镉镍蓄电池寿命预测的PF-LSTM建模方法研究  

PF-LSTM modeling method for life prediction of Ni-Cd battery

作　　者：成庶[1] 甘沁洁 赵明 毕福亮 王家捷 王国良 于天剑[1] CHENG Shu;GAN Qinjie;ZHAO Ming;BI Fuliang;WANG Jiajie;WANG Guoliang;YU Tianjian(School of Traffic and Transportation Engineering,Central South University,Changsha 410075,China;CRRC Changchun Railway Vehicles Co.,Ltd,Changchun 130062,China;Asiantongdai Railway Equipment Co.,Ltd,Qingdao 266000,China)

机构地区：[1]中南大学交通运输工程学院,湖南长沙410075 [2]中车长春轨道客车股份有限公司,吉林长春130062 [3]青岛亚通达铁路设备有限公司,山东青岛266000

出　　处：《铁道科学与工程学报》2020年第7期1825-1832,共8页Journal of Railway Science and Engineering

基　　金：国家十三五重点研发计划项目(2017YFB1200902-11)。

摘　　要：对动车组用蓄电池进行寿命预测,能够评估电池状态,降低故障的危害性和运用维护成本,指导修订修程。相较于在线预测模型,离线预测模型无法适应影响因素的不断变化,提出一种基于粒子滤波(PF)与长短期记忆网络(LSTM)融合的在线预测方法。传统的PF方法依赖经验方程作为状态转移方程,而精确的经验方程难以得到,利用已有数据训练LSTM模型,模型得到的退化方程作为PF的状态转移方程,解决了PF依赖经验方程的问题,同时PF能给出不确定性表达。研究结果表明,该方法模型更新简单有效,预测精度好,弥补了镉镍蓄电池寿命模型研究的缺失,对蓄电池剩余寿命研究的发展有着重要意义。The life prediction of a battery for an EMU can evaluate the battery status,avoid the occurrence of failures,reduce the cost of investment,and guide the inspection and repair process.Compared with the online prediction model,the offline prediction model can not adapt to the changing conditions and other factors.This paper proposed an online estimation method based on particle filter(PF)and long short-term memory network(LSTM).The traditional PF method relies on the empirical equation as the state transition equation,but the exact empirical equation is difficult to obtain.This paper used the existing data to train the LSTM model,the degenerate equation obtained by the model was used as the state transition equation of PF.The advantages of the approach can solve the problem of PF dependent empirical equations.The PF can give the uncertainty expression.The results show that the method model is simple and effective,and the prediction accuracy is good,which makes up for the lack of research on the remaining useful life model of cadmium-nickel battery.It has important significance for the development of battery residual life research.

关 键 词：蓄电池 剩余寿命 在线预测 长短期记忆网络 粒子滤波 

分 类 号：TM912[电气工程—电力电子与电力传动]