无监督聚类在锂离子电池分类中的应用  

A new method based on unsupervised clustering for lithium-ion battery classification

作　　者：申建斌[1] 唐有根[1] 李玉杰[1] 谢正和[1] 

机构地区：[1]中南大学化学化工学院,湖南长沙410083

出　　处：《计算机与应用化学》2007年第3期305-308,共4页Computers and Applied Chemistry

基　　金：国家重点基础研究发展规划863计划(2001AA501433)

摘　　要：单体电池的一致性,决定了电池组的性能,如何选出性能一致的单体电池又一直是电池组研究中的重点所在。本文采集了100个合格锂离子电池的6项性能指标(老化前后电压、容量、内阻、1C放电平台、电芯厚度),运用主成分分析(PCA)、核主成分分析(KPCA)、随机森林(RF)3种无监督聚类方法,对数据结构进行了研究。结果表明,数据指标之间存在复杂的非线性关系,主成分分析和核主成分分析,均未能形成明显聚类,但随机森林数据在低维空间显然形成4类,任意从中选4个电池组成电池组作循环性能仿真测试,结果显示由由该方法挑选出的单体电池具有较好的一致性。The performance of battery pack largely depends on the uniformity of single batteries and the selection of batteries with uniform properties has become the focus in the research of battery pack. Three unsupervised clustering methods （ PCA, KPCA, RF） are employed to search the structures of a set of lithium-ion battery data including several parameters such as aging voltage, capacity, resistance, 1 C discharging time and thickness. Results show there are complex non-linearity between parameters so that PCA and KPCA failed to cluster. However four class clearly appear when data are projected on the low dimensional space using RF. Four batteries are randomly selected from four classes to assemble four battery packs. The cycling performance of those battery packs gives a satisfactory result.

关 键 词：无监督聚类 随机森林 锂离子电池 电池组 一致性 

分 类 号：TM911.14[电气工程—电力电子与电力传动]