Data-driven analysis and forecasting of highway traffic dynamics

Area of Science:

• Dynamical Systems Theory
• Transportation Engineering
• Data Science

Background:

• Vehicular traffic systems are complex, high-dimensional, nonlinear dynamical systems influenced by unpredictable factors like human behavior and weather.
• Developing accurate predictive models for traffic flow is challenging due to this inherent complexity.
• Increasing transportation demands necessitate advanced methods for traffic analysis and forecasting.

Purpose of the Study:

• To present Koopman mode decomposition as a model-free, data-driven technique for traffic dynamics analysis.
• To demonstrate the application of this method for forecasting highway network conditions.
• To identify and characterize spatiotemporal patterns within traffic data.

Main Methods:

• Utilized Koopman mode decomposition on traffic data from the Federal Highway Administration and the California Department of Transportation.
• Applied data-driven analysis to reconstruct observed traffic data.
• Identified growing and decaying patterns within the traffic system.

Main Results:

• Successfully reconstructed observed traffic data using Koopman mode decomposition.
• Distinguished significant growing and decaying patterns in traffic dynamics.
• Uncovered a hierarchy of known and novel spatiotemporal patterns.
• Demonstrated the capability to forecast highway network conditions.

Conclusions:

• Koopman mode decomposition provides a powerful, model-free approach for understanding and predicting traffic dynamics.
• This data-driven methodology offers a robust solution for traffic agencies facing increasing transportation demands.
• The technique facilitates the identification of complex spatiotemporal patterns crucial for effective traffic management.