Learning and managing stochastic network traffic dynamics with an aggregate traffic representation

This study estimates and manages the stochastic traffic dynamics in a bi-modal transportation system, and gives hints on how increasing data availability in transport and cities can be utilized to estimate transport supply functions and manage transport demand simultaneously. In the bi-modal system, travelers’ mode choices are based on their perceptions of the two travel modes: driving or public transit. Some travelers who have access to real-time road (car) traffic information may shift their mode based on the information received (note that real-time information about public transit departures/arrivals is not considered here). For the roadway network, the within-day traffic evolution is modeled through a Macroscopic Fundamental Diagram (MFD), where the flow dynamics exhibits a certain level of uncertainty. A non-parametric approach is proposed to estimate the MFD. To improve traffic efficiency, we develop an adaptive pricing mechanism coupled with the learned MFD. The adaptive pricing extends the study of Liu and Geroliminis (2017) to the time-dependent case, which can better accommodate temporal demand variations and achieve higher efficiency. Numerical studies are conducted on a one-region theoretical city network to illustrate the dynamic evolution of traffic, the MFD learning framework, and the efficiency of the adaptive pricing mechanism.