Research Papers

Neural Network-based Optimization of Traffic Safety and Operations for Roadway Corridors

Filename Alecsandru.pdf
Filesize 335 KB
Version 1
Date added June 6, 2010
Downloaded 1 time/fois
Category 2010 CMRSC XX Niagara
Tags Session 3B
Author/Auteur Ciprian Alecsandru, Behzad Rouhieh


Currently, most transportation agencies design signal timing plans for intersections with the main objective of minimizing vehicular traffic delay while ensuring compliance with basic safety guidelines. Often times along urban roadways where automobiles share the space with large volumes of non. motorized users (i.e. pedestrians and cyclists), reaching a balance between delays and safety of all road users is a challenging task. In this study, different approaches are presented to address potential improvements on traffic operations and safety of intersections serving more than one mode of transportation. A methodology is proposed to reach a desired compromise between the safety and efficiency of either an isolated intersection or a corridor of independent/coordinated intersections. An integrated delay. safety (DS) indicator is used in combination with a neural network based tool. The proposed methodology was applied to a urban arterial in downtown Montreal, along which a bicycle path was recently built. The study area was evaluated using VISSIM, a microscopic traffic simulator, by coding traffic signal timing plans along the arterial to perform independently, or coordinated. The objective is to advance with minimum delay a specified transportation mode (i.e. automobiles or bicycles). A Multi.Layer Perceptron (MLP) was built to identify what type of signal timing plan yields the best tradeoff between automobile delay and safety of non.motorized users. Based on traffic data collected from and from simulations, a large data set of input/output pairs was used to train and test the MLP neural network. It was found that for 99.8% of the tested cases the neural network identifies correctly the configuration of signal timing plan that yields the optimal DS value.

Ciprian Alecsandru and Behzad Rouhieh