Publication Detail
Design of a Recurrent Neural Network for Analyzing Route Choice Behavior in the Presence of Information System
|
UCD-ITS-RP-95-43 Presentation Series Download PDF |
Suggested Citation:
Reddy, Prasuna D., Kenneth M. Vaughn, Mohamed A. Abdel-Aty, Ryuichi Kitamura, Paul P. Jovanis (1995) Design of a Recurrent Neural Network for Analyzing Route Choice Behavior in the Presence of Information System. Institute of Transportation Studies, University of California, Davis, Presentation Series UCD-ITS-RP-95-43
Proceedings of 1995 Annual Meeting of ITS America, Washington, DC
As a cost-effective alternative to field studies, computer simulation is an often-used methodology to study travel behavior. In this study, a PC-based computer simulation was used to study the effects of information on drivers' route choice and learning. Building on a prior stage of simulation efforts, a new set of experiments was developed with an expanded traffic network and various levels of information given to subjects. This framework allows one to investigate both en route and pre-trip route-choice behavior and capture the effect of different levels of information on drivers' learning and adaptive processes.
The experiments were conducted in two generations (stages). In the first-generation experiments, a simple, two-route-alternative traffic network was developed. Experiments conducted with this network provided the authors with a set of extensive comments from participants. These insights were modeled using object-oriented programming techniques to produce a better subsequent design. Data from the first-generation experiments were analyzed using neural network techniques, and the neural network was trained using the back-propagation method. The second-generation experiments used a multiple-route, expanded network with varying levels of information. Data obtained in this stage are being analyzed using recurrent neural networks. This paper describes the redesign of the network simulation with the experience gained in the first-generation experiments. The paper also analyzes data obtained from the experiment.
Design of the network simulation involved the following steps: requirements analysis, data base design, specifications of user-computer interface, design of shortest-path module, software development, and prototype testing and refinement. The simulator was developed using an object-oriented programming language, C++. A recurrent neural network has been built for modeling of the data obtained in the second generation experiments. This neural network will be used to predict subjects' choices of whether or not to follow the system-provided advice, depending on their past experience. An important feature of this neural network is that decisions at previous nodes will be used as an input for the neural network at subsequent nodes. This allows one to model participants' route-choice behavior at every node that approximates a traffic intersection.
As a cost-effective alternative to field studies, computer simulation is an often-used methodology to study travel behavior. In this study, a PC-based computer simulation was used to study the effects of information on drivers' route choice and learning. Building on a prior stage of simulation efforts, a new set of experiments was developed with an expanded traffic network and various levels of information given to subjects. This framework allows one to investigate both en route and pre-trip route-choice behavior and capture the effect of different levels of information on drivers' learning and adaptive processes.
The experiments were conducted in two generations (stages). In the first-generation experiments, a simple, two-route-alternative traffic network was developed. Experiments conducted with this network provided the authors with a set of extensive comments from participants. These insights were modeled using object-oriented programming techniques to produce a better subsequent design. Data from the first-generation experiments were analyzed using neural network techniques, and the neural network was trained using the back-propagation method. The second-generation experiments used a multiple-route, expanded network with varying levels of information. Data obtained in this stage are being analyzed using recurrent neural networks. This paper describes the redesign of the network simulation with the experience gained in the first-generation experiments. The paper also analyzes data obtained from the experiment.
Design of the network simulation involved the following steps: requirements analysis, data base design, specifications of user-computer interface, design of shortest-path module, software development, and prototype testing and refinement. The simulator was developed using an object-oriented programming language, C++. A recurrent neural network has been built for modeling of the data obtained in the second generation experiments. This neural network will be used to predict subjects' choices of whether or not to follow the system-provided advice, depending on their past experience. An important feature of this neural network is that decisions at previous nodes will be used as an input for the neural network at subsequent nodes. This allows one to model participants' route-choice behavior at every node that approximates a traffic intersection.