面向负荷特征分析的地理分布式协同聚类方法  

Geo-distributed Collaborative Clustering Method for Load Characteristic Analysis

作　　者：刘家丞 吴江[1] 刘鹏远 徐占伯 李晓鹏 管晓宏[1] LIU Jiacheng;WU Jiang;LIU Pengyuan;XU Zhanbo;LI Xiaopeng;GUAN Xiaohong(Ministry of Education Key Lab for Intelligent Networks and Network Security(Xi'an Jiaotong University),Xi'an 710049,China;Xi'an Power Supply Company of State Grid Shaanxi Electric Power Company,Xi'an 710048,China)

机构地区：[1]智能网络与网络安全教育部重点实验室(西安交通大学),陕西省西安市710049 [2]国网陕西省电力公司西安供电公司,陕西省西安市710048

出　　处：《电力系统自动化》2022年第15期112-120,共9页Automation of Electric Power Systems

基　　金：国家重点研发计划资助项目(2016YFB0901900);国家自然科学基金资助项目(61803297)。

摘　　要：能源互联网架构下,用户数据传输延迟和电力公司管理规定促使电力数据中心在全国各地建立,电力数据因而呈现地理分布式,对此研究了在地理分布式情景下的用户负荷特征聚类算法。首先,对于地理节点内用户负荷的特征分析,在采用主成分分析与负荷指标特征加权组合算法的基础上,提出了考虑密度峰值信息的K-means算法,并为地理节点间的信息共识提供了支撑。其次,针对用户需求的地理分布式网络化感知结构,构建了考虑特征迁移的分布式聚类模型框架,提出将节点局部信息通过参数共识得到全局聚类模型的分布式K-means算法,在节点间仅传递公开信息的前提下,实现了用户特征的全局聚类。最后,通过爱尔兰、中国北方部分城市的负荷数据对模型及算法进行验证。结果显示,分布式K-means能利用全局信息、考虑不同区域的差异来更好地识别典型用电曲线,并且算法具有较好的可迁移性。Under the framework of Energy Internet.The transmission delay of user data and management requirements of power grid companies promote the establishment of power data centers all over the country.Therefore,the power data naturally presents geographic distribution.In this regard,the clustering algorithm of user load characteristics in geo-distributed scenarios is studied.Firstly,for the characteristic analysis of user load in geographical nodes,based on the principal component analysis(PCA)-load index feature weighted combination algorithm,a K-means algorithm considering the density peak information is proposed,which provides the support for information consensus among geographical nodes.Secondly,according to the geo-distributed network perception structure of user requirement,the framework of distributed clustering model considering feature migration is constructed,and the distributed K-means algorithm is proposed to obtain the global clustering model by using the local information of nodes through parameter consensus,which realizes the global clustering of user characteristics on the premise that only public information is transmitted between nodes.Thus,the global clustering of user characteristics is realized.Finally,the model and algorithm are verified by user load data from the cities in Ireland and northern China.The results show that distributed K-means can use global information and consider differences in different regions to better identify typical power consumption curves,and the algorithm has better transferability.

关 键 词：分布式聚类 负荷特征 参数共识 密度峰值聚类 迁移学习 

分 类 号：TM714[电气工程—电力系统及自动化] TP311.13[自动化与计算机技术—计算机软件与理论]