Journal
MATHEMATICAL POPULATION STUDIES
Volume 21, Issue 4, Pages 205-219Publisher
TAYLOR & FRANCIS INC
DOI: 10.1080/08898480.2013.836418
Keywords
reinforcement learning; uncertain network; real-time information; experiment
Ask authors/readers for more resources
Nonrecurring disruptions to traffic systems caused by incidents or adverse conditions can result in uncertain travel times. Real-time information allows travelers to adapt to actual traffic conditions. In a behavior experiment, subjects completed 120 days of repeated route choices in a hypothetical, competitive network submitted to random capacity reductions. One scenario provided subjects with real-time information regarding a probable incident and the other did not. A reinforcement learning model with two scale factors, a discounting rate of previous experience and a constant term, is estimated by minimizing the deviation between predicted and observed daily flows. The estimation combines brute force enumeration and a subsequent stochastic approximation method. The prediction over 120 runs has a root mean square error of 1.05 per day per route and a bias of 0.14 per route.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available