USING NEURAL NETWORKS TO SYNTHESIZE ORIGIN-DESTINATION FLOWS IN A TRAFFIC CIRCLE

The traffic circle is a classic transportation problem for traffic engineers. Although it is easy to determine the volume of vehicles entering and exiting the circle at all points, it is difficult to determine the actual flow pattern of these vehicles. In other words, although it is easy to determine how many vehicles enter the circle from a given street, it is difficult to determine how many of those vehicles will leave the circle at each possible exit point. Currently, the only method of accurately determining this traffic flow is to visually track each vehicle as it enters and exits the circle, a laborious method of collecting data. However, emerging neural network technologies give researchers another approach. The capability of neural networks to handle subtle or contradictory information by organizing and capturing complex relationships, optimizing and generating analytical models, and learning and adapting the model when new data become available has made them increasingly popular in transportation and traffic flow models. The objective is to describe the development of a neural network model for generating origin-destination (O-D) information for traffic circles based on observed flow volumes on approaching and exiting legs. The quality of the model is evaluated with respect to the different methods used to train the model. Observations about the synthesized O-D matrices and the corresponding errors generated by the neural network model are also described.

Language

  • English

Media Info

  • Features: Figures; References; Tables;
  • Pagination: p. 134-142
  • Monograph Title: Part 1: 1994 TRB Distinguished Lecture, Adolf D May; Part 2: Traffic flow and capacity
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00677656
  • Record Type: Publication
  • ISBN: 0309061008
  • Files: TRIS, TRB
  • Created Date: May 11 1995 12:00AM