Spatio-Temporal Heterogeneity-Oriented Graph Convolutional Network for Urban Traffic Flow Prediction.

Area of Science:

• Urban vehicular ad hoc networks (VANETs)
• Traffic prediction and forecasting
• Graph convolutional networks (GCN)

Background:

• Urban vehicular ad hoc networks (VANETs) utilize cross-domain data for enhanced traffic prediction.
• Spatial and temporal heterogeneity in data complicates normalization and prediction model construction.
• Dynamic external factors introduce cumulative impacts on traffic pattern prediction.

Purpose of the Study:

• To propose a Spatio-Temporal Heterogeneity-oriented Graph Convolutional Network (SHGCN) to address challenges in urban traffic prediction.
• To leverage spatial heterogeneity and external factors like air quality for improved traffic forecasting.
• To investigate cross-correlation characteristics using a hybrid GCN-GRU model.

Main Methods:

• Developed SHGCN to analyze spatial heterogeneity beyond simple adjacency for traffic stream correlations.
• Integrated air quality data as external factors for street-level traffic forecasting.
• Employed a hybrid Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU) model to capture cross-correlation characteristics.

Main Results:

• The SHGCN model demonstrated significant improvements over baseline models, with Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) reductions between 2.91% and 41.26%.
• Ablation studies confirmed that incorporating air quality factors enhances traffic prediction performance.
• The model effectively captures complex correlations between air pollutants, traffic dynamics, and road network topology.

Conclusions:

• The proposed SHGCN effectively handles spatio-temporal heterogeneity in urban traffic data.
• Integrating air quality data improves the accuracy and robustness of traffic prediction models.
• The SHGCN approach offers a validated method for understanding intricate relationships within urban transportation systems.