With the rapid development of the Internet of Things (IoT) and the acceleration of urbanization, traffic flow prediction has become a key research focus in intelligent transportation systems (ITS). However, owing to the highly nonlinear and dynamic spatiotemporal dependencies of traffic flow, achieving accurate and timely predictions, especially for long-term forecasting, remains a significant challenge. To address this, we propose AGCGCN, a novel prediction model designed to effectively model sensor relationships and capture spatiotemporal dependencies. By incorporating node features, local features, and global features, AGCGCN comprehensively describes sensor relationships. The model leverages a self-attention mechanism to capture directed spatial dependencies dynamically, assigning greater weights to key nodes to reflect real-time traffic flow status and directional trends. Furthermore, its multilevel hierarchical structure facilitates efficient information propagation and integrates these insights into a predictive output layer. Extensive experiments demonstrate that AGCGCN outperforms several state-of-the-art baseline models in terms of prediction accuracy, highlighting its effectiveness in traffic flow forecasting. The source code is available at https://github.com/mengfanyu-hd/AGCGCN .