Residential Load Pattern Clustering Based on Smart Meter Data Using Weighted Self-Organizing Map

With arrival of big data of smart meters, a large number of residential power consumption data are collected according to different sampling frequency, namely Residential Load Profiles (RLPs). In this paper, RLPs of smart meter customers are analyzed by clustering, which is of great significance to load management of smart grid. A two-stage Weighted Self-Organizing Map (WSOM) clustering algorithm and a clustering performance evaluation method, SSE-DBI, combining Sum of Squares Error (SSE) and Davies-Bouldin (DBI) are proposed. In first stage, Principal Component Analysis (PCA) is used to reduce the dimension of the RLPs. When dimension reduced data is fed into SOM network for clustering, update of weights of SOM is weighted according to PCA, and these clustering centers, namely Typical Residential Load Profiles (TRLPs) of each customer are obtained after some iterations of training. In second stage, above processing is repeated for TRLPs of each customer, TRLPs of all customer are obtained. According to SSE-DBI, final optimal cluster number and clustering performance score of the model are determined. Compared with several benchmark methods, the proposed method obtains optimal performance.