Traffic Flow Prediction Based on Dynamic Environment-Aware and Adaptive Multi-Scale Feature Extraction

With the rapid development of intelligent transportation systems, accurate traffic flow prediction has become a key issue in improving urban traffic management efficiency. To address the limitations of existing attention-based spatio-temporal graph convolutional networks (ASTGCN) in modeling external environmental factors and multi-scale features, this paper proposes an improved model called Dynamic Environment-aware and Multi-Scale Spatio-Temporal Graph Convolutional Network (DA-MASTGCN). First, a dynamic environment-aware attention mechanism is designed to integrate external environmental factors, such as weather and holidays, with spatio-temporal features, enhancing the model’s capability to perceive unstructured external factors. Second, an adaptive multi-scale spatio-temporal feature extraction module is introduced to capture dynamic traffic patterns at different temporal and spatial scales, utilizing multi-scale convolutions and a fusion mechanism to dynamically adjust feature weights. Experimental results demonstrate that the proposed model significantly improves prediction accuracy on multiple real-world traffic datasets, particularly under peak and abnormal traffic conditions. This study provides a novel modeling framework for complex traffic flow prediction and theoretical support for optimizing intelligent transportation systems.