Traffic Flow Prediction in Attention Mechanism and Temporal Graph Convolutional Networks

To address the challenges posed by the intricate dynamic spatiotemporal dependencies within traffic networks, and the nonlinear nature of traffic data, we propose a traffic flow prediction model that uses an attention mechanism and temporal graph convolutional network (AMTGCN). First, an encoder-decoder architecture and a Graph Convolutional Recurrent Unit (NTGCCU) have been used to capture spatiotemporal dependencies effectively. Then, an attention mechanism module is employed to enhance the prioritization and emphasis on critical spatiotemporal characteristics. Finally, a memory network module has been added to discern abrupt changes in traffic conditions. To validate the efficacy and practicality of the proposed model, experimental analyses are conducted on real-world datasets. Experimental results show that the proposed method exhibits a higher prediction accuracy than current mainstream traffic forecasting models.