A data-driven dynamic route choice model under uncertainty using connected vehicle trajectory data

This paper proposes a data-driven dynamic route choice model to understand traveler’s routing behavior in a time-dependent network under uncertainty using connected vehicle trajectory data over many days. Different from existing efforts on stochastic route choice models using a random term with a given distribution, this paper directly uses connected vehicle trajectory data over many days without knowing the underlying distribution in a data-driven stochastic optimization model. Specifically, the authors apply a Bayesian risk formulation for parametric underlying distributions that optimizes a risk measure taken with respect to the posterior distribution estimated from the connected vehicle trajectory data. Two risk measures (i.e. Value-at-Risk and Conditional Value-at-Risk) of the travel time uncertainty are considered in the proposed data-driven dynamic route choice model. Based on the risk measures, the proposed model allows a flexible choice on the risk preferences of individual users (i.e. from risk-neutral to risk-averse). To test the data-driven dynamic route choice model in a large network, the authors implement the model in Southeast Michigan using a high-resolution (i.e. 0.1 seconds) trajectory dataset of connected vehicles from the Safety Pilot Model Deployment (SPMD) project over many days