An attention mechanism-based method for predicting traffic flow by GCN

Traffic flow prediction is an important part of Intelligent Transportation System (ITS), and studying it is of great significance to alleviate traffic congestion and improve traffic conditions. However, due to the complexity and dynamic changes of the urban road network, it is difficult to accurately predict using a single model. This paper proposes a RES2GCN traffic prediction model based on stacked graph convolutional layer (GGCN) and Attention model. The stacked graph convolutional layer (GGCN) consists of graph convolutional network (GCN) and gated linear unit (GLU) composition, used to extract the main features of the urban road network, the attention mechanism adjusts the time weight to output the traffic flow prediction results. In this paper, pems08 data set and Seattle data set are used for prediction. Experimental analysis and comparison show that, compared with other baseline methods, the accuracy of pems08 data set is improved by 2.50%, and the accuracy of Seattle data set is improved by 4.3%.