Modelling the Impact of Bottlenecks on Arterial Travel Time Using Neural Networks

Abstract

Bottlenecks along signalized arterials are a major cause of capacity reduction and delay, which directly impact travel time along specific routes within an urban network. This research addresses the impacts of bottlenecks on arterial travel time under different traffic demand and geometric conditions. Neural network models are developed that quantify the impact of different types of bottlenecks on travel time. Different combinations of conditions are studied including variation in number of lanes, traffic demand (volume), length and position of bottlenecks, and presence of heavy vehicles. An extensive database of synthetic traffic data generated from microscopic traffic simulation is used. Link travel times are observed for different traffic demand levels/geometrics/bottleneck combinations. Different architectures of a back propagation neural network are evaluated. Results show that the neural network models are able to capture travel time with high accuracy. For comparison purposes, linear regression models are developed as well. The neural network models significantly outperformed the regression models. The results are a clear demonstration that neural network models can be a valuable tool for predicting travel time, a necessary piece of information for traffic routing and emergency evacuations under different traffic and geometric conditions.