PT Journal
AU van den Haak, W
   Rothkrantz, L
   Wiggers, P
   Heijligers, B
   Bakri, T
   Vukovic, D
TI Modeling Traffic Information using Bayesian Networks
SO Transactions on Transport Sciences
PY 2010
BP 129
EP 136
VL 3
IS 3
DI 10.2478/v10158-010-0018-09
DE Bayesian Networks; prediction; vehicle speed; inductive loop detector data.
AB Dutch freeways suffer from severe congestion during rush hours or incidents. Traffic congestion increases travel time, resulting in a delay for travelers. To avoid these delays, rerouting traffic around congested areas is an option. Reliable travel time predictions are essential for dynamic routing and travel information. Travel time can be calculated from vehicle speed measurements (van Lint, 2004). These speed measurements are acquired from dual inductive loop detectors collected by the Dutch Monitoring Casco (MONICA) data system. In this paper, the predictability of average vehicle speed by Bayesian Networks is investigated in a case study. We propose a general Bayesian Network model and evaluate several simplified versions of this model on a well known traffic bottleneck in the Netherlands. We show that our Bayesian Network is capable of predicting the start and end of a congestion for a prediction horizon of 30 minutes with an accuracy of 14%. Furthermore, we present a prediction model based on historical data which is evaluated on the same bottleneck. This prediction model based only on historical data and our Bayesian Network are combined in a hybrid model, where we evaluate performance as well. This hybrid model is able to predict congestion with an accuracy of 85% for a rather long prediction horizon of 2.5 hours in our case study.
ER