可重构卫星锂离子电池剩余寿命预测系统研究  

Study on the reconfigurable remaining useful life estimation system for satellite lithium-ion battery

作　　者：周建宝[1] 王少军[1] 马丽萍[2] 杨思远[2] 彭宇[1] 彭喜元[1] 

机构地区：[1]哈尔滨工业大学自动化测试与控制系,哈尔滨150080 [2]上海空间电源研究所,上海200233

出　　处：《仪器仪表学报》2013年第9期2034-2044,共11页Chinese Journal of Scientific Instrument

基　　金：部委预先研究课题(51317040302)资助项目

摘　　要：针对卫星锂离子电池剩余寿命预测问题,提出一种基于FPGA的可重构卫星锂离子电池剩余寿命预测系统设计方法。首先利用具备不确定性表达能力的相关向量机实现锂离子电池的RUL预测,进而采用FPGA动态重构技术,实现了基于相关向量机的预测算法的嵌入式计算,解决了核函数矩阵和矩阵求逆的计算方法和结构设计等关键问题,为解决硬件计算资源有限条件下的机器学习算法计算问题提供了一种新颖的思路。实验结果表明,在与PC平台保持相近计算精度的条件下,利用FPGA实现的剩余寿命预测计算效率提升了4倍,同时证明了机器学习的可重构计算方法在嵌入式计算体系中的应用具有良好的前景。This paper proposes a FPGA based reconfigurable system design methodology to realize the remaining use- ful life （RUL） estimation for the lithium-ion battery on satellite. With Relevance Vector Machine （ RVM ）, which possesses the uncertainty expression capacity of the prediction results, the RUL estimation of lithium-ion battery is re- alized. Then ,the dynamic Reconfigurable Computing （RC） technique based on FPGA is applied to realize the em- bedded RVM computation. The key challenges, including the computing method and architecture design of the kernel function matrix formulation and matrix inversion, are also solved. The research work contributes a novel idea for the computing of machine learning algorithms under limited hardware resource condition. Experimental results on the bat- tery data set show that with almost the same computing accuracy as that on a PC platform, the proposed method can get a 4x speed up over the PC solution. And this also indicates that the reconfigurable computing method can be well applied to the embedded computing of machine learning algorithms and has bright prospect.

关 键 词：卫星 锂离子电池 剩余寿命预测 可重构计算 故障预测和健康管理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]