A Data-Driven Approach for Lithium-ion Battery Lifetime Classification Based on Early Cycles

In recent years, the global demand for electric energy has been increasing year by year. In order to cope with increasingly serious problems such as grid-connected new energy generation and increasing dispatching pressure of power grid systems, the demand for energy storage systems is growing day by day. However, in the grid-side energy storage system, huge amounts of battery cells are grouped into modules, packs, and systems. This presents a problem: cells with inconsistent lifetimes are assembled into a battery system, which will seriously affect the overall lifetime of the whole battery system. Therefore, early classification for lifetime of cells is important, futher more, fast and accurateclassification about health status of cells is also challenging. In this paper, the XGBoost model and the transformed voltage curves extracted from early cycles are combined to realize the early classification of the end of life(EOL) of batteries in the early aging process. By extracting the features from transformed voltage curves of discharge cycles in the first four weeks (depth of discharge, DOD=100), the EOL of 64 groups of batteries was classified into three categories (large/middle/small, L/M/S). It is worth mentioning that the feature interval that has the greatest influence on the lifetime classification is found to be 3.2V-3.4V through PCA transformation, which indicates the range of feature extraction for future lifetime prediction and classification. Finally, experiments on the ISU-ILCC Battery Aging Dataset show that for small feature requirements, the prediction accuracy of the method reaches more than 85%.