基于多循环特征的储能电池SOH估计模型  

A SOH estimation model for energy storage batteries based on multiple cycle features

作　　者：张宣梁 何霆 朱文龙 王屾 曾建华 徐泉[3] 牛迎春 ZHANG Xuanliang;HE Ting;ZHU Wenlong;WANG Shen;ZENG Jianhua;XU Quan;NIU Yingchun(School of Computer Science and Technology,Huaqiao University,Xiamen 361021,Fujian,China;ZhongHai Energy Storage Technology(Beijing)Co.Ltd.,Beijing 102308,China;State Key Laboratory of Heavy Oil,China University of Petroleum,Beijing 102299,China)

机构地区：[1]华侨大学计算机科学与技术学院,福建厦门361021 [2]中海储能科技(北京)有限公司,北京102308 [3]重质油国家重点实验室,中国石油大学(北京),北京102299

出　　处：《储能科学与技术》2023年第11期3488-3498,共11页Energy Storage Science and Technology

基　　金：国家自然科学基金国际(地区)合作与交流项目(52211530034)。

摘　　要：准确估计电化学储能电池的SOH(state of health)对于确保电池的安全可靠工作至关重要。数据驱动方法在SOH估计领域得到广泛应用,但现有方法忽略了电池循环过程中多个连续循环之间的时序健康信息和特征挖掘,以及这些特征与SOH值之间的关系。为解决这些问题,本研究提出了一种名为MCNet(multi-cycle net)的新型SOH估计模型。该模型不需要手动提取健康特征,只需输入电池充电阶段的电流和电压,即可自动挖掘单次循环中与SOH估计相关的特征,并提取多个连续循环之间的相关特征,进而融合上述特征进行SOH估计。首先,为了构造模型的多循环张量输入数据并提升模型的收敛速度,将每个循环内充电阶段长度不同的采样数据进行长度对齐、最大-最小归一化以及拼接多个连续历史循环数据的预处理;其次,将上述预处理后的张量数据作为模型的输入,建立MCNet模型对公开电池数据集进行预测,平均绝对误差都在1%以内、均方根误差都在1.5%以内;最后,将本文所提出的模型与其他常用的序列预测模型进行比较,并与仅使用单次循环下的电流、电压数据作为输入进行了对比实验,结果表明,本文提出的模型具有较高的SOH估计精度,并且使用多个循环的电流、电压数据作为输入可以提升估计精度。Accurate estimation of the state of health(SOH)of electrochemical energy storage batteries is crucial for ensuring their safe and reliable operation.Data-driven methods have been widely used for SOH estimations.However,existing methods overlook the temporal health information and feature extraction between multiple consecutive cycles of battery operation and the relationship between these features and the SOH value.This study proposes a novel SOH estimation model called multi-cycle net(MCNet)to address these issues.This model does not require manual extraction of health features;it only takes current and voltage measurements during the charging phase of the battery as input.It automatically extracts features relevant to the SOH estimation within each cycle,extracts relevant features between multiple consecutive cycles,and then combines them for SOH estimation.First,to construct the multicycle tensor input data and improve the convergence speed of the model,the sampled data from the charging phase of each cycle with varying lengths were preprocessed through length alignment,maximum-minimum normalization,and concatenation of multiple consecutive historical cycle data.Second,the preprocessed tensor data were used as input to build the MCNet model for predicting the SOH using a publicly available battery dataset.The average-absolute-and rootmean-square errors were within 1%and 1.4%,respectively.Finally,the proposed model was compared with other commonly used sequence prediction models,and a comparative experiment was conducted using only single-cycle current and voltage data as inputs.The results demonstrate that the proposed model achieves higher accuracy in SOH estimation,and using multiple cycles'current and voltage data as input improves the estimation accuracy.

关 键 词：电化学储能电池 健康状态 多循环特征 TRANSFORMER 门控循环单元 

分 类 号：TM911[电气工程—电力电子与电力传动]