摘要
由于动力系统特性的不同,针对传统内燃机车辆制定的保养维护策略,对于电动汽车来说并不适用。根据电动汽车的历史运行数据,预测其未来循环里程的变化趋势,可提出个性化的保养维修建议,延长电池系统的使用寿命。首先,基于实车运行的慢充数据,绘制不同充电片段下各电池单体的IC曲线,利用皮尔森相关性分析,提取与里程具有高相关性的特征,形成电池组的IC特征带。此外,研究表明等效循环次数、充电时间和平均温度与循环里程具有非常高的相关性。然后,利用上述特征和IC特征带的均值和宽度,可构造5维特征,以该多维特征为输入量,以循环里程为输出量,利用多种机器学习算法建立不同的数据驱动模型。结果表明,各模型均有较好的预测精度,循环里程预测的平均误差均小于3%,其中支持向量回归模型的预测效果最好,平均误差小于1%。通过对未来循环里程的预测,可有效识别出电池老化衰减速率,为电动汽车的保养维修提供指导和建议。
Due to the different characteristics of power systems,the maintenance strategy for traditional internal combustion engine vehicles is not suitable for electric vehicles.According to the historical operation data of electric vehicles,the change trend of its future cycle mileage can be predicted,and personalized maintenance suggestions can be put forward to prolong the service life of battery system.First,based on the slow charging data of the real electric vehicles,the IC curve of each cell under different charging segments is drawn.Pearson correlation analysis is used to extract features that have high correlations with mileage,then the IC feature zone of the battery pack can be constructed.In addition,the number of equivalent cycles,charging time,and average temperature are proven with a high correlation to cycle mileage.Using the above features and the average value and width of the IC feature zone,a 5-dimensional feature set can be constructed.Then,by using the multi-dimensional features as the input and the cycle mileage as the output,different data-driven models based on a variety of machine learning algorithms can be established.The results show that each model has good prediction accuracy,and the average error of cycle mileage prediction is less than 3%.Among them,the support vector regression model has the best prediction accuracy,and its average error is less than 1%.By predicting the future cycle mileage,the rate of battery degradation can be effectively identified,which provides guidance and suggestions for the maintenance of electric vehicles.
作者
邓忠伟
肖伟
李阳
黄勇
贾俊
胡晓松
DENG Zhongwei;XIAO Wei;LI Yang;HUANG Yong;JIA Jun;HU Xiaosong(Department of Automotive Engineering,Chongqing University,Chongqing 400044;State Key Laboratory of Control and Simulation of Power Systems and Generation Equipment,Tsinghua University,Beijing 100084;Sichuan Hongwei Technology Co.,Ltd.,Chengdu 610041;Tsinghua Sichuan Energy Internet Research Institute,Chengdu 610213)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2021年第24期250-258,共9页
Journal of Mechanical Engineering
基金
国家自然科学基金(51875054)
重庆市自然科学基金(cstc2019jcyjjq0010)
博士后(cstc2020jcyj-bsh0040)
四川省科技计划(20ZDYF0274)资助项目