UC Irvine

The forthcoming Connected Vehicles (CV) technology promises to substantially aid in managing traffic congestion and improving users’ mobility along transportation corridors. Through CV technology, a global estimate of the corridor’s traffic flow state can be obtained through analyzing data from its constituent components, which can enable globally-optimized traffic management strategies that efficiently utilize existing transportation infrastructure resources. In this paper, we propose a novel integrated corridor management (ICM) methodology that incorporates the underutilized infrastructure of park and ride facilities into its global optimization strategy. Firstly, we discuss how vehicle-to-infrastructure (V2I) communication protocols like basic safety messages (BSM) and traveler information messages (TIM) can be tailored to collect the state of downstream traffic and advertise park and ride advisories to upstream traffic respectively. Then, we model the system in terms of potential delays that can be experienced by vehicles traversing the corridor, and accordingly, we implement a novel centralized deep reinforcement learning (DRL) solution to control how and when such messages are advertised, with the aim of maximizing throughput and minimizing carbon emissions and travel time. We simulate our ICM strategy on a realistic model of Interstate 5 using the Veins simulation software, and the DRL agent converges to a strategy that provides marginal improvement to throughput, freeway travel time, and carbon emissions but at the cost of added travel delay for those using park and ride services.