Application of Advanced Traffic Information Systems

With increased congestion in most urban networks, providing reliable trip times to commuters has emerged as a critical challenge for all existing advanced traffic information systems. This paper presents the development and evaluation of a hybrid travel-time prediction model in a demonstration system sponsored by the Maryland State Highway Administration. The automated real-time travel-time prediction system is located on a 25-mi stretch of I-70 eastbound from MD-27 to I-695, which includes seven interchanges and 10 traffic detectors. To contend with the resource constraints of more than 1 mi in detector spacing, the proposed prediction model consists of two components: a multitopology neural network model with a clustering function as the main model and an enhanced k nearest neighbor model as the supplemental model. The evaluation results based on estimated travel times and field data collected by a third party indicate that the proposed hybrid model can provide reliable predictions of travel times in real-time operations in uncongested, congested, and transition traffic conditions.