Estimating Battery Life From First-Cycle Charging Data Using HiLoSegNet Model

In electric vehicles, some short-life lithium-ion batteries with a comparatively higher aging rate could adversely affect the service life of the large-scale battery pack. This study proposes the use of first-cycle charging data to estimate the battery life, providing an effective indicator for battery grouping. Herein, twelve features were extracted based on the battery’s first-cycle charging data. Random forest (RF) was used to quantify the Gini impurity variation in each feature regarding the battery life, and key features were accordingly selected. By aiming at the problem of less data in the first cycle, random noise was introduced to form spurious data based on the training dataset to increase the number and diversity of training samples. Owing to relatively small differences in the charging data of different batteries in the first cycle, the high-dimensional mapping technology, regular segmentation, and local decision-making were combined to establish a model called “High-Dimensional Local Segmentation Network” (HiLoSegNet), which was applied to small samples. By employing high-dimensional mapping, regular segmentation mechanism, and local regression analysis, the model enhances prediction accuracy and generalization capability in the presence of sparse features. The HiLoSegNet estimation model was pseudo trained based on pseudo data and fine-tuned based on actual features. Two sets of NCA and NCM data from Tongji University were used as research objects to verify the proposed method. Among them, the maximum error, minimum error, root-mean-square error (RMSE), mean absolute error (MAE), and R-Square (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula>) between the estimated battery life and actual value corresponding to the NCM dataset were found to be 7.6667, 0.2334, 5.2541, 4.8333, and 0.90, respectively. The method proposed in this study, which takes 1.2 s without additional battery testing, can provide battery life indicators for battery grouping.