Analytical Probabilistic Traffic Models for Large-scale Network Optimization

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

Major transportation agencies in the U.S. and Europe have recognized the importance of measuring and optimizing the reliability and robustness of our networks. Evaluating reliability and robustness metrics involves the use of probabilistic network models. This project formulates, validates and uses probabilistic network models. Case studies based on actual metropolitan areas will illustrate the importance of accounting for uncertainty in large-scale transportation network analysis. The use of these methods can inform the design and operations of the considered networks, helping to mitigate congestion along with its economic, environmental and health impacts. These case studies on complex regional networks will illustrate the contributions to transportation practice of the methodologies. The findings of this project will be shared through various activities with transportation researchers, transportation stakeholders, the general public and with young engineers interested in learning about and contributing to the transportation challenges of the future. This project formulates an analytical stochastic kinematic wave model for general network topologies. It formulates a model that is suitable to address large-scale network optimization problems. First, the project formulates stochastic link models that are consistent with the kinematic wave model. Two types of models are formulated: (i) models with a complexity that is linear in the link?s space capacity, (ii) models with a complexity that is independent of the link?s space capacity. This is achieved through a combination of ideas from the fields of traffic flow theory, queueing network theory, transient queueing theory, and more generally operations research. Second, the project formulates a network decomposition approach that enables the link models to be used for large-scale network analysis. Third, this project plans a technique to approximate the joint network distribution of a given performance measure based on lower-dimensional subnetwork distributions. The case studies of this project contribute to the modeling of between-link dependency structures, as well as to their use to mitigate congestion for large-scale networks.