电工技术学报2025，Vol.40Issue(11)：3349-3361,13.DOI:10.19595/j.cnki.1000-6753.tces.240835
磷酸铁锂电池荷电状态估计的多源数据特征提取
Multi-Source Data Feature Extraction Method for State of Charge Estimation of LiFePO4 Battery
摘要
Abstract
The flat open-circuit voltage versus state of charge(SOC)curve of LiFePO4 batteries leads to difficulties in achieving an accurate estimation of SOC using only electrical signals.In addition,there are limitations in SOC estimation methods for single electrical,thermal,and acoustic data sources.In view of this,a multi-source data feature extraction method for SOC estimation of LiFePO4 battery was proposed.A comprehensive feature extraction was carried out on the electro-thermal-acoustic multi-source data obtained from different angles.Considering the advantages of different feature selection methods,a new feature selection method integrating Spearman correlation coefficient,mutual information,category boosting and least absolute shrinkage and selection operator regression was proposed.The joint selection of electro-thermal-acoustic key features was realized to improve the accuracy of SOC estimation. Firstly,an experimental platform for LiFePO4 batteries was built.Electro-thermal-acoustic multi-source data were acquired.The transient features and short-term variation features of electrical and thermal signals,as well as the time-domain,frequency-domain,and time-frequency-domain features of ultrasonic signals were extracted,respectively.Secondly,in order to select the key features more accurately,a new method of feature selection incorporating Spearman correlation coefficient,mutual information,category boosting,and least absolute shrinkage and selection operator regression was proposed.In order to verify the performance of the proposed method,the proposed method was compared with SOC estimation results using all features and SOC estimation results under different feature selection methods.The effect of SOC estimation using single data source features versus multi-source data features was compared.The feasibility of the proposed method was verified at different magnifications and under different operating conditions.Finally,Gaussian white noise with different signal-to-noise ratios was added to the raw ultrasound signals acquired under dynamic stress test(DST)conditions and new european driving cycle(NEPC)conditions,respectively,to verify the applicability of the proposed method under high-intensity noise. The results show that using the new method of feature selection proposed can effectively select the features that are important for SOC estimation with higher accuracy than SOC estimation using all features.With the same number of features,the SOC estimation accuracy of this method is improved compared with that of a single feature selection method.The model constructed using electric-thermal-acoustic multi-source key features has higher SOC estimation accuracy compared to single data source features.When using the BiGRU model,the mean absolute error and root-mean-square error of SOC estimates are 0.58%and 0.72%,respectively.The method performs well under a single operating condition.The method also shows good applicability at different discharge multipliers and under multiple operating conditions.Under DST conditions and NEDC conditions,the mean absolute error of SOC estimation is 0.91%and 0.98%,and the root mean square error is 1.03%and 1.13%,respectively,which verifies the validity and accuracy of the method.After adding noise with different signal-to-noise ratios to the original signals of different working conditions,the wavelet noise reduction can resist the noise interference in the actual environment to a certain extent and maintain the accuracy of SOC estimation.关键词
磷酸铁锂电池/荷电状态/多源数据/特征提取/特征选择Key words
LiFePO4 battery/state of charge/multi-source data/feature extraction/feature selection分类
信息技术与安全科学引用本文复制引用
刘素贞,任佳乐,袁路航,徐志成,张闯..磷酸铁锂电池荷电状态估计的多源数据特征提取[J].电工技术学报,2025,40(11):3349-3361,13.基金项目
河北省自然科学基金项目(E2024202010)、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(河北工业大学)优秀青年创新基金项目(EERI_OY2023007)资助. （E2024202010）
