A traffic flow simulator for driver information system

Abstract

A traffic flow simulator (TFS) is proposed in this thesis for deriving the prevailing traffic condition of the road network based on partial traffic count. The TFS integrates origin-destination (OD) matrix estimation with the probit-type stochastic user equilibrium (SUE) assignment problem. When short-term traffic forecasts are available at the locations with historical and prevailing detector data, the TFS can be extended for prediction of short-term traffic condition under the context of the driver information system (DIS) as well as fleet management and logistics. In summary, the first contribution of this thesis is the incorporation of link flow variance and covariance information in the OD estimation. One of the advantages of using probit-type SUE model in the TFS is the availability of variance and covariance matrix of link flows. Previous OD estimation models are based only on partial traffic count and prior OD matrix, but link flow variance and covariance information have not been considered. It is shown that additional information on link flow variance and covariance can help to improve the performance of OD estimation and the short-term forecasting of link flows and travel times for DIS. Calibration of the TFS is the second contribution of this thesis. As the variance and covariance of link flow is induced by disthbution of driver perception error and travel demand variation, two parameters in the TES, the perceived error coefficient and the OD variation coefficient, need to be calibrated before the TFS can be applied in practice. Various calibration measures are discussed and appropriate measures are identified for TFS calibration. Due to the complexity of the probit-type SUE model, a stochastic optimizing technique, the genetic algorithm (GA), is employed to calibrate the TFS. The third contribution of this thesis is two extensions made for the TFS, The distribution of perception error on path travel time of one driver may not be the same as another owing to different associated characteristics such as trip regularity, driving experience or trip purpose. Hence, multiple-user-class dimension is introduced to the TFS so as to accommodate different route choice behaviour across drivers. The multiple-user-class TFS is then used to assess the impacts of the DIS particularly on network travel time reliability that is defined as the probability of a trip can be finished within a given period of time. With the deployment of DIS, some drivers can access the DIS information which can help them to choose the better route and save travel time. Therefore, drivers can be classified into two user classes, with and without DIS information. The TFS is used to evaluate network travel time reliability with respect to different DIS market penetration, i.e. proportion of drivers with DIS information. Another extension made for the TFS is integration of the TFS with transport network calibration (i.e. updating the parameters of link travel time functions). In the conventional approach, OD matrix estimation and network calibration are carried out iteratively and a stable solution may not be achieved. As the TFS can serve as an OD estimation model, the TFS is extended for integration with network calibration in this study. Finally, a case study in Hong Kong has been undertaken to illustrate the performance of the proposed TFS for large-sca1e network and to demonstrate the short-term link flow and travel time forecasts for DIS.