摘要
针对锂离子电池寿命预测过程中内部参数测量困难、特征构造过程复杂、预测结果不精确等问题,提出了一种结合核主元分析法(KPCA)特征提取与自适应神经网络模糊推理系统(ANFIS)的锂电池剩余寿命预测方法。该方法利用KPCA算法从电池在线可测的参数(电压、电流等)中提取出相互独立的主元特征,通过计算特征得分率和Spearman秩相关系数筛选出准确反映电池退化规律的主元,将其作为锂离子电池的健康指标输入ANFIS神经网络进行容量估计和剩余寿命预测。实验结果表明,基于KPCA-ANFIS的锂电池剩余寿命预测算法所提取的主元特征能够显著反应电池退化特性。此外,通过与PCA-ANFIS算法对比,所设计算法的剩余寿命预测精度得到显著提高(均方误差提高1.73倍,平均绝对误差提高1.38倍)。
As for the RUL prediction of lithium-ion batteries, the complex feature construction processes, capacity measurement difficulties and inaccurate prediction results, they pose some challenges. To solve these issues, a new RUL predicting method is proposed by integrating the kernel principal components analysis(KPCA) based feature extraction strategy and ANFIS-NN predicting method. The method firstly uses KPCA feature extraction algorithm to extract the independent principal component features from the battery′s on-line measurable parameters(voltage, current, etc.), and selects the principal component that accurately reflects the battery degradation law by calculating the feature score rate and the Spearman rank correlation coefficient. Finally, with the input of the health indicator extracted above, an ANFIS neural network is constructed for capacity estimation and remaining life prediction. The result proves that the feature extracted by KPCA-ANFIS algorithm can significantly reflect the degradation characteristics of the battery. By compared with the PCA-ANFIS, the proposed prediction method has more accurate prediction result, with the MSE increasing by 1.73 times and MAE increasing by 1.38 times.
出处
《电子测量与仪器学报》
CSCD
北大核心
2018年第10期26-32,共7页
Journal of Electronic Measurement and Instrumentation
基金
山西省重点研发计划重点项目(201703D111011)
山西省研究生教育改革研究项目(2018JG62)
山西省青年自然科学基金(201601D021075)
山西省高等学校科技创新(2014143)
山西省回国留学人员科研项目(2015-083)
中北大学自然科学基金(2016032,2017025)资助项目
关键词
锂离子电池
剩余使用寿命
健康指标
KPCA算法
ANFIS神经网络
lithium-ion battery
remaining useful life
health indicators
kernel principal components analysis(KPCA)
adaptive network-based fuzzy inference system(ANFIS)
