基于改进k-means的电力信息系统异常检测方法

1)国网四川省电力公司信息通信公司,四川成都 610015; 2)电子科技大学计算机科学与工程学院,四川成都 611731

电力信息系统; 模式识别; 异常检测; 数据压缩; k-means算法; 聚类

An anomaly detection method for electric power information system based on improved k-means
HUANG Lin1, CHANG Jian1, YANG Fan1, LI Yi2, and NIU Xinzheng2

1)State Grid Sichuan Electric Power Company Information and Communication Corporation, Chengdu 610015, Sichuan Province, P.R.China2)School of Computer Science and Engineering,University of Electronic Science and Technology of China, Chengdu 611731, Sichuan Province, P.R.China

electric power information system; pattern recognition; anomaly detection; data compression; k-means method; clustering

DOI: 10.3724/SP.J.1249.2020.02214

备注

电力信息系统可用于管控电力设备,检测电力信息系统的异常对维持电力设备的稳定运行具有重要意义,但传统的异常检测方法难以检测电力信息系统中存在的多个指标综合异常的情况,为解决该问题,提出一种基于改进k-means算法的异常检测方法. 将数据空间划分为网格,以网格均值点映射该网格内所有样本点来压缩数据,减少了计算量; 通过引入基于聚类边界密度和簇密度移动聚类边界的机制,提高k-means算法的准确率,以准确识别正常模式; 通过计算数据与正常模式的偏离程度,检测异常. 实验结果表明,该方法能准确挖掘多指标综合异常,与其他异常检测方法比较,检测运行时间由16.44 s减少到0.55 s,异常检测的准确率提高了5.2%,在电力运维异常检测领域具有良好的工程应用前景.

The electric power information system is usually used to control the power equipment. The anomaly detection of electric power information system is very important for maintaining the stable operation of power equipment. However, the traditional anomaly detection method is difficult to detect the comprehensive anomalies of multi-indicator in the electric power information system. In order to solve this problem, an anomaly detection method based on improved k-means algorithm is presented in this paper. To reduce the amount of calculation, the data space is compressed by dividing the data space into multiple grids and all the sample points in same grid are mapped by the mean point of grid. In order to accurately identify the normal mode, the accuracy of k-means algorithm is improved by the mechanism of moving cluster boundaries based on the cluster boundary density and cluster density. Then the anomalies are detected by calculating the deviation degree between the data and normal mode. The experimental results show that the proposed method can accurately mine the comprehensive anomalies on multi-indicator. Compared with other anomaly detection methods, the running time of our method is reduced from 16.44 seconds to 0.55 seconds and the accuracy of anomaly detection is improved by 5.2%. Our method has good application prospects in the field of power operation and maintenance anomaly detection.

·