Short-term load forecasting study of distribution network based on FADJ-ASTGCN

With the construction of new power systems, the distribution network is gradually shifting from the receiving power network to the active distribution network. It needs to undertake more load management tasks, which puts forward higher requirements for load forecasting at various nodes of distribution network. A node load prediction method based on fully connected graph neural network model is proposed in this article to address the problems of insufficient spatiotemporal joint representation ability and prediction accuracy in existing power load forecasting models. First, a global information transfer mechanism is introduced in Attention Based Spatial-temporal Graph Convolutional Networks (ASTGCN), where each distribution node of the distribution network is connected to other nodes to form a fully connected graph structure; Then, the graph structure plus the input hourly, daily and weekly components together with the fully connected matrix is input into the spatiotemporal module composed of spatiotemporal attention mechanism and spatiotemporal convolutional network; Finally, the three component features are fused using a multicomponent attention mechanism to enhance prediction accuracy. The effectiveness of the proposed method is verified by comparing it with the existing methods using the U.S. real distribution network dataset and the U.S. transportation datasets PeMSD4 and PeMSD8 data as examples.