Stochastic dynamic traffic assignment for robust online operations under real -time information systems

Chao Zhou, Purdue University

Abstract

Dynamic Traffic Assignment (DTA), which aims at determining paths for traffic network users, plays a key role in Advanced Traveler Information Systems and Advanced Traffic Management Systems. Most existing DTA models are inadequate for on-line deployment because of their unrealistic assumptions on the availability of on-line information, their inability to account for the randomness in unfolding on-line conditions, and their computational requirements. In this research, a stochastic formulation of the dynamic traffic assignment problem is proposed to incorporate the inherent stochasticity within the traffic systems for on-line deployment. A stochastic quasi-gradient (SQG) algorithm is proposed to solve the problem. Simulation experiments indicate that the proposed stochastic quasi-gradient algorithm is superior to the previously proposed algorithms. To ensure on-line computational feasibility, a new paradigm for deployable dynamic traffic assignment is proposed. Called the hybrid strategy, it combines both off-line and on-line components. The off-line component uses the a priori optimization technique and SQG to seek a robust initial solution vis-à-vis randomness in origin-destination (O-D) demands using historical data. Heuristics are proposed to dynamically update the initial solution on-line based on unfolding demand and supply conditions. The hybrid approach circumvents the need for accurate on-line demand and network supply prediction models, while exploiting historical demand and network supply data off-line. Thereby, the computationally intensive components are executed off-line and are coupled with highly efficient on-line update heuristics. Extensive simulation studies highlight the robustness of the hybrid approach with respect to on-line variations in demand, its ability to address stochastic incident situations effectively, and its on-line efficiency.

Degree

Ph.D.

Advisors

Peeta, Purdue University.

Subject Area

Civil engineering|Industrial engineering|Operations research

COinS