Congestion is familiar to anyone who relies on a privately owned or rented automobile, taxi, or public transit for commuting, shopping and errand running. Historically, engineers and scientists exploring traffic networks frequently build mathematical models with the intent of coaxing from them insights revealing how congestion may evolve over time. Unfortunately, such models may easily become so large and complex that they are unwieldly, and simplifications are needed in order to provide passengers and drivers with accurate and rapidly computable information pertinent to route choice and departure time selection. Toward that goal, this project will employ modern statistics, simulation experiments, and notions of competition among traffic network users for available road capacity to better depict and more efficiently compute the behaviors of drivers who rely on road networks. The broader impacts of this research will be substantial. In particular, the results of this research will allow commuters and urban freight carriers to make more informed travel decisions, and governmental organizations to better regulate travel decisions within heavily congested major metropolitan regions. This study will also provide system-level experiential learning opportunities for students entering the transportation workforce.
Specifically, through a combination of experiments and machine learning and model development, this project will aim to depict the noncooperative exploration of available routes and departure times by drivers and passengers seeking to fulfill their travel demands via metropolitan road networks. A key goal of the intended research will be the efficient computation of solutions to the most prevalent type of dynamic traffic assignment (DTA), namely so-called dynamic user equilibrium (DUE). It is the lack of closed-form travel-delay operators that makes DUE computation tedious and slow. The plan is to replace the existing, differential algebraic equation (DAE) system representing travel delay with closed-form, approximate delay operators based on a form of statistical learning known as Kriging. Ad hoc experiments based on such an approach show great promise for small networks, but are not definitive. The PIs will develop the envisioned models and make developed software available as free-ware or inexpensive apps.
|Effective start/end date||8/1/17 → 7/31/21|
- National Science Foundation: $246,049.00