基于加权表决集成聚类的居民用电行为回归分析

Regression Analysis of Residential Electricity Consumption Behavior Based on Weighted Voting Ensemble Clustering

居民用电行为分析是深度挖掘居民需求响应潜力,提升精准电力服务水平的基础。针对居民用户电力日负荷曲线数据,提出一种基于加权表决的集成聚类方法。将4种常用聚类算法视作选民成员进行投票表决,并根据聚类有效性指标赋权从而集成成员算法的聚类结果,以结合不同算法的性能优势。提取负荷曲线特性指标对居民负荷曲线加权表决聚类得到6种典型用电模式,采用多元逻辑回归方法分析居民用电模式与其家庭特征之间的驱动联系。案例分析结果表明所提方法提高了负荷曲线聚类效果,鲁棒性更优,且用电模式与多项家庭特征间表现出显著的正或负相关联系。

The analysis of residential electricity consumption behavior is the basis for exploring the potential of residential demand response and improving the level of electricity service. An ensemble clustering method based on weighted voting is proposed for residential customers’ electricity daily load curve. Four commonly used clustering algorithms are treated as members for voting, and the clustering results of the member algorithms are ensembled by assigning weights according to the clustering validity index to combine the advantages of algorithms. The load curve characteristic index is extracted to obtain six typical electricity consumption patterns by weighted voting clustering of residential load curves, and multiple logistic regression is used to analyze the link between electricity consumption patterns and their household characteristics. The results of the case show that the proposed method improves the clustering effect with better robustness and show significant positive or negative correlations between electricity consumption patterns and several household characteristics.