SHORT-TERM PREDICTION METHOD OF WIND POWER CLUSTERS BASED ON GRAPH CONVOLUTION NEURAL NETWORK UNDER SPITIO-TEMPORAL CHARACTERISTICS
In order to solve the problems that the traditional wind power clusters prediction methods ignore the meteorological correlation characteristics of different locations and the single site prediction cannot quickly obtain the overall power of the wind power clusters,and fully consider the complex spatio-temporal characteristics of wind power clusters coupling,a short-term prediction method of wind power clusters based on attention mechanism and spatio-temporal graph convolution neural network is proposed.Initially,the mutual information between the historical power of wind farms in the region is calculated,the feature adjacency matrix is extracted,and the meteorological characteristic variables that affect the cluster power,which are converted into meteorological graph data.Furthermore,a graph convolution network(GCN)model is constructed to extract the correlation characteristics of meteorological graph nodes from non-European space.The gated recurrent unit(GRU)network,which incorporates the attention mechanism(AM),is fed to enhance the contribution of key information in the temporal features to the power of wind power clusters.Finally,the progressiveness and adaptability of the proposed method is verified based on the actual operation data of the wind power cluster in a certain province in Western China.
wind powergraph data structuresdeep learningspatio-temporal characteristicsgraph convolutional neural networkattention mechanism
NETL
NSTL
万方数据
