Regional Distributed Photovoltaic Power Forecasting Method Based on Two-Stage Clustering and Hierarchical Graph Network Modeling

Abstract

Accurate regional distributed Photovoltaic (PV) power forecasting provides data support for power grid management and optimal operation. Existing graph structures are unable to effectively mine spatio-temporal correlation information or lose a lot of valid information in the modeling process. For this reason, this paper proposes an ultra-short-term regional distributed PV power forecasting method based on two-stage clustering and Hierarchical Graph Convolutional Neural network. First, a two-stage clustering method is used to optimize the clustering results on the basis of the existing K-means clustering method to maximize the use of PV power data to mine the deep similarity between PV power stations; then, based on the optimal clustering results, sub-graph and global graph are constructed by taking into account the hierarchical structure of distributed PV systems; finally, a Hierarchical Graph Convolutional Neural network forecasting model is constructed to further mine deep spatio-temporal correlation between PV power stations to achieve ultra-short-term regional distributed PV power forecasting. The superiority of the proposed method is verified by a real dataset in China.