Estimation of OD Demand for Dynamic Assignment with Route Choice and Departure Time Choice



Estimation of OD Demand for Dynamic Assignment with Route Choice and Departure Time Choice

Authors

LINDVELD C D R AND ZIJPP N VAN DER, Delft University of Technology, The Netherlands

Description

It is well known that increased congestion levels not only lead to increased dispersion of OD-demand over multiple routes, but also to shifts in departure time. On an aggregate level this leads to the widely observed peak-spreading phenomenon.

Abstract

It is well known that increased congestion levels not only lead to increased dispersion of OD-demand over multiple routes, but also to shifts in departure time. On an aggregate level this leads to the widely observed peak-spreading phenomenon.

Departure time choice may be modelled simultaneously with route choice, using a utility maximisation framework. In such an approach, each traveller has a preferred departure- and arrival time. A traveller may deviate from these times at the cost of a certain disutility. This type of assignment is called the Dynamic User Optimum Departure time and Route choice (DUO-D&R) assignment.

Where the DUO assignment needs a dynamic Origin-Destination (O-D) matrix with fixed departure times, the DUO-D&R assignment needs a dynamic O/D matrix with preferred departure times under uncongested conditions, and a function that gives the schedule delay cost associated as a function of arrival time. The schedule delay cost measures the disutility of a traveller who departsor arrives at a time other than the preferred arrival or departure time. We will focus on departure time schedule delay, and we will call the matrix of the demand stratified by preferred departure time period the uncongested O-D matrix (uOD).

This paper is concerned with the estimation of the uncongested Origin- Destination (uOD) demand matrix which reflects preferred rather then realised departure times. In order to carry out this estimation, the Dynamic User- Optimal Departure time and Route choice (DUO-D&R) problem is formulated and solved. The schedule delay function and the departure-time choice model are assumed to be known.

Our hypothesis is that dynamic traffic assignment based on the use of the uOD matrix provides a better basis for predicting how traffic flows will respond to changes in the level of service. The reason is that travellers have at least two different ways to respond to L.O.S. changes: route diversion and departure time shift. On a network scale this results in re-routing or peak spreading.

We propose to test this hypothesis using the following numerical experiments. First a DUO-D&R is calculated on a known uOD matrix, next the resulting flows are perturbed and two versions of the dynamic O/D matrix are estimated back from the perturbed DUO-D&R: one based on the assignment map from the DOU-D&R, the other based on the assignment map of an ordinary DUO assignment. The matrix estimation procedure will ensure that both versions will reproduce the network flows. Next the network is changed, and the resulting "true" flows are calculated using the DUO-D&R.

Two aspects of the method are evaluated. Firstly the accuracy of the estimated uOD demand and its sensitivity to errors in the input data are evaluated. For this purpose traffic counts are generated using the DUO-D&R model. These are perturbed to reflect inaccuracies in the model and to simulate inaccuracies in data collection. The uOD demand is then estimated from these data and is compared to the OD demand from which the data were generated.

Secondly the accuracy is evaluated with which flow levels can be predicted for network variants. For this purpose the uOD demand estimated in the first series of experiments is assigned to the original network after increasing capacity at a number of congested links. The predicted flow levels are compared to those that result when the original OD demand is assigned.

Publisher

Association for European Transport