Comparison of Machine Learning Methods in SOC Estimation of Lithium-ION Batteries

Abstract

In recent years, demand for fuel cells and storage batteries is increasing. In particular, lithium ion batteries are very popular as batteries of small electronic devices such as mobile phones and laptop computers because they have very high energy density and can easily output high voltage. As effects of temperature, charging / discharging cycle and degradation, It is difficult to accurately determine the SOC (State of Charge) of the battery under use. Equivalent circuit model is usually used as the SOC estimation method. Although it has high robustness, it has the drawback that design of equivalent circuits and calculations are complicated and difficult. In this study, the impedance of the lithium ion battery was measured by AC impedance method. The measured impedance locus was used as the input layer, and the output layer was SOC. Machine learning methods such as SVM(Support Vector Machine) and SVR(Support Vactor Regerssion) were used to estimate the SOC. It is well proved that the estimation of SOC is possible by the machine learning method. According to the comparation of the result of each method, an optimum method was obtained.