摘要

Pattern recognition of residential electricity consumption refers to discover different electricity consumption patterns from electricity consumption data (ECD), which can provide valuable insights for developing personalized marketing strategies, supporting targeted demand side management, and improving energy utilization efficiency. To improve the efficiency and effectiveness of ECD analysis, we proposed an improved K-means algorithm, in which principal component analysis (PCA) was used to reduce the dimensions of smart meter time series data and the initial cluster centers were optimized. 3000 daily electricity consumption profiles (ECPs) of 1000 residents, obtained from the smart metering electricity customer behavior trials of Irish, and 2000 yearly residential ECPs from Jiangsu Province, China, were used in the experiments. The ECPs were divided into 7 and 4 clusters respectively based on their ECPs, and the characteristics of each cluster were extracted. In addition, the changes of residential electricity consumption are also reflected in the shape variation of ECPs. However, traditional similarity measurements cannot find the shape similarity of ECPs. Therefore, a shape-based clustering method was also proposed to group ECPs with similar shapes and the detailed algorithm procedures were provided. The results showed that the shape-based clustering method can effectively find similar shapes and identify typical electricity consumption patterns based on daily ECPs.