Alternating Current Impedance Probing Capacity of Lithium‐Ion Battery by Gaussian Process Regression

Abstract

Alternating current (AC) impedance is an important and promising feature for lithium‐ion battery state estimation and prediction. Herein, a new battery capacity estimation method using AC impedance with Gaussian process regression (GPR) is proposed. A bunch of high‐energy 18 650‐type batteries with a nominal capacity of 3.5 Ah are cycled at 25, 35, and 45 °C until the capacity drops below 2.6 Ah. Two single‐frequency points are found which are highly correlated with the battery residual capacity regardless of the cycling temperatures. Machine learning methods are used to probe the battery capacity with the real and imaginary impedance of two single‐frequency points. The best model achieves a test root‐mean‐squared error of 0.5% (17.36 mAh) with GPR. This work provides a new perspective to predicting the complex dynamical behavior of batteries by combining electrochemical impedance with data‐driven methods.

Application of Electrochemical Impedance Spectroscopy to Degradation and Aging Research of Lithium-Ion Batteries

“…Locorotondo et al found that there may be a certain frequency point that can be considered as a reliable indicator of SOH through data clustering . Zhu et al first investigated the relations between battery capacity and impedance, which found that the impedance at two frequencies (0.25 and 398 Hz) exhibited the highest correlation coefficient with the capacity fade . Therefore, they used the real and imaginary part of the two single-frequency impedance, i.e., four parameters, as features combined with different machine learning algorithms, including linear regression (LR), support vector machine (SVM), and Gaussian process regression (GPR).…”

Advances of LiCoO2 in Cathode of Aqueous Lithium‐Ion Batteries

“…The introduction of machine learning methods can effectively overcome these limitations and provide a faster, efficient and accurate optimization strategy. [ 208,209 ]…”

“…The introduction of machine learning methods can effectively overcome these limitations and provide a faster, efficient and accurate optimization strategy. [208,209] First of all, machine learning methods can quickly screen out LiCoO 2 cathode materials with potentially excellent performance by establishing a correlation model between material structure and performance. With a large amount of experimental data and computational results, highly accurate predictive models can be trained to provide targeted guidance in material design.…”