State of Charge Estimation Method for Lithium-Ion Batteries Using Machine Learning Considering Thermal Variation

Abstract

Lithium-ion Batteries are now widely used and very useful. However, it is difficult to accurately estimate the state of charge (SoC) of a Lithium-ion Battery in use. Possible causes include the effects of temperature, charge/discharge cycles, and degradation. The conventional method uses the Electrochemical Impedance Spectroscopy to estimate the SoC by fitting the equivalent circuit model of the lithium-ion battery used. This method has the disadvantage of making the equivalent circuit model more complex and computationally intensive in order to accurately represent the battery. Therefore, we propose a new SoC estimation method based on machine learning. In this study, Lithium-ion Batteries were placed in a thermostatic chamber between 0°C and 20°C, and the impedance at each temperature was measured using the Electrochemical Impedance Spectroscopy while charging and discharging repeatedly. The input layer was set to the measured impedance and the output layer to the SoC, and the estimation was performed using a neural network. Grid search was used to optimize the hyperparameters and to reduce the errors. As a result, the error was 4.174 % when the temperature was 20 ℃. Considering the temperature characteristics of the Lithium-ion Battery, the error was 4.220 % when the temperature range was changed from 0 ℃ to 20 ℃. The proposed method, a neural network, can be used to estimate the SoC even under the temperature change.