基于驾驶行为量化因子的电动汽车剩余续驶里程预测  

Prediction of the remaining mileage of electric vehicles based on the quantitative factors of driving behavior

作　　者：李骏 孙亚诚 李继秋 单丰武 刘俊宇 曾建邦 LI Jun;SUN YaCheng;LI JiQiu;SHAN FengWu;LIU JunYu;ZENG JianBang(School of Electromechanical and Vehicle Engineering,East China Jiaotong University,Nanchang 330013,China;School of Automobile Studies,Tongji University,Shanghai 201804,China;New Energy Vehicle Corporation,Jiangxi Jiangling Motors Group,Nanchang 330013,China)

机构地区：[1]华东交通大学机电与车辆工程学院,南昌330013 [2]同济大学汽车学院,上海201804 [3]江西江铃集团新能源汽车有限公司,南昌330013

出　　处：《中国科学：技术科学》2024年第5期955-967,共13页Scientia Sinica(Technologica)

基　　金：国家自然科学基金(批准号:51806066);江西省重点研发计划(编号:20223BBE51016)资助项目。

摘　　要：针对基于聚类算法提取驾驶行为难以考虑聚类类别中每个样本点对能耗的影响,建立了一种基于驾驶行为量化因子的电动汽车剩余续驶里程预测模型.首先,从车速和加速度等影响能耗方面选取了七个驾驶行为评价指标,使用随机森林算法对指标进行赋权,将指标归一化后的加权和作为驾驶行为量化因子;其次,将电池荷电状态、驾驶行为量化因子、环境温度、电器用电负荷率和工况信息作为模型输入,建立电动汽车剩余续驶里程预测模型.结果表明:考虑驾驶行为量化因子的预测模型相比未考虑时预测的均方根误差和平均绝对误差更小;与支持向量回归、前馈神经网络和循环神经网络算法相比,长短期记忆网络(LSTM)的预测效果最好.最后,通过实车运行数据,验证了该模型可提高剩余续驶里程预测的准确度.可见,本文所取得的研究成果对改善驾驶员驾驶体验具有重要意义.A model for predicting the remaining mileage of electric vehicles was established by addressing the challenge of overlooking the impact of each sample point in a clustering category on energy consumption when using clustering algorithms to extract driving behaviors.The model is based on the quantitative factors of driving behavior.First,seven driving behavior evaluation indicators,such as vehicle speed and acceleration affecting energy consumption,were selected.The random forest algorithm was employed to assign weights to these indicators,and the normalized weighted sum was used as the quantitative factor of driving behavior.Second,a remaining mileage prediction model for electric vehicles was constructed using the state of charge of the battery,quantitative factors of driving behavior,environmental temperature,electrical load rate,and operational information as inputs.Results showed that the prediction model that considers the quantitative factors of driving behavior has a smaller root mean square error and mean absolute error than the prediction model that does not consider the quantitative factors of driving behavior.Compared with the support vector regression,backpropagation,and recurrent neural network algorithms,the long short-term memory network exhibits the best predictive performance.Finally,using real vehicle operation data,this study validated that the model improves the accuracy of predicting the remaining mileage.This research achievement is significant in enhancing the driving experience of drivers.

关 键 词：电动汽车 剩余续驶里程 驾驶行为 量化因子 LSTM 

分 类 号：U469.72[机械工程—车辆工程]