基于奇异值分解和改进FCM的日负荷聚类方法  

A daily load clustering method based on singular value decomposition and improved FCM

作　　者：赵省军 和建军 姚黄金 宋长城 侯世杰 欧星良 ZHAO Shengjun;HE Jianjun;YAO Huangjin;SONG Changcheng;HOU Shijie;OU Xingliang(State Grid Gansu Electric Power Company Wuwei Power Supply Company,Wuwei 733000,China;College of Electrical and Electronic Information,Xihua University,Chengdu 610039,China)

机构地区：[1]国网甘肃省电力公司武威供电公司,甘肃武威733000 [2]西华大学电气与电子信息学院,四川成都610039

出　　处：《电气应用》2022年第2期20-26,共7页Electrotechnical Application

基　　金：国网甘肃省电力公司科技项目(52273018000L)。

摘　　要：随着电力体制改革的不断深入,为争夺市场份额、吸引潜在用户购电并提高自身收益,售电公司愈发重视用户的用电体验。对用户日负荷曲线的聚类分析能够有效挖掘用户的用电行为特性,进而为售电公司提供决策依据。针对FCM算法运行时间较长、对初始数据敏感、容易陷入局部最优、需要人为给定类簇数以及聚类结果不稳定等问题,提出了一种基于奇异值分解(Singular Value Decomposition,SVD)和改进FCM的日负荷聚类方法。首先对日负荷数据进行奇异值分解降维;然后,利用KNN和DPC算法形成初始类簇中心矩阵,并在FCM算法的迭代寻优过程中通过局部密度和邻近点对隶属度进行修正;最后,以某地区工商业用户日负荷曲线进行算例分析。结果表明,与传统聚类算法相比,该方法的聚类结果更准确、更稳定,运行速度更快。With the development of national electric power market, retailers pay more attention to the user’s electricity experience for market share, attracting potential users and improving their own profits. The clustering analysis of daily load curves can effectively mining users’ electricity consumption behavior, and then provide decision basis for retailers. A daily load clustering method based on singular value decomposition( SVD) and improved FCM is proposed to solve the problems of FCM,such as long calculation times, sensitivity to initial data,low efficiency and local optimum, setting cluster number and unstable clustering results. Firstly, the dimension of daily load data is reduced by SVD. Then, the initial cluster center matrix is formed by KNN and DPC algorithms,and the membership degree is corrected through local density and adjacent points in the iterative optimization process of FCM algorithm. The experimental results show that clustering results obtained from the proposed method perform better in accuracy, stability and calculation speed.

关 键 词：用电行为分析 负荷曲线聚类 奇异值分解 FCM 密度峰值聚类 KNN 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TM714[自动化与计算机技术—计算机科学与技术]