Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MHGNet: Multi-Heterogeneous Graph Neural Network for Traffic Prediction

Mei Wu, Yiqian Lin, Tianfan Jiang, Wenchao Weng

TL;DR

MHGNet targets traffic prediction on multi-heterogeneous graphs by decoupling single-pattern signals into multi-pattern representations using daily/weekly embeddings and a node embedding, enabling efficient clustering into pattern-based groups. It then constructs dynamic, cluster-specific spatiotemporal subgraphs (DSTGG) and applies gated, multi-hop subgraph convolutions (SIE) with skip connections to fuse information across scales, followed by a regression-based prediction head. The approach achieves superior performance on four benchmarks and is supported by extensive ablations showing the necessity of each module, notably the Node Clusterer and DSTGG, for capturing multi-dimensional spatiotemporal heterogeneity. This work advances traffic forecasting by converting complex heterogeneous graphs into simpler, tractable subgraphs while preserving rich spatiotemporal information, enabling more accurate and scalable ITS applications.

Abstract

In recent years, traffic flow prediction has played a crucial role in the management of intelligent transportation systems. However, traditional forecasting methods often model non-Euclidean low-dimensional traffic data as a simple graph with single-type nodes and edges, failing to capture similar trends among nodes of the same type. To address this limitation, this paper proposes MHGNet, a novel framework for modeling spatiotemporal multi-heterogeneous graphs. Within this framework, the STD Module decouples single-pattern traffic data into multi-pattern traffic data through feature mappings of timestamp embedding matrices and node embedding matrices. Subsequently, the Node Clusterer leverages the Euclidean distance between nodes and different types of limit points to perform clustering with O(N) time complexity. The nodes within each cluster undergo residual subgraph convolution within the spatiotemporal fusion subgraphs generated by the DSTGG Module, followed by processing in the SIE Module for node repositioning and redistribution of weights. To validate the effectiveness of MHGNet, this paper conducts extensive ablation studies and quantitative evaluations on four widely used benchmarks, demonstrating its superior performance.

MHGNet: Multi-Heterogeneous Graph Neural Network for Traffic Prediction

TL;DR

MHGNet targets traffic prediction on multi-heterogeneous graphs by decoupling single-pattern signals into multi-pattern representations using daily/weekly embeddings and a node embedding, enabling efficient clustering into pattern-based groups. It then constructs dynamic, cluster-specific spatiotemporal subgraphs (DSTGG) and applies gated, multi-hop subgraph convolutions (SIE) with skip connections to fuse information across scales, followed by a regression-based prediction head. The approach achieves superior performance on four benchmarks and is supported by extensive ablations showing the necessity of each module, notably the Node Clusterer and DSTGG, for capturing multi-dimensional spatiotemporal heterogeneity. This work advances traffic forecasting by converting complex heterogeneous graphs into simpler, tractable subgraphs while preserving rich spatiotemporal information, enabling more accurate and scalable ITS applications.

Abstract

In recent years, traffic flow prediction has played a crucial role in the management of intelligent transportation systems. However, traditional forecasting methods often model non-Euclidean low-dimensional traffic data as a simple graph with single-type nodes and edges, failing to capture similar trends among nodes of the same type. To address this limitation, this paper proposes MHGNet, a novel framework for modeling spatiotemporal multi-heterogeneous graphs. Within this framework, the STD Module decouples single-pattern traffic data into multi-pattern traffic data through feature mappings of timestamp embedding matrices and node embedding matrices. Subsequently, the Node Clusterer leverages the Euclidean distance between nodes and different types of limit points to perform clustering with O(N) time complexity. The nodes within each cluster undergo residual subgraph convolution within the spatiotemporal fusion subgraphs generated by the DSTGG Module, followed by processing in the SIE Module for node repositioning and redistribution of weights. To validate the effectiveness of MHGNet, this paper conducts extensive ablation studies and quantitative evaluations on four widely used benchmarks, demonstrating its superior performance.
Paper Structure (14 sections, 11 equations, 5 figures, 7 tables)

This paper contains 14 sections, 11 equations, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. STD Module
  4. Node Clusterer
  5. DSTGG Module
  6. SIE Module
  7. Output and prediction
  8. Experiments
  9. Datasets
  10. Baseline Methods
  11. Experiment Settings
  12. Experiment Results
  13. Ablation Experiments
  14. Conclusion

Figures (5)

  • Figure 1: The nodes have different sizes of spatiotemporal relationship similarities. Nodes of the same type have similar traffic patterns.
  • Figure 2: Transforming multi-heterogeneous graph into simple subgraph convolution, with different types of nodes contributing traffic features from different traffic pattern structures.
  • Figure 3: Overall framework diagram (MHGNet)
  • Figure 4: (a) Constructing the limit point. (b) Clustering the nodes using the Euclidean distance.
  • Figure 5: (a) Node clustering and node repositioning demonstration. (b) SIE Module schematic diagram.