This paper presents a formal model of travel time learning, driven by the notion that insight into dynamic travel time perceptions is crucial for the understanding of travelers time use and their (re-)scheduling of activity-travel patterns. It presents two methodological contributions to Bayesian literature on travel time learning. First, previous work has focused on learning a routes mean travel time, reflected by a variance associated with this mean which gradually decreases as more observations are made. Our model acknowledges that travelers may learn a routes day-to-day travel time variability as well. This is reflected in an additional variance term that gradually approaches the true travel time variance as more observations are made. Secondly, the model incorporates the notion that travelers may learn that travel times on different (nearby) routes are likely to be correlated. Mathematically, this implies that this paper model how travelers update the full covariance matrix associated with multiple routes travel times, together with their means. Numerical simulations show that the outcomes of the model are consistent with intuitions regarding actual traveler behavior. For example, it is found that travelers learn more rapidly the travel time distribution for routes with a low level of travel time variability.
Abstract