摘要
直流备用电源是变电站安全稳定运行的重要保证,厂站中目前常用的铅酸蓄电池存在着寿命低、温度性能差的问题。锂离子电池的长循环寿命、高能量密度等特点,近年随着技术不断成熟,有望成为替代方案。电池健康状态(state of health,SOH)是锂离子电池储能系统可靠运行所需的核心参数,而电化学阻抗谱(electrochemical impedance spectroscopy,EIS)作为一种无损检测的方法,可用来评估电池的SOH并分析其老化的主要机制。针对静态EIS在电池工作情况下获取困难、带直流偏置的快速EIS可解释性不足的问题,本研究提出了一种基于快速阻抗谱可解释性增强的锂离子电池健康状态估计方法,在基本不影响直流电源工作的情况下快速完成电池老化预测与老化机制分析。首先,利用卷积-长短期记忆网络模型实现了动态到静态的EIS预测,卷积网络提取关键特征,长短期记忆神经网络捕捉序列间依赖关系,以实现电池老化机理解析;其次,提出了一种基于极限梯度提升算法及EIS的电池SOH估计方法,捕捉静态EIS与SOH之间的高度非线性映射关系,完成了电池SOH的在线评估,并依靠特征分裂增益量化不同频域特征的贡献以分析EIS的不同形式在预测结果中的重要性。实验表明,所提静态EIS预测方法的平均绝对误差(mean absolute error,MAE)为1.75×10-5;电池SOH估计结果的MAE仅为2.43%,电解液损失是所用电池老化的主要原因。
Substation DC power supply is an important guarantee for the safe and stable operation of the substation,providing a stable power source for control,protection,and other key equipment.When a substation fails and AC power is interrupted,the system can supply power to DC equipment uninterruptedly and stably within a set period.At present,lead-acid batteries are widely used in plants and substations in China,but their short effective life and poor temperature performance limit their reliable operation in diverse environments.Thanks to their high energy density,low self-discharge rate,long cycle life,absence of memory effect,and environmental friendliness,lithium-ion batteries are expected to become an ideal replacement for lead-acid batteries in DC power systems.The health status and aging mechanism of lithium-ion batteries are key attributes for evaluation.State of health(SOH)reflects the battery's current health condition,while the aging mechanism essentially explains the cause of battery performance degradation.Electrochemical impedance spectroscopy(EIS),as a non-destructive testing method,can be used to predict battery SOH and analyze aging mechanisms by obtaining the impedance characteristics of the battery at different frequencies through a small excitation signal.To address the issues that static EIS cannot be acquired during battery operation and fast EIS with DC bias is not conducive to aging mechanism analysis,this paper proposes a lithium-ion battery SOH prediction method based on enhancing the interpretability of fast impedance spectra.We use a convolutional long short term memory network model to predict static EIS from fast EIS with DC bias.The convolutional layers automatically extract key local frequency-domain features,and the LSTM captures long-range dependencies between frequency points,enabling accurate reconstruction of static EIS for subsequent aging mechanism interpretation.The mean absolute error(MAE)of the prediction result is 1.75×10-5.Furthermore,we employ an extreme gradient boosting algorithm to predict battery SOH based on the static EIS,which captures the highly nonlinear mapping between impedance characteristics and SOH,and uses the feature splitting gain to quantify the contribution of different frequency-domain components and EIS representations.The MAE of the prediction result is only 2.43%.In the aging mechanism analysis,we select a stationary equivalent circuit model(ECM)to analyze the EIS.By analyzing the distribution of different components and the variation of parameters,we conclude that electrolyte loss is the main aging factor of the battery.
作者
陈轲娜
刘小江
卜祥航
潘禹昊
李祎婧
孟锦豪
CHEN Kena;LIU Xiaojiang;BU Xianghang;PAN Yuhao;LI Yijing;MENG Jinhao(State Grid Sichuan Electric Power Company Research Institute,Chengdu 610041,Sichuan,China;Power Internet of Things Key Laboratory of Sichuan Province,Chengdu 610041,Sichuan,China;National Innovation Platform for Industry-Education Integration of Energy Storage Technology,Xi'an Jiaotong University,Xi'an 710049,Shaanxi,China;School of Electrical Engineering,Xi'an Jiaotong University,Xi'an 710049,Shaanxi,China)
出处
《储能科学与技术》
北大核心
2025年第12期4709-4720,共12页
Energy Storage Science and Technology
基金
国网四川省电力公司科技项目(521997230036)。
关键词
快速阻抗谱
电池健康状态预测
老化机制分析
卷积-长短期记忆网络
极限梯度提升算法
fast impedance spectra
battery health state prediction
aging mechanism analysis
convolutional-long and short-term memory networks
extreme gradient boosting algorithms
