I propose structuring my contribution in two presentations—one dealing with model fundamentals and mathematical aspects, and another dealing with applications of the models to real-time control of traffic systems.
Models: The rationale and structure of a simulation-based framework for real-time traffic estimation and prediction will be presented. Combining Kalman filter methods for O-D estimation and prediction, with simulation logic for flow modeling, along with simple control-theoretic procedures for both short-term and long-term adjustment of predictions, the model system we describe has been developed and tested over a 15-year period. Driven by both historical information and real-time measurements from various sources in a traffic network, the model system is implemented in an asynchronous rolling-horizon framework, and is intended to support a host of management and control strategies ranging from traffic control to personalized user information systems.
Applications: The range of applications we focus on include real-time information to users, route guidance, dynamic pricing and traffic signal control, including recent real-world applications to severe weather mitigation. These are all predictive strategies that directly rely on the traffic flow, speed and travel time predictions generated by our system. We describe the challenges in such real-world applications, how those have been overcome, and the development of decision support tools such as scenario manager software found to be essential for field application. We describe the growing role of data mining, machine learning and clustering techniques in the development of operational scenarios intended to trigger appropriate historical data inputs as well as control action recommendations.