Statistical approach for activity-based model calibration based on plate scanning and traffic counts data

Abstract

This thesis aims to contribute to the rapidly growing research area of route travel time estimations and travel demand model calibrations from traffic observations (vehicle identification data and traffic counts). The method, which is generally adopted in this thesis, is based on maximum-likelihood estimations. To obtain the reliable calibration/estimation results, this thesis also aims to contribute to the design on the suitable locations of traffic observations. The thesis, first, presents a new method for estimating route travel times and route choice parameters based on vehicle identification data (sensor-to-sensor travel time). This method is developed, which is based on maximum likelihood estimations and solved by the Expectation-Maximization (EM) algorithm. To extend such estimation method for the related application of travel behaviour model, the thesis proposes the new method for updating of travel behaviour model parameters and vehicle trip-chain demands based on vehicle identification data (sensor-to-sensor travel time). In addition, there is a growing importance on development of an advance travel-demand model (activity-based model) for predicting the realistic travel behaviour of travellers, which is better than the traditional 4-step model (trip generation/trip distribution/mode split/traffic assignment). However, one of the main obstacles of implementing the advance travel demand is the model calibrations. This thesis also aims to propose a statistical method for the activity-based model calibrations (the calibrations of synthetic population and parameters in activity-based model) based on the vehicle identification data and traffic counts. In this thesis, compared to the use of link counts in traditional model calibration approaches, vehicle identification data (license plate scanning) is shown to be more informative and suitable for the calibration of activity-based model deriving the travel demand from activity behaviour sequences and trip chains. Finally, the thesis focuses on finding the suitable locations for vehicle identification observations (license plate scanning observations). Updating of optimal license plate scanning locations with link counts to enhance the origin-destination matrix estimations under budget limitation was adopted in this thesis. The license plate scanning location problem is formulated in the integer-programming model, in which a binary variable is presented by the set of suitable links installed with plate scanning sensors. The results illustrate the efficiency of the proposed plate scanning location model and its potential for large and complex applications.