Medium- and long-term Load Forecasting of Residential User Groups Based on Graph Convolutional Neural Network and DBSCAN Clustering

Accurate load forecasting for residential customer clusters is an important foundation for promoting demand-side management of smart distribution grids and assisting grid companies to realize peak shaving and valley filling. However, most of the existing customer-level load forecasting methods utilize the temporal correlation of historical load sequences to construct a data-driven model, while ignoring the potential spatial correlation that exists between the electricity consumption behaviors of neighboring customers. Therefore, a residential customer cluster load forecasting method based on DBSCAN clustering and adaptive spatio-temporal synchronization map convolutional neural network is proposed. First, DBSCAN clustering is used to divide the residential clusters into different groups according to the similarity of electricity use behavior; then, based on the number of subgroups in the residential clusters, the historical load data of the residential users in each group, and the correlation between the load sequences of the residential users in each group, spatio-temporal maps are constructed for the load prediction of the residential clusters; and finally, adaptive spatiotemporal synchronization map convolutional neural network is used to achieve the medium- and long-term load prediction of the residential clusters. Finally, an adaptive spatiotemporal synchronization map convolutional neural network is used to realize medium- and long-term load forecasting for residential user clusters. The accuracy and effectiveness of the proposed method are tested and verified by the open data set of residential loads, and the experimental results show that the proposed method can fully exploit and utilize the spatio-temporal correlation between different residential loads to improve the accuracy of residential load forecasting compared with the benchmark forecasting methods.