Data-enhanced machine recognition model of DC serial arc in electric vehicle power system

作     者：Xia, Kun Zhang, Zihan Liu, Bangzheng Guo, Haotian Yu, Wei Xu, Jingjun Dong, Hui 

作者机构：Univ Shanghai Sci & Technol Elect Engn Dept 516 JunGong Rd Shanghai 200093 Peoples R China Hella Shanghai Elect Co Ltd R&D Ctr Shanghai 201201 Peoples R China 

出 版 物：《IET POWER ELECTRONICS》 (IET Power Electron.)

年 卷 期：2020年第13卷第19期

页      面：4677-4684页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 

基　　金：HELLA RD Foundation [3A16302050, 3A17302064] National Natural Science Foundation of China [51207091, 51637001] 

主　　题：arcs (electric) fault diagnosis electric vehicles support vector machines fault currents Fourier transforms power engineering computing direct current arc arc detection algorithm support vector machine model DC serial arc detection EV power system pre-detection algorithm false detection rate DC arc fault arc fault data enhancement model arc fault current data SVM classification model high power experiment data-enhanced machine recognition model electric vehicle power system windowed Fourier transform classification 

摘      要：Electric vehicle (EV) power system is the key to the development of EVs. If direct current (DC) arc occurs in the power system, it is difficult to extinguish at zero point. The arc fault will release a huge amount of energy and continuous sparking, which may cause spontaneous combustion or even explosion. In this study, an arc detection algorithm based on the classification of windowed Fourier transform and support vector machine (SVM) model is proposed for DC serial arc detection of EV power system. In order to optimise the arc detection algorithm, the authors use the pre-detection algorithm, which can effectively reduce the false detection rate of DC arc fault and ensure the reliability of detection algorithm. In addition, they propose an arc fault data enhancement model, which can generate arc fault current data. Finally, the experimental results show that the arc detection algorithm has a high accuracy and a false detection rate of 0%. After data enhancement, it has generalisation of the SVM classification model under the condition of high power experiment.