DT-ASGNRU: A Comprehensive Framework for Traffic Flow Forecasting

Traffic prediction plays a crucial role in alleviating congestion, enhancing road utilization, and reducing traffic accidents. In order to tackle the challenge of dynamic spatiotemporal correlations in traffic data, this paper introduces a new traffic flow prediction model, DT-ASGNRU. This model combines Gated Recurrent Units (GRU), a multi-attention mechanism, and a Graph Convolutional Network (GCN), while introducing an expansion convolution layer to enhance the multi-attention mechanism. This modification expands the acceptance domain, which enhances the model’s capacity to grasp local temporal dependencies. The expanded convolution based on multi-head attention mechanism enhances the sensitivity of the model to time series data and captures the temporal correlation in traffic flow effectively. Additionally, the integration of attention mechanisms with graph convolutions allows for dynamic modeling of spatial relationships within traffic networks. We conducted assessment trials utilizing an extensive collection of genuine traffic information, and the outcomes indicate that DT-ASGNRU surpasses all comparative methodologies in terms of performance.