Spatiotemporal Analysis of Urban Traffic Congestion Via Deep Convolutional Architectures

Urban traffic congestion is a key issue affecting traffic efficiency, environmental sustainability, and economic productivity. Traditional transportation analysis methods mainly rely on predefined static adjacency matrices and often separately process spatial and temporal features (CNN), making it difficult to capture the inherent complex spatiotemporal dependencies in urban traffic patterns. This paper proposes a deep learning-based spatiotemporal traffic congestion analysis method using Graph Convolutional Networks (GCN). We design a novel Deep Spatiotemporal Traffic Network (DSTN) that models the spatiotemporal dynamic features of traffic by integrating 3D convolution and attention mechanisms, enhancing the model's ability to capture cyclical patterns and distinguish dynamic peak periods. The model is trained and evaluated on a large-scale traffic dataset collected using ring coil detectors and GPS trajectories.