ANN-Bayes-Based Travel Time Prediction Method for Signalized Corridors

Uncertainties always lie in predicting travel time along a signalized corridor or dynamic urban network. Frequent interrupted traffic flows, varying signal timing schemes, crossing traffic, and dynamic route choices make arterial travel time estimations much more challenging than freeways. This paper presents a study of developing an Artificial Neural Network techniques and Bayes algorithms for signalized corridors travel time prediction. Two types of models are tested with the data source obtained along the US-27 corridor in Ohio, including hourly instantaneous travel time prediction model and realized route travel time prediction model. The study suggests that the proposed methods can effectively capture the travel time patterns by combining a base profile and an ANN-Bayes-trained dynamic profile, and make it more sensitive to a variation aroused by non-recurring congestion. The testing results indicate a good performance of the predicted travel time models that the most errors fall into the range of 0.608 to 0.485, much lower than the threshold standard deviation (2.24) and close to 6% of mean value (0.077).