Browse Prior Art Database

Neural Network Modeling of Driver's Managed Lane Preference Behavior

IP.com Disclosure Number: IPCOM000236941D
Publication Date: 2014-May-22
Document File: 4 page(s) / 108K

Publishing Venue

The IP.com Prior Art Database

Abstract

High Occupancy Toll (HOT) lanes are utilized on high speed highways to reduce congestion and maintain the service levels. At each toll gate, drivers need to make the decision whether or not to enter the HOT lane by paying the toll. It is very important to be able to model this driver behavior accurately so that the correct toll can be applied on the HOT lanes for overall optimum traffic flow. Without an accurate behavior model, it is very difficult to develop an effective pricing algorithm. The commonly-used driver preference models are random utility models, which are based on the utility functions for the HOT and the General-purpose (GP) lanes. An individual driver will choose to travel on the HOT lane if the utility value of HOT is higher than that of GP. On the other hand, the driver will stay in the GP lanes if the utility value of HOT is not greater than that of GP. There are different forms of utility functions for such choice selection, and it is very difficult to validate the utility function and justify which one is closer to reality. Thus, we introduce a Neural Network (NN) to represent the driver’s preference behavior model to avoid finding out the exact forms for utility functions. This idea proposes to use Neural Networks (NN) to model drivers' behavior to predict whether a driver will chose a high occupancy toll (HOT) lane or a general purpose (GP) lane. Use of NN eliminates the need to determine exact functions and parameters to model the drivers' behavior

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 38% of the total text.

Neural Network Modeling of Driver's Managed Lane Preference Behavior

High Occupancy Toll (HOT) lanes are utilized on high speed highways to reduce congestion and maintain the service levels.  At each toll gate, drivers need to make the decision whether or not to enter the HOT lane by paying the toll. It is very important to be able to model this driver behavior accurately so that the correct toll can be applied on the HOT lanes for overall optimum traffic flow. Without an accurate behavior model, it is very difficult to develop an effective pricing algorithm. The commonly-used driver preference models are random utility models, which are based on the utility functions for the HOT and the General-purpose (GP) lanes. An individual driver will choose to travel on the HOT lane if the utility value of HOT is higher than that of GP. On the other hand, the driver will stay in the GP lanes if the utility value of HOT is not greater than that of GP. There are different forms of utility functions for such choice selection, and it is very difficult to validate the utility function and justify which one is closer to reality. Thus, we introduce a Neural Network (NN) to represent the driver’s preference behavior model to avoid finding out the exact forms for utility functions. This idea proposes to use Neural Networks (NN) to model drivers' behavior to predict whether a driver will chose a high occupancy toll (HOT) lane or a general purpose (GP) lane. Use of NN eliminates the need to determine exact functions and parameters to model the drivers' behavior

This idea proposes a NN to model the driver’s preference behavior to avoid figuring out the exact forms for utility functions of HOT and GP. Firstly, factors that affect drivers’ decisions have to be determined, e.g., value of time (VOT), travelling time, and toll rate etc. Then, prepare the data to train the NN in order for the NN to capture the relationship between the factors and drivers’ decisions. The training data can be obtained either by demographic survey or designed simulation experiment. In the same way, another set of data is needed to validate the trained NN. After the NN is validated within the acceptable prediction error, the NN will be used to model the driver’s preference behavior.

Background:

The fact that distinguishes HOT transportation system from others is that drivers decide whether or not to travel in the HOT lane based on utility functions for HOT and GP (General-Purpose) lane. Equations (1) and (2) are examples of utility functions for HOT and GP respectively.

 

    Equation 1

(1)

 

                Equation 2

(2)

Where,  is the value of time. TR is the toll rate.  and  are the travelling time on HOT and GP lanes respectively.

If , then drivers will choose to enter the HOT lane. Otherwise, they remain on the GP lane. Also there are a couple of utility functions introduced, but it is very hard to justify whether or not they match the real world. Therefore...