期刊
ETRANSPORTATION
卷 17, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.etran.2023.100243
关键词
Lithium-ion battery; Capacity degradation trajectory; Bayesian optimization; Deep learning
This study proposes a data-centric method to predict the capacity degradation trajectory of lithium-ion cells using early single-cycle data. A deep learning-based model is used to predict a few knots at specific retention levels, which are then interpolated to reconstruct the trajectory. The proposed method is effective in predicting capacity degradation trajectories and can help establish appropriate warranties or replacement cycles in the battery manufacturing and diagnosis processes.
Early degradation prediction of lithium-ion batteries is important to guarantee safe operations and avoid unexpected failure in manufacturing and diagnosis processes. However, long-term capacity trajectory prediction involves several extrapolation predictions, which can result in failure due to cumulative errors and noises. To reduce the efforts for qualification testing of batteries in the industry, this study proposes a data-centric method to predict the capacity degradation trajectory of lithium-ion cells using early single-cycle data. A few knots at specific retention levels are predicted using a deep learning-based model and interpolated to reconstruct the trajectory. The retention levels of two to four knots are identified using two approaches: uniformly divide the retention up to the end of life and find the optimal locations using Bayesian optimization. The proposed model is validated using the experimental data of 169 cells with five-fold cross-validation. The simulation results show that the mean absolute percentage errors in trajectory prediction are less than 1.60% for all cases of knots. The proposed model can directly estimate the overall capacity degradation trajectory by predicting only the cycle numbers of at least two knots based on the early single-cycle charge and discharge data. Further experiments on the impact of the number of input cycles suggest the use of three-cycle input data to obtain robust and efficient predictions even when there is noise in the input data. Finally, the proposed method is applied to predict various shapes of capacity degradation patterns of additional experimental data of 82 cells. We demonstrate that regardless of the shapes of the trajectories, it is sufficient to collect only the cycle information of a few knots while training the prediction model and a few early cycle information for predicting the capacity degradation, which can help establish appropriate warranties or replacement cycles in the battery manufacturing and diagnosis processes.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据