基于电化学阻抗谱及弛豫时间分布的锂电池异常识别与诊断

Identification and Diagnosis of Abnormal Lithium-ion Batteries Based on Electrochemical Impedance Spectroscopy and Distribution of Relaxation Time Analysis

摘要:

针对锂离子电池模组中单体电池的状态识别与诊断问题，基于电化学阻抗谱和弛豫时间分布曲线，引入仿射传播（AP）聚类算法进行电池模组异常识别，并与基于密度噪声鲁棒空间聚类（DBSCAN）算法进行对比，以10个正常样本、多个异常样本进行识别。结果表明，AP聚类算法在精度、鲁棒性、参数敏感性方面（数据重叠、密度不均等）表现得比DBSCAN算法更好。另外，引入极端梯度提升（XGBoost）回归器，在存储该电池对应的一定数据后，对同样电池进行识别时，直接通过XGBoost回归器进行电池异常诊断。结果表明，异常检出率为100%，异常种类识别准确率超过92%。最后，提出了包括数据收集、特征提取、识别诊断等关键环节的电池模组异常识别和诊断系统。

Abstract:

To address the issues of state identification and diagnosis for cells in lithium-ion battery modules， this paper proposes using electrochemical impedance spectroscopy and distribution of relaxation time curves with the affinity propagation （AP） clustering algorithm for abnormal identification of battery modules. The AP algorithm is compared with the density-based spatial clustering of applications with noise （DBSCAN） algorithm using 10 normal samples and multiple abnormal samples. The results show that AP performs better than DBSCAN in terms of accuracy， robustness， and parameter sensitivity （overlapping data， uneven density， etc.）. In addition， the extreme gradient boosting （XGBoost） classifier is introduced， and after storing a certain amount of data corresponding to the battery， the same battery can be directly diagnosed for abnormalities through the XGBoost classifier. The anomaly detection rate is 100%， and the accuracy of identifying anomaly types exceeds 92%. Finally， a battery module abnormal identification and diagnosis system is proposed， which includes key steps such as data collection， feature extraction， identification， and diagnosis.