE decreased from 6.88 K to 1.58 K. The measurement system was found to be able to measure impedance with accuracy close to that of a general frequency response analyzer.