面向海量用户用电特性感知的分布式聚类算法

Abstract

智能电表的普及促进了配用电大数据的发展。通过对用户用电数据的挖掘和用电特性的感知，能够有效识别用户用电模式、评估需求响应潜力、指导电价制定等。然而，用户用电数据一方面随时间不断更新，增长迅速，呈海量态势；另一方面，数据采集点分布在用户侧，具有极强的分散性。针对海量、分散的用电数据带来的挑战，文中提出一种新的分布式聚类算法。首先利用自适应k-means聚类算法对分布在各区域的用电数据进行局部聚类分析，提取各局部数据的典型负荷曲线，构建局部模型；然后利用传统聚类算法对获取的局部模型进行二次聚类分析，获取全局的典型负荷曲线，构建全局模型；最后向局部数据中心反馈全局聚类结果，实现全局聚类分析。通过爱尔兰实际量测用电数据证明了所提出算法的有效性。

The popularity of smart meters promotes the development of big data in smart power distribution and consumption systems. Data mining for smart meter data and awareness of electricity consumption characteristics are of great significance for consumption patterns recognition, demand response potential evaluation, and electricity price design. However, on one hand, the volume of smart meter data will grow dramatically with higher data collection frequency; on the other hand, the collected smart meter data has strong dispersion. To tackle the challenges brought by the massive and distributed smart meter big data, a novel distributed clustering algorithm is proposed. Firstly, the adaptive k-means algorithm is applied to each local data center so the typical load profiles can be extracted and the local model can be built. Then slightly revised traditional clustering algorithms are applied to the local models for secondary clustering analysis, thus the global model is built. Finally, the effectiveness of the proposed algorithm is verified by an actual example from Ireland.