Clustering of Daily Load Curves Based on Graph Convolutional Network

Fluctuating wind turbine load data reflect the power generation patterns of turbines and the behavioral characteristics of users. Clustering of daily load curves facilitates the extraction of characteristics, excavates the electricity consumption pattern, and provides a foundation for grid planning and operation. In allusion to the constraints imposed by traditional methods, including inefficiency and challenges of intricate nonlinear relationships, this paper proposes a clustering method based on graph convolutional network (GCN). Firstly, cosine similarity is used to construct an adjacency matrix, measuring the similarity between different node load data. Secondly, GCN maps the node features to a new space, extracting the potential data features. Finally, the load clustering task cooperates with K-means. The case study is based on data from the four-machine, two-zone simulation model built by SEU, with clustering results quantitatively analyzed by DBI and SC. The results indicate that this method significantly improves the clustering performance compared to K-means, PCA+K-means, and the other neural network models, such as LSTM and VAE. It can effectively extract temporal features and accurately analyze electricity consumption patterns.