摘要
锂离子电池(lithium-ion batteries, LIBs)的健康状态(state of health, SOH)评估对电动汽车(electric vehicles, EVs)和储能系统的安全性与经济性至关重要。针对传统SOH估计方法依赖全频段电化学阻抗谱(electrochemical impedance spectroscopy, EIS)数据、计算复杂度高且跨工况鲁棒性不足的问题,本研究提出了一种基于EIS几何解析与分段特征提取的锂离子电池SOH评估方法。通过弛豫时间分布(distribution of relaxation times, DRT)分析识别电池极化过程,构建分段等效电路模型(equivalent circuit model, ECM),在高、中、低频段分别提取欧姆内阻(R_(o))、电荷转移阻抗(charge transfer resistance, R_(ct))和扩散斜率(β)等9维特征参数,显著降低了数据存储与计算需求。为消除温度与荷电状态(state of charge, SOC)对特征参数的干扰,设计多层感知机(multilayer perceptron, MLP)模型将多工况特征映射至标准工况(25℃, 60%SOC),并结合随机森林(random forest, RF)算法建立SOH预测模型。实验结果表明,该方法在多种工况下的SOH评估平均绝对误差(mean absolute error, MAE)和平均绝对百分比误差(mean absolute percentage error, MAPE)分别低至0.00662和0.251735%,优于梯度提升决策树(extreme gradient boosting, XGBoost)等对比算法。特征重要性分析显示,R_(ct)、R_(o)和β对SOH预测的贡献显著,且单特征参数的独立评估误差仍低于0.42%,验证了该方法的工程适用性。本研究为嵌入式电池管理系统(battery management system, BMS)提供了高精度、低复杂度的SOH评估方案,尤其适用于有限硬件资源的实车应用场景。
Accurate assessment of the state of health(SOH)of lithium-ion batteries is essential for ensuring the safety,reliability,and cost-effectiveness of electric vehicles and energy storage systems.To overcome the limitations of conventional SOH estimation methods-such as high computational complexity and poor robustness across operating conditions due to their dependence on full-frequency electrochemical impedance spectroscopy(EIS)data-this study proposes a novel SOH evaluation framework based on geometric analysis and segmented feature extraction of EIS.The distribution of relaxation times is employed to decouple overlapping polarization processes,enabling the construction of a piecewise equivalent circuit model that extracts nine-dimensional features from high,medium,and low-frequency EIS segments.This segmentation significantly reduces data storage and computational demands.A multilayer perceptron model is then used to normalize these features to standard conditions(25℃;60%state of charge,SOC),accounting for variations in temperature and SOC,followed by a random forest model for SOH prediction.Experimental results demonstrate that the proposed method achieves a mean absolute error of 0.00662 and a mean absolute percentage error of 0.251735%under diverse operating conditions,outperforming benchmark algorithms such as eXtreme Gradient Boosting.Feature importance analysis identifies R_(ct),R_(o),andβas the most influential parameters in SOH estimation.Moreover,individual features yield prediction errors below 0.42%,confirming the method's practicality for deployment in resource-constrained battery management systems.Overall,the proposed framework offers a high-accuracy,low-complexity solution for embedded SOH monitoring in real-world applications.
作者
严芷涵
王学远
魏学哲
戴海峰
YAN Zhihan;WANG Xueyuan;WEI Xuezhe;DAI Haifeng(School of Automobiles,Tongji University,Shanghai 201804,China)
出处
《储能科学与技术》
北大核心
2025年第12期4732-4742,共11页
Energy Storage Science and Technology
基金
国家自然科学基金(52207242)。
关键词
锂离子电池
健康状态
电化学阻抗谱
几何解析
随机森林
lithium-ion batteries
state of health
electrochemical impedance spectroscopy
geometric analysis
random forest
