Real Time Toll Optimization based on Predicted Traffic Conditions

Road pricing is an effective method of demand management. Pricing on highway managed lanes is usually implemented as time-of-day or dynamic tolling in practice. Toll rates are usually updated according to latest traffic measurement and based on pre-defined rules. Researches on highway pricing can be generally categorized as analytical, reactive or optimization-based approaches. The limitations of current studies are compared and discussed in this thesis. A new framework is proposed which aims to develop an adaptive integrated simulation-optimization framework that brings together several enhancements: real time, predictive, simulation-based and consistent. The main components of the framework include DTA model, DynaMIT, for evaluating control strategies, optimization module solving for optimal solution and real-life traffic system providing surveillance data. Optimization problem is formulated with rolling horizon scheme, and presented with basic models for revenue maximization. Close-loop testing approach is proposed by replacing traffic system with a microscopic simulator, MITSIM. Tests are first conducted on a two-path synthetic network to demonstrate the capability of the framework with changing demand and different behavior parameters. Then a case study is performed on NTE Express Lanes network in Texas. Calibration of the network with multiple sources of traffic data is discussed, and initial calibration results with sensor data are presented. Also, the models are extended to account for the regulation rules imposed by the local government. Optimization results for morning peak period on a typical weekday are presented, and the resulting revenue is compared with the benchmark case. Finally, potential improvement in solution algorithm is discussed for the system's real time computational requirements. The main contribution of the thesis includes: 1) identifying the limitations of