基于LSTM与KPCA的退役动力电池分选方法  

orting method for retired power batteries based on LSTM and KPCA

作　　者：黄志礼 蔺红[1] Huang Zhili;Lin Hong(School of Electrical Engineering,Xinjiang University,Urumqi 830017,China)

机构地区：[1]新疆大学电气工程学院,乌鲁木齐830017

出　　处：《国外电子测量技术》2023年第12期152-159,共8页Foreign Electronic Measurement Technology

基　　金：新疆维吾尔自治区重点研发专项(2022B01020-3)资助。

摘　　要：针对退役动力电池的快速分选问题,提出了一种能够满足电网储能要求的基于长短期记忆网络(long short-term memory,LSTM)、核主成分分析(kernel principal component analysis,KPCA)和优化K-means聚类的退役动力电池快速分选方法。采用二阶RC模型和混合脉冲功率测试(hybrid pulse power characterization,HPPC)数据,利用带遗忘因子的最小二乘法(forgetful factor recursive least squares,FFRLS)获取电池参数;然后结合LSTM网络,估计电池健康状态(state of health,SOH)和最大可用容量;根据电池参数和梯次利用范围进行初分选,剔除不适用于电网储能的电池;最后依据KCPA处理的特征变量,采用改进的K-means聚类算法进行再分选。仿真结果表明,该方法准确度超过97%,具有较高应用价值。A rapid sorting method for retired power batteries is proposed to meet the energy storage requirements of the power grid,which utilizes long short-term memory(LSTM)networks,kernel principal component analysis(KPCA),and optimized K-means clustering.The method employs a second-order RC model and hybrid pulse power characterization(HPPC)data to estimate battery parameters using the forgetful factor recursive least squares(FFRLS)algorithm.The LSTM network is then utilized to estimate the battery's state of health(SOH)and maximum available capacity.Based on the battery parameters and utilization range,an initial sorting is performed to exclude batteries unsuitable for grid energy storage.Finally,with the feature variables processed by KPCA,an optimized K-means clustering algorithm is applied for further sorting.Simulation results demonstrate an accuracy exceeding 97%,highlighting its high practical value.

关 键 词：退役动力电池 脉冲功率测试 长短期记忆网络 核主成分分析 分选 

分 类 号：TN2[电子电信—物理电子学]