电工技术学报  2020, Vol. 35 Issue (18): 3979-3994    DOI: 10.19595/j.cnki.1000-6753.tces.190750
电化学储能 |
基于改进粒子滤波算法的锂离子电池状态跟踪与剩余使用寿命预测方法
焦自权, 范兴明, 张鑫, 罗奕, 刘阳升
桂林电子科技大学电气工程及其自动化系 桂林 541004
State Tracking and Remaining Useful Life Predictive Method of Li-ion Battery Based on Improved Particle Filter Algorithm
Jiao Ziquan, Fan Xingming, Zhang Xin, Luo Yi, Liu Yangsheng
Department of Electrical Engineering & Automation Guilin University of Electronic and Technology Guilin 541004 China
全文: PDF (67157 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 针对标准粒子滤波算法在锂离子电池剩余使用寿命预测方面出现的估计精度不高、算法不稳定及计算效率低等问题,该文提出一种改进粒子滤波算法的状态跟踪与剩余使用寿命预测估计方法。选取电池容量衰退经验物理模型为基础,通过贝叶斯理论对历史样本进行状态跟踪建模,优化训练算法辨识物理模型参数与重采样策略。采用状态跟踪训练优化后最新量测信息取代序贯重要性采样过程中未考虑观测噪声的量测信息,指导产生新的提议分布更新粒子重要性权值计算的方法来改善粒子退化现象,同时基于马尔科夫链-蒙特卡洛(MCMC)方法中的M-H(Metropolis-Hastings)抽样算法丰富采样粒子多样性,改良重采样策略来解决由其引起的粒子枯竭问题,并通过仿真揭示出不同跟踪集S和粒子数M等模型参数对预测结果的影响规律,继而构建实时更新提议分布、MCMC方法与粒子滤波算法优化融合的状态跟踪与剩余使用寿命预测模型——基于MCMC的更新改进粒子滤波融合算法模型。仿真实验结果表明,该文提出的改进算法具有状态跟踪拟合度好、预测精度高及计算效率性能优良等特点,并通过设计出不同类型电池容量和算法模型等多种组合方案的仿真,验证了改进算法较强的稳定鲁棒性、泛化适应性和通用有 效性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
焦自权
范兴明
张鑫
罗奕
刘阳升
关键词 改进粒子滤波算法马尔科夫链-蒙特卡洛方法锂离子电池状态跟踪剩余使用寿命预测    
Abstract:A standard particle filter (PF) algorithm has the problems of low estimation accuracy, unstable algorithm and low computational efficiency in predicting the remaining useful life (RUL) of li-ion batteries. This paper presents a state tracking and RUL prediction estimation method based on an improved particle filter algorithm. Using the empirical physical model of battery capacity degradation, Bayesian theory is used to model the state tracking of historical samples and optimize training algorithm to determine physical model parameters and resampling strategy. The latest measurement information optimized by the state tracking training is used to replace the measurement information without considering the observation noise in the sequential importance sampling, which guides the generation of new proposed distributions and updates particle importance weights to improve particle degradation. Meanwhile, Metropolis-Hastings sampling algorithm based on Markov Chain Monte Carlo (MCMC) method enriches the diversity of sampling particles, and improves Sampling Importance Resampling (SIR) to solve the problem of particle depletion. The simulation reveals the influence law of model parameters on the prediction results, such as different tracking sets S and particle numbers M. Then the optimal fusion model of state tracking and RUL prediction based on real-time update proposal distribution is constructed combined with MCMC method and PF algorithm. It is an updated and improved particle filter fusion model based on MCMC. The simulation results show that the improved algorithm has good state tracking fitting, high prediction accuracy and good computational efficiency. The simulations of different types of battery capacity and algorithm models under various combined schemes are designed. It is proved that the improved algorithm has strong stability robustness, generalization adaptability and general availability.
Key wordsImproved particle filter algorithm    Markov Chain Monte Carlo method    li-ion battery    state tracking    remaining useful life prediction   
收稿日期: 2019-06-21     
PACS: TM912  
基金资助:国家自然科学基金(61741126)和广西研究生教育创新计划(YCBZ2019050)资助项目
通讯作者: 范兴明 男,1978年生,教授,博士生导师,研究方向为智能化电器和高电压新技术。E-mail: fanxm_627@163.com   
作者简介: 焦自权 男,1986年生,博士研究生,研究方向为智能化电器。E-mail: jiaomaomao202@163.com
引用本文:   
焦自权, 范兴明, 张鑫, 罗奕, 刘阳升. 基于改进粒子滤波算法的锂离子电池状态跟踪与剩余使用寿命预测方法[J]. 电工技术学报, 2020, 35(18): 3979-3994. Jiao Ziquan, Fan Xingming, Zhang Xin, Luo Yi, Liu Yangsheng. State Tracking and Remaining Useful Life Predictive Method of Li-ion Battery Based on Improved Particle Filter Algorithm. Transactions of China Electrotechnical Society, 2020, 35(18): 3979-3994.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.190750          https://dgjsxb.ces-transaction.com/CN/Y2020/V35/I18/3979