Data-driven modeling of traffic flow in macroscopic network systems

Area of Science:

• Urban planning and transportation engineering
• Computational modeling and simulation
• Data science and machine learning

Background:

• Existing urban traffic models struggle to balance realism and scalability.
• Microscopic simulators are detailed but computationally expensive.
• Macroscopic models are efficient but often oversimplify traffic dynamics.

Purpose of the Study:

• To develop a data-driven macroscopic traffic model that overcomes limitations of current approaches.
• To simulate traffic phenomena like congestion, bottlenecks, and spillbacks.
• To provide a scalable and interpretable framework for urban traffic forecasting.

Main Methods:

• Proposed a discrete-time load-exchange process over flow networks for traffic simulation.
• Utilized road-type attributes, network structure, and observed traffic density.
• Employed evolutionary optimization for parameter learning without assuming latent travel demand.

Main Results:

• The model effectively captures traffic phenomena including bottlenecks and spillbacks.
• Parameter learning adapted the model to both synthetic and real-world traffic data.
• Evaluated on diverse networks including London, Istanbul, and New York.

Conclusions:

• The developed framework offers a scalable and interpretable alternative for urban traffic forecasting.
• It balances predictive accuracy with computational efficiency.
• The model performs well across diverse network conditions and data types.