基于分布式计算的海量用电数据分析技术研究  

Research on Power Usage Behavior Analysis Based on Distributed Computing

作　　者：蒋菱[1] 王旭东[1] 于建成[1] 袁晓冬[2] 

机构地区：[1]国网天津市电力公司,天津300010 [2]江苏省电力科学研究院,江苏南京210036

出　　处：《计算机技术与发展》2016年第12期176-181,共6页Computer Technology and Development

基　　金：国家自然科学基金资助项目(51407025);江苏省科技支撑计划(社会发展)(BE2013737);天津电力公司科技项目(SGTJDK00DWJS1500033)

摘　　要：用电行为分析技术对供电企业掌握用户用能方式、调整生产计划以及进行电网规划有着较大的现实意义。传统用电行为分析多利用少量样本数据,由于数据源覆盖面的问题往往容易造成结果偏差。借助大数据技术,可以利用海量用电数据提高用电行为分析的准确性。针对用电行为分析在处理海量数据时效率低下的问题,提出了基于MapReduce技术的模糊C均值聚类(FCM)并行算法,通过将FCM算法的迭代过程分解到Map和Reduce两个步骤中,可以有效地提高聚类过程中数据对象和聚类中心的相似度计算效率。在此基础上,利用所提出的FCM并行算法对居民用电数据的四个特征进行聚类分析。实验结果表明,所提算法可以提高海量用电数据聚类分析的效率,证明了计算模型的可行性。The power usage behavior analysis technology can be used to acquire costumer power usage pattern, adjust power generation schedule and plan gird development. Thus, it is meaningful to power grid company. Traditional power usage behavior analysis only uses small volume of data. The limited data will draw to inaccurate result. This problem can be solved by using large scale of data. In allusion to the problem about electricity behavior analysis in the low efficiency of dealing with huge amounts of data,the Fuzzy C-Means clustering （FCM） parallel algorithm based on MapReduce is put forward. By decomposing the iterative process of FCM algorithm into two steps of Map and Reduce,it can effectively improve the efficiency of similarity computing between the data objects and the clustering cen- ters. On this basis, the four characteristics of resident electrical data are clustering analyzed by using the proposed FCM parallel algorithm. The experimental results show that the proposed algorithm can improve the efficiency of mass data clustering analysis and also proves the feasibility of the model.

关 键 词：MAPREDUCE 模糊C均值聚类 用电行为分析 大数据 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]