Efficient screening of retired lithium-ion batteries with a lightweight stacking model via time-frequency data fusion
点击次数：
影响因子：10.7
DOI码：10.1016/j.est.2026.123818
教研室：H. Xiao, G. Luo, W. Ma, J. Wu, etc.
发表刊物：Journal of Energy Storage
关键字：Data fusion; Retired battery screening; Simplified transformer; Stacking architecture
摘要：With the rapid expansion of electric vehicles and renewable energy systems, the global volume of retired batteries has reached an unprecedented scale and continues to grow at an accelerating pace. However, low detection efficiency, limited datasets, and absence of historical data present significant challenges for large-scale echelon utilization of retired batteries. It is imperative to develop a rapid and accurate screening method for retired batteries to overcome these limitations. Here, we introduce integrating charge–discharge tests with electrochemical impedance spectroscopy (EIS) for effective battery screening. Specifically, charging profiles and EIS data were first collected from 870 commercially retired batteries at a recycling enterprise. To address the high redundancy present in the raw data, a lightweight encoder-only Transformer was then designed to identify informative data segments and extract effective features. Time–frequency domain fusion was achieved through a hierarchical stacking framework that models the complementary and class-dependent prediction behaviors of heterogeneous modalities. Finally, a stacking ensemble architecture was constructed, wherein two simplified Transformer branches serve as base learners with complementary category-wise strengths, and a random forest meta-classifier learns the nonlinear interactions between their confidence distributions to suppress inter-modal interference. Experimental results show that the proposed fusion method completes battery screening within 4 min with an accuracy of 97.01%, demonstrating its ability to support multidimensional information analysis under real-world time-constrained conditions.
论文类型：期刊论文
学科门类：工学
文献类型：J
卷号：178
页面范围：123818
是否译文：否
发表时间：2026-07-22
发布期刊链接：https://www.sciencedirect.com/science/article/pii/S2352152X26034821