电源技术2026，Vol.50Issue(1)：129-142,14.DOI:10.3969/j.issn.1002-087X.2026.01.016
基于Transformer-XGBoost框架的轨交车辆电池多视角数据健康诊断研究
Health diagnosis of rail transit vehicle batteries using Transformer-XGBoost-based multi-view data analysis
摘要
Abstract
Lithium-ion batteries have been extensively employed in rail transit and energy storage sys-tems due to their high energy density and long cycle life.However,their state of health(SOH)gradu-ally deteriorates with increasing charge-discharge cycles,posing safety risks and maintenance chal-lenges for battery management.Conventional SOH prediction approaches predominantly rely on single-view incremental capacity analysis(ICA)or standard data-driven models,which struggle to fully capture the multiscale variations in electrochemical characteristics and temporal dynamics dur-ing battery degradation.This limitation hinders both prediction accuracy and robustness.To address these challenges,this paper proposes an SOH prediction method based on multi-view data analysis.By integrating information from incremental capacity(IC)curves under both voltage and temporal views,multi-view health indicators(HI)are constructed.A prediction framework combining Trans-former and extreme gradient boosting(XGBoost)is then designed.In this framework,the Trans-former incorporates a dynamic time-window adjustment and a dual-scale attention mechanism to adaptively extract temporal features at different degradation stages.Meanwhile,XGBoost introduces physical constraints to enhance prediction stability and robustness.On the PL13 battery training da-taset from the University of Maryland,the proposed method achieves a root mean square error(RMSE)of only 3.13×10-3 and a coefficient of determination(R2)of 0.997.On the PL11 battery testing dataset,the method maintains a low RMSE of 4.57×10-3 and an R2 of 0.994,demonstrating its supe-rior performance in multi-view feature fusion and dynamic temporal modeling.关键词
健康状态/多视角数据分析/Transformer/XGBoost/电池管理系统Key words
state of health/multi-view data analysis/Transformer/XGBoost/battery management system分类
信息技术与安全科学引用本文复制引用
王健,毛建,唐超伟,孙小康,候晓双,王春生,廖垠钦..基于Transformer-XGBoost框架的轨交车辆电池多视角数据健康诊断研究[J].电源技术,2026,50(1):129-142,14.基金项目
国家自然科学基金资助项目(62103443) （62103443）
湖南省自然科学基金资助项目(2022JJ40630) （2022JJ40630）
中南大学中央高校基本科研业务费专项资金资助项目(2025ZZTS0734) （2025ZZTS0734）
