Road Network-Guided Fine-Grained Urban Traffic Flow Inference.

Area of Science:

• Computer Science
• Transportation Engineering
• Data Science

Background:

• Accurate fine-grained traffic flow inference from coarse-grained data is crucial for optimizing traffic management and reducing sensor costs.
• Previous methods often overlook or inadequately incorporate the significant correlation between traffic flow and road network topology.
• There is a need for advanced models that can effectively utilize road network information for enhanced traffic flow prediction.

Purpose of the Study:

• To propose a novel road-aware traffic flow magnifier (RATFM) for accurate fine-grained traffic flow inference.
• To explicitly leverage road network topology as prior knowledge to model the spatial distribution of traffic flow.
• To improve the accuracy of traffic flow prediction by integrating road network features with coarse-grained flow data.

Main Methods:

• A multidirectional 1-D convolutional layer is employed to extract semantic features from road networks.
• Road network features and coarse-grained flow data are combined to regularize short-range spatial distribution modeling.
• A transformer architecture utilizes road network features as queries to capture long-range spatial dependencies in traffic flow.

Main Results:

• The proposed RATFM method generates high-quality fine-grained traffic flow maps.
• Experimental results on three real-world datasets demonstrate superior performance compared to state-of-the-art models.
• The road-aware inference mechanism significantly enhances the accuracy of traffic flow prediction.

Conclusions:

• The RATFM model effectively integrates road network information for accurate fine-grained traffic flow inference.
• This approach offers a cost-effective solution by potentially reducing the number of required traffic monitoring sensors.
• The study highlights the importance of incorporating road network topology in traffic flow prediction models.