Ultra-Short-Term Forecasting for Large-Scale Distributed Photovoltaics Based on K-Medoids-DGCN-GRU

Due to the technical and economic limitations of real-time communication, only several units can be equipped with real-time information transmission equipment in large-scale distributed photovoltaics (PVs). In order to make ultra-short-term forecasting of photovoltaic power output with high precision, reasonable installation position of communication equipment is firstly chosen, and then the corresponding forecasting framework is constructed in this paper. In the planning stage of communication equipment, multivariate variational mode decomposition algorithm is used to process photovoltaic power sequence. The sub-modes obtained by decomposition are taken as input features. K-medoids algorithm is used to cluster. The communication equipment is installed in the cluster center. In the PV ultra-short-term forecasting stage, the dynamic adjacency matrix is constructed firstly. The dynamic graph convolutional network is applied to process the PV modal characteristics. Then we use gate recurrent unit to forecast the PV output power. The forecasting result shows that root mean square error of this method is 8.37%, better than two comparison forecasting methods, which proves the effectiveness of proposed method.