一种针对电力数据异常检测的改进谱聚类算法  

An Improved Spectral Clustering Algorithm for Anomaly Detection of Power Data

作　　者：高书强[1] 李晨 GAO Shu-qiang;LI Chen(School of Information Engineering,Zhengzhou University,Zhengzhou Henan 450052,China;School of Electronic and Information Engineering,Nanjing University of Information Science&Technology,Nanjing Jiangsu 210044,China)

机构地区：[1]郑州大学信息工程学院,河南郑州450052 [2]南京信息工程大学电子与信息工程学院,江苏南京210044

出　　处：《计算机仿真》2019年第11期239-242,304,共5页Computer Simulation

基　　金：河南省科技攻关项目(172102410065)

摘　　要：近几年,数据挖掘中的异常检测在不同的研究领域发挥着越来越重要的作用,大量的研究工作也致力于数据的异常检测,包括基于距离、密度和聚类的异常检测。然而,现有的方法都将耗费较多的计算时间。因此,提出了一种改进的谱聚类算法用于异常检测,上述算法具有较高的性能。另外,上述算法可以实现对数据进行剪枝,剪枝后的数据可以减少计算的复杂性,结合基于杰卡德相似系数的方法获得相似度数值。通过真实的电力数据进行实验验证,并与现有方法在运行时间、精确度等进行对比,实验结果表明改进方法优于现有方法。In recent years, outlier detection in data mining plays a more and more important role in different areas of research, a lot of research work is also dedicated to the data anomaly detections based on distance, density and clustering. However, the existing methods consume more computation time. Therefore, this paper proposes an improved spectral clustering algorithm for anomaly detection, and the algorithm has higher performance. In addition, the algorithm can be used to prune the data, and after pruning the data can reduce the computational complexity, and the similarity value can be obtained by the method based on the Jaccard similarity coefficient. The experimental results were validated by real power data, and compared with the existing methods in terms of running time and accuracy. The experimental results show that the proposed method is superior to the existing methods.

关 键 词：数据挖掘 异常检测 谱聚类 杰卡德相似系数 

分 类 号：TP393.0[自动化与计算机技术—计算机应用技术] TP391.9[自动化与计算机技术—计算机科学与技术]