摘要
针对触电事故具有随机性、难以识别的问题,提出一种基于循环功率谱密度特征和聚类分析的触电事故识别方法。首先利用循环功率谱分析获得触电前、后剩余电流信号的循环功率谱三维图,对剩余电流中150Hz成分对应的循环谱切面分析,提取特征循环频率1200Hz的线谱占比,定义四种循环频谱特征;然后利用K-means聚类分析对不同维度的循环频谱组合特征进行聚类,提出了触电识别判据,同时提出加偏置修正的欧式距离测度提高了聚类识别准确率。结果表明:单相电路时循环谱2、3、4组合特征与其他维度特征相比对触电事故的识别率最高,为94.67%,对应的触电前、后剩余电流聚类中心分别为20.597、57.682、4.773和4.102、11.387、0.923;三相电路时最佳识别特征为循环谱特征4,对应的触电前、后剩余电流聚类中心分别为16.136和2.197;加偏置项修正后的欧氏距离对触电事故识别率达到了99.33%。研究结果为有效识别触电事故提供一定理论参考。
In order to solve the problems that electric shock accidents are difficult to identify due to its’randomness,this paper proposes a new method based on cyclic power spectral density characteristics and cluster analysis.First,the cyclic power spectrum is used to analyze the residual current signal to obtain a three-dimensional cyclic spectrum diagram of the residual current signal before and after the electric shock.According to the slice spectral analysis of the 150Hz component of the residual current,the line spectrum proportion of the characteristic cycle frequency,1200Hz,is extracted,and four kinds of cycle spectrum features are defined to describe electric shock accidents.In order to extract the shock recognition criterion,K-means clustering analysis was used to cluster the combined features of the cyclic spectrum of different dimensions.At the same time,it is proposed that the Euclidean distance measure with offset term improves the accuracy of cluster recognition.The result shows that the combined features of cyclic spectrum 2,3,and 4 in single-phase circuits have the highest recognition rate for electric shock accidents compared with other dimensional features,which is 94.67%.The corresponding cluster centers of residual current before and after electric shock are 20.597,57.682,4.773 and 4.102,11.387,0.923 respectively.The best recognition feature for three-phase circuits is the cyclic spectrum feature 4,the corresponding cluster centers of residual current before and after electric shock are 16.136 and 2.197 respectively.The Euclidean distance with offset term has a 99.33%recognition rate for electric shock accidents.The research results provide some theoretical reference for effective identification of electric shock accidents.
作者
李春兰
罗杰
王长云
王海杨
杜松怀
Li Chunlan;Luo Jie;Wang Changyun;Wang Haiyang;Du Songhuai(College of Mechanical and Electrical Engineering,Xinjiang Agricultural University,Urumqi,830052,China;School of Mechanical and Electronic Engineering,Xinjiang Vocational University,Urumqi,830013,China;College of Information and Electrical Engineering,China Agricultural University,Beijing,100083,China)
出处
《电工技术学报》
EI
CSCD
北大核心
2021年第22期4677-4687,共11页
Transactions of China Electrotechnical Society
基金
国家自然科学基金资助项目(51467021)。
关键词
触电
剩余电流
循环功率谱
K-MEANS聚类
特征识别
Electric shock
residual current
cyclic power spectral
K-means clustering
feature recognition
