基于增量学习相关向量机的锂离子电池SOC预测方法  

A Prediction Method of Li-ion Batteries SOC Based on Incremental Learning Relevance Vector Machine

作　　者：范兴明[1] 王超[1] 张鑫[1] 高琳琳 刘华东[1] Fan Xingming;Wang Chao;Zhang Xin;Gao Linlin;Liu Huadong(YCBZ2019050)和桂林电子科技大学研究生优秀论文培育项目(16YJPYSS02)资助.)

机构地区：[1]桂林电子科技大学电气工程及其自动化系

出　　处：《电工技术学报》2019年第13期2700-2708,共9页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金(61741126);广西制造系统与先进制造技术重点实验室主任课题(16-380-12-006Z);广西研究生教育创新计划项目(YCBZ2019050);桂林电子科技大学研究生优秀论文培育项目(16YJPYSS02)资助

摘　　要：针对锂离子电池荷电状态(SOC)预测精度不高以及在线适应性差的问题,提出一种改进的增量学习相关向量机模型对锂离子电池SOC 进行在线预测。选择锂离子电池电压、充放电电流和表面温度作为模型的输入,SOC 作为模型的输出,构造模型的训练集。选用快速序列稀疏贝叶斯学习算法进行训练,并结合增量学习法建立增量学习相关向量机模型进行锂离子电池SOC在线预测方法研究。研究发现,通过自动调整核参数的方法,可以保证有较高的预测精度。算法验证实验表明,核参数可以控制算法的预测精度和计算效率,该算法预测精度高、计算速度快、通用性强,可为锂离子电池SOC 的预测与应用提供参考。Aiming at the problems of low prediction accuracy and poor online adaptability for state of charge (SOC) of li-ion battery, an improved incremental relevance vector machine (RVM) model is proposed to predict the SOC of li-ion battery online. The measured voltage, discharge current and surface temperature are selected as the model input, and the SOC is used as the model output, both of them are constructed the training set of the model. A fast-sequence sparse Bayesian learning algorithm is chosen to train RVM, and the incremental learning relevance vector machine model was established by connecting RVM algorithm with incremental learning method to research the prediction for SOC of liion battery online. The study found that this model can guarantee a higher prediction accuracy by adjusting the kernel parameters automatically. The result of the experiment indicates that the kernel parameters can control the prediction accuracy and calculation efficiency of the algorithm, and the IRVM algorithm has the characteristics of high prediction accuracy, fast calculation speed and strong universality, in terms of prodiction and application of the SOC li-ion battery, the algorithm can provide a reference for it.

关 键 词：相关向量机 增量学习法 核参数 计算效率 锂离子电池 SOC预测 

分 类 号：TM912[电气工程—电力电子与电力传动]