基于软分类的直流充电桩故障预警方法  

Fault Early Warning Method of DC Charging Pile Based on Soft Classification

作　　者：李芝娟 吴丹 雷珽 王宁[3] 段艳 马洪恩 LI Zhijuan;WU Dan;LEI Ting;WANG Ning;DUAN Yan;MA Hongen(State Grid Shanghai Pudong Electric Power Supply Company,Shanghai 200122,China;State Grid Shanghai Municipal Electric Power Company,Shanghai 200122,China;School of Automotive Studies,Tongji University,Shanghai 201804,China)

机构地区：[1]国网上海浦东供电公司,上海200122 [2]国网上海市电力公司,上海200122 [3]同济大学汽车学院,上海201804

出　　处：《电气传动》2023年第7期78-84,共7页Electric Drive

基　　金：上海市软科学基金项目(21692100700)。

摘　　要：针对目前直流充电桩故障频发、维护不及时的问题,提出一种基于软分类模型的直流充电桩故障预警方法。基于充电桩充电过程中的电流电压序列,构建充电功率的差分序列,结合充电过程中车端的荷电状态(SOC)序列作为该过程的充电特征,与充电桩的基础信息共同作为故障识别模型的输入特征;集成随机森林(RF)、极端梯度提升(XGBoost)和K近邻(KNN)算法构建充电桩故障识别模型,模型采用软分类方法计算充电桩的充电特征属于各类故障状态的概率值。通过设定预警阈值,在充电桩发生故障的概率超过预警阈值时,向相关维护人员发出预警信息。采用从车联网平台获取的直流充电桩数据进行验证,结果表明,该方法可对即将发生故障的直流充电桩进行预警,预警的准确率和召回率满足要求,验证了该方法的有效性和实用性,该方法对直流充电桩的运维工作具有一定的实用价值。Aiming at the problems of frequent failures and untimely maintenance of DC charging pile,a fault early warning method of DC charging pile based on soft classification model was proposed.Based on the current sequence and voltage sequence of the charging pile in the charging process,the differential sequence of charging power was constructed,the state of charge(SOC)sequence of the electric vehicle was also regarded as the feature of the charging process,above characteristics were taken together with the basic information of the charging pile as the input feature of the fault identification model;based on the integration of random forest(RF),extreme gradient boosting(XGBoost)and K-nearest neighbor(KNN)algorithms,a charging pile fault identification model was constructed,the soft classification method was used in the model to calculate the probability that the charging characteristics of the charging pile belong to various fault states.By setting the early warning threshold,when the probability of failure of the charging pile exceeded the early warning threshold,an early warning message was sent to the relevant maintenance personnel.The DC charging pile data obtained from the internet of vehicles platform was used for experiments,the results show that the proposed method can carry out early warning for the DC charging pile that is about to fail,and the precision and recall of early warning meet the requirements,which verifies the effectiveness and practicability of this method.The proposed method has certain practical value for the operation and maintenance of DC charging pile.

关 键 词：直流充电桩 故障预警 软分类 集成模型 

分 类 号：TM715[电气工程—电力系统及自动化]