Spatial-temporal synchronous graphsage for traffic prediction

Xian Yu,Yinxin Bao,Quan Shi
DOI: https://doi.org/10.1007/s10489-024-05970-5
IF: 5.3
2024-12-07
Applied Intelligence
Abstract:The application of intelligent transportation systems (ITSs) relies heavily on accurate traffic prediction, which hinges on effectively capturing spatial-temporal features. Current methodologies often address spatial and temporal dependencies separately, which limits their ability to synchronize modeling efforts. Moreover, existing graph convolutional network (GCN) approaches primarily support transductive learning and fall short in inductive tasks. To address these challenges, this paper introduces a novel spatial-temporal synchronous GraphSAGE (STS-GraphSAGE) model for traffic prediction. By integrating spatial and temporal correlations into a unified graph structure, STS-GraphSAGE achieves synchronous learning of these dependencies. Specifically, we introduce the Spearman correlation coefficient to compensate for the spatial adjacency matrix, facilitating the construction of an inclusive spatial graph. Coupled with a causal temporal graph, this forms a spatial-temporal synchronous graph that is capable of capturing intricate dependencies across both dimensions. Furthermore, our model employs multiple STS-GraphSAGE layers equipped with attention mechanisms to inductively aggregate spatial-temporal features from neighboring nodes. Extensive experiments on real-world datasets validate the effectiveness of STS-GraphSAGE, which significantly outperforms state-of-the-art baselines in traffic prediction tasks.
computer science, artificial intelligence
What problem does this paper attempt to address?