ON THE USE OF NEURAL NETWORKS TECHNIQUES FOR TRAFFIC FLOW MODELING
Today, most large cities in industrialized countries are confronted with chronic traffic congestion. To cope with this problem, traffic management in urban areas seems to be a necessity. The most recently designed urban traffic control systems are fully adaptive, in the sense that they modify the signal states on the basis of a control law to respond to variation of the traffic conditions. Their policy is to evaluate the consequences of the application of a particular set of commutations on traffic evolution. Only the use of a traffic flow model makes this projection in the future possible. No unified theory of traffic flow exists, rather there are several theoretical approaches to describe this phenomenon. The traffic flow models, most widely applied, are analytic models, based on the laws of physics and mathematics. Usually, they require preliminary knowledge about traffic flow. Moreover, due to the limitations of the hypotheses they are built on, they are inaccurate to represent the traffic dynamics under some circumstances. To overcome these difficulties, a traffic flow model based on neural networks is proposed. Neural nets are introduced here to model the queue length on a link whose output is controlled by signals. They are nonlinear data driven models, capable of inferring, from data that are relevant to the computation of the queue length, the queue length function. The purpose of this paper is to present the results of the investigations on the potential of neural networks to model traffic flow. The first part is dedicated to an overall presentation of adaptive real time UTC systems. The second part presents the limits of conventional modeling techniques used in such systems. Then, neural networks are further detailed. In the fourth part, their abilities to approximate the queue length function are underlined. The benefits of their use over more conventional techniques are clear. Finally, based on numerical simulations, their effectiveness to queue modeling is discussed.
-
Supplemental Notes:
- Five volumes of papers and one volume of abstracts comprise the published set of conference materials.
-
Corporate Authors:
VERTIS
TORANOMOM 34 MORI BUILDING 1-25-5
TORANOMON, MINATOKU, TOKYO 105 Japan -
Authors:
- LEDOUX, C
- Boillot, F
- Sellam, S
-
Conference:
- Steps Forward. Intelligent Transport Systems World Congress
- Location: Yokohama, Japan
- Date: 1995-11-9 to 1995-11-11
- Publication Date: 1995-11
Language
- English
Media Info
- Pagination: p. 1915
Subject/Index Terms
- TRT Terms: Highway traffic control; Intelligent transportation systems; Neural networks; Queuing theory; Real time control; Traffic simulation; Urban areas
- Geographic Terms: France
- Old TRIS Terms: Queueing models; Real-time systems
- Subject Areas: Highways; Operations and Traffic Management; I71: Traffic Theory;
Filing Info
- Accession Number: 00724401
- Record Type: Publication
- Report/Paper Numbers: Volume 4
- Files: TRIS
- Created Date: Aug 7 1996 12:00AM