考虑驾驶行为的电动汽车电池电压预测方法  

Battery Voltage Prediction Method for Electric Vehicle Considering Driving Behavior

作　　者：朱曼[1,2] 谢宗锐 张晖[1,2] 陈枫 马枫[1,2] 李少鹏 ZHU Man;XIE Zongrui;ZHANG Hui;CHEN Feng;MA Feng;LI Shaopeng(Intelligent Transportation Systems Research Center,Wuhan University of Technology,Wuhan 430063,China;National Engineering Research Center for Water Transport Safety,Wuhan University of Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学智能交通系统研究中心,武汉430063 [2]武汉理工大学国家水运安全工程技术研究中心,武汉430063

出　　处：《交通信息与安全》2022年第6期137-147,共11页Journal of Transport Information and Safety

基　　金：国家重点研发计划项目(2019YFB1600800);武汉理工大学三亚科创园开放基金(2020KF0041);中央高校基本科研业务费专项资金(WUT:2021CG021);湖北省科技重大专项项目(2020AAA001)资助。

摘　　要：电动汽车电池参数的准确预测是电池故障精准诊断的前提,是电动汽车安全运行的重要技术保障。电池状态的估计受多种因素影响,驾驶行为是其中的1个重要因素,但现有电动汽车电池电压预测的研究较少考虑微观驾驶行为及驾驶工况的影响。研究提出了综合考虑驾驶行为参数的动力电池电压预测方法,针对现有电动汽车国标网联数据颗粒度低的问题,设计并开展了自然驾驶实验,采集多维精细化18辆电动汽车运行数据;通过截取典型驾驶工况数据片段,分析驾驶行为参数与动力电池参数关联性,以动力电池参数中的电池组电压作为基准参数,提取与其相关的特征指标及与其相关的特征指标,分别得到与电池组电压相关系数:0.978(电池单体电压)、0.853(电池剩余电量SOC)、0.691(加速踏板行程值)、0.683(总电流)、0.616(车速)。基于长短时记忆神经网络(LSTM)构建了以驾驶行为参数、电池组电压相关的特征参数作为输入,电池组电压作为输出的预测模型,并对LSTM超参数进行了优化。筛选出16辆电动汽车共700万条数据训练并测试预测模型,训练集与测试集的数据比例为8:2,测试结果中预测电压与实际电压间的均方误差(MSE)为0.46,平均相对误差(MRE)为0.13%。利用其余2辆车共2万条数据验证模型精度,验证结果的MSE分别为0.50,0.55,MRE分别为0.15%,0.17%。智能网联环境下基于电动汽车精准驾驶行为数据的电池电压预测模型精度明显优于传统的LSTM、递归神经网络模型。The main challenge for battery fault diagnosis for electric vehicle is the accurate prediction of battery parameters,is also critical to safe operation of electric vehicles and traffic safety.Driving conditions especially induced by driving behaviors are important factors which can affect the estimation of battery voltage of electric vehicles.However,this effect of driving behavior is rarely investigated in related works.Thus,a battery voltage prediction method considered driving behavior parameters is proposed in this study.Eighteen electric vehicles are recruited to participant the naturalistic driving study by installing the driving behavior sensors with high resolution compared with the National connected electric vehicles data collection Standard.Data segments with regard to the typical driving mode is extracted first,Then,the correlation between driving behavior parameters and power battery parameters is analyzed,and the battery pack voltage in the power battery parameters is taken as the reference parameter for correlation analysis;characteristic variables related to battery pack voltage are extracted,the correlation coefficients between characteristic variables and battery pack voltage are:0.978(battery voltage),0.853(state of charge,SOC),0.691(pedal travel value),0.683(total current),0.616(vehicle speed).Besides,a prediction model is developed based on the long short-term memory(LSTM)neural network and hyperparameters are optimized,taking driving behavior parameters and characteristic variables related to battery pack voltage as input data and taking the values of battery pack voltage as output data.The prediction model is trained and tested using a total of 7 million of data from 16 electric vehicles,and the ratio of training set and test set is 8:2.The mean square error(MSE)and mean relative error(MRE)of the proposed model are 0.46 and 0.13%,respectively.The data of the other two vehicles are used to verify the accuracy of the model.The MSE of the verification results are 0.50 and 0.55,and the MRE of t

关 键 词：交通安全 电动汽车 电压预测 长短时记忆神经网络 驾驶行为 

分 类 号：U491.5[交通运输工程—交通运输规划与管理]