A Screening Method for Retired Lithium-Ion Batteries Based on Support Vector Machine With a Multi-Class Kernel Function

Abstract

Abstract With the retirement of a large number of lithium-ion batteries from electric vehicles, their reuse has received increasing attention. However, a retired battery pack is not suitable for direct reuse due to the poor consistency of in-pack batteries. This paper proposes a method of retired lithium-ion battery screening based on support vector machine (SVM) with a multi-class kernel function. First, ten new NCR18650B batteries were used to carry out the aging experiments for collecting the main parameters, such as capacity, voltage, and direct current resistance. Second, an SVM based on a multi-class kernel function was proposed to screen retired batteries. To improve the screening efficiency, a capacity/voltage second-order conductance curve was adopted to extract their capacity features quickly, and four new feature points were selected as the input of the SVM to classify retired batteries. Finally, the retired batteries are accurately divided into four classes by the trained model, and the classification accuracy can reach 97.0%. Compared with the traditional method, the feature extraction time can be reduced by four-fifths, and the screening efficiency is greatly improved.