摘要
容量是衡量电池性能的关键指标,当前容量估计存在特征实用性差、模型准确度与泛化性不足的问题。鉴于此,该文提出一种结合极化衰减特征与通道关注混合神经网络的锂离子电池容量在线估计方法。首先,利用恒压电流和弛豫电压衰减的去极化特性,提取不受充电起点影响的多维实用特征,同时引入相关系数法和主成分分析法对特征进行预处理以用于容量在线估计;其次,通过融合深度置信网络(DBN)、长短期记忆网络(LSTM)和挤压-激励(SE)机制,构建具有自适应通道关注能力的混合神经网络以提高容量估计精度;最后,利用多种工况、多种材料的电池数据,对所提的方法进行了验证。结果表明,容量估计的平均绝对百分比误差、方均根百分比误差分别在1.2%、1.5%以内,验证了该方法的准确性与有效性。
The capacity of a lithium-ion battery is a key indicator of the state of health(SOH)and remaining useful life(RUL)of the battery,as well as the state of charge(SOC)and equalization management within the battery pack.Therefore,accurate estimation of battery capacity is crucial for the battery management system(BMS).However,in practice,the battery capacity often cannot be measured directly,and needs to be estimated indirectly by combining relevant methods and accessible parameters.Therefore,this paper extracts multidimensional features reflecting the polarization strength of the battery based on a data-driven approach,and constructs a hybrid neural network with adaptive channel focusing capability,so as to realize online accurate estimation of lithium-ion battery capacity.First,considering that the microscopic manifestation of battery capacity decay is the increase of battery polarization internal resistance,and the constant voltage charging and voltage relaxation processes are directly related to the battery depolarization reaction,the depolarization characteristics of constant voltage current and relaxation voltage decay are utilized to extract the multidimensional practical features that are not affected by the charging starting point.Data preprocessing of the original multidimensional features is performed by correlation coefficient method and principal component analysis(PCA)to complete the screening and dimensionality reduction and fusion of the features,and finally four indirect features are identified as model inputs.Next,the squeeze-excitation(SE)module is integrated into each hidden layer of the deep belief network(DBN)to construct an enhanced DBN network with channel attention capability,and the higher-order feature information extracted from the DBN is used to perform time-series prediction using a long-and short-term memory(LSTM)network to capture the mapping relationship between the battery capacity and the input features.Finally,six sets of simulation experiments are designed for different data structures,prediction models and input features to The simulation results show that the method proposed in this paper can effectively improve the accuracy of capacity estimation.Under different training and prediction data structures,the mean absolute percentage error(MAPE)and root mean square percentage error(RMSPE)of capacity estimation are controlled within 1.2%and 1.5%,respectively,and the model has the ability to simultaneously estimate the battery capacity under different charging conditions with good generalization.For different prediction models,the fusion of DBN network,SE mechanism and LSTM network effectively improves the accuracy of capacity estimation,especially the SE module contributes more significantly to the stability and accuracy of the model,but the LSTM network is equally indispensable,and the synergistic effect of both of them improves the performance of the model by more than 25%.Meanwhile,compared with other advanced algorithms,the model proposed in this paper demonstrates better stability while ensuring higher estimation accuracy,which verifies the effectiveness of the model improvement strategy in this paper.Compared with the original input features,the capacity estimation accuracy is improved by about 30%by combining the complementary features of constant pressure and relaxation dual depolarization processes and data preprocessing methods,which validates the effectiveness of the work done on feature engineering in this paper.
作者
徐志成
杨达
张闯
陈占群
张献
Xu Zhicheng;Yang Da;Zhang Chuang;Chen Zhanqun;Zhang Xian(State Key Laboratory of Intelligence Power Distribution Equipment and System,Hebei University of Technology,Tianjin,300130,China;Key Laboratory of Electromagnetic Field and Electrical Apparatus Reliability of Hebei Province,Hebei University of Technology,Tianjin,300130,China;Baoding UNT Electric Co.Ltd,Baoding,071051,China)
出处
《电工技术学报》
北大核心
2025年第17期5683-5702,共20页
Transactions of China Electrotechnical Society
基金
国家自然科学基金项目(52307238)
省部共建电工装备可靠性与智能化国家重点实验室优秀青年创新基金项目(EERI_OY2023007)
河北省省级科技计划(225676163GH)
河北省燕赵青年科学家项目(E2024202109)
天津市重点项目(22JCZDJC00620)资助。
关键词
锂离子电池
容量在线估计
极化衰减特征
混合神经网络
通道注意力
Lithium-ion battery
capacity estimation online
polarization decay feature
hybrid neural network
channel attention
