Spatio-Temporal Aware Model Based on Multi-Scale Information Aggregation

Current research exhibits limitations in modeling multi-scale characteristics within temporal dependencies, where researchers predominantly employ attention mechanisms or convolutional neural networks to capture either global or local temporal relationships, consequently leading to the loss of multiscale features. To address the dynamic variations and periodic patterns in traffic flow, this paper proposes a Spatio-Temporal Aware Model based on Multi-Scale Information Aggregation (STAMSA), which integrates attention mechanisms with multiscale feature extraction to exploit and derive dynamic features from node attributes, thereby capturing multi-scale temporal dependencies in traffic flow data.