华东师范大学学报(自然科学版) ›› 2020, Vol. 2020 ›› Issue (4): 79-87.doi: 10.3969/j.issn.1000-5641.201921012

• 计算机科学 • 上一篇    下一篇

基于改进K-means的电力数据异常检测算法

吴蕊, 张安勤, 田秀霞, 张挺   

  1. 上海电力大学 计算机科学与技术学院, 上海 200090
  • 收稿日期:2019-08-25 发布日期:2020-07-20
  • 通讯作者: 张安勤,女,副教授,硕士生导师,研究方向为数据挖掘、普适计算.E-mail:aqz612@sina.com E-mail:aqz612@sina.com
  • 基金资助:
    国家自然科学基金(61772327, 61532021)

Anomaly detection algorithm based on improved K-means for electric power data

WU Rui, ZHANG Anqin, TIAN Xiuxia, ZHANG Ting   

  1. College of Computer Science and Technology, Shanghai University of Electric Power, Shanghai 200090, China
  • Received:2019-08-25 Published:2020-07-20

摘要: 异常检测方法在电力领域有着广泛的应用, 如设备故障检测和异常用电检测等. 改进了传统K-means聚类随机选择初始聚类中心的策略; 结合数据对象的密集度与最大近邻半径, 选择更加接近实际簇中心的数据点作为初始聚类中心, 并在此基础上提出了一种基于改进K-means算法的电力数据异常检测新方法. 实验表明, 上述算法具有更优的聚类效果和异常检测性能, 并且在应用于电力领域时, 算法可以有效地检测出异常电力数据.

关键词: 初始聚类中心, 密集度, 异常检测

Abstract: Anomaly detection methods are widely used for applications in the field of electric power, such as equipment fault detection and abnormal electricity consumption detection. The proposed algorithm combines densities of data objects with the maximum neighborhood radius to select data points that are closer to actual cluster centers for the initial selection; this, in turn, improves random selection of the initial cluster centers. In addition, a new anomaly detection method based on an improved K-means algorithm for electric power data is proposed. Experiments show that the algorithm is more suitable in both clustering performance and anomaly detection. When this algorithm is applied to the field of electric power, abnormal data can be effectively detected.

Key words: initial cluster centers, density, anomaly detection

中图分类号: