Dynamic OD matrix estimation using floating car data : Floating car data used for a-priori estimation and route choice analysis

Travel demand in terms of an Origin-Destination (OD) matrix is an essential input for network traffic assignment and traffic simulation models. Several different methods that estimate OD matrices exist. For most of them, a-priori information (in the form of matrices) is necessary. Until this day, that information has been acquired from noncurrent data sources, such as household surveys and road interviews. Due to possible changes in the network or the traffic demand there is no guarantee that these data are still accurate. The quality of OD matrix estimations depends largely on the quality of the input data. This thesis is aimed at improving the input data and hence improving the estimations themselves. Detailed traffic data has been collected in the city centre of Chengdu in China. The data comprise traffic counts, video camera data and Floating Car Data (FCD) that come from taxis. In this thesis the FCD are used to derive a-priori matrices and analyze route choices and Trip Length Distributions (TLD). From the results it can be concluded that FCD can indeed be used for estimating a-priori matrices and analysing the route choices. However, due to the fact that FCD are only sample data, they lack some information. This lack of information can be compensated with traditional a-priori matrices and route choice analyses. In that way, additional, current information is added to the traditional input data, which increases their reliability. The TLD analyses made with the FCD are consistent between days, but due to the limited size of the study area, short trips are overestimated. With a larger study area the TLD could be used to scale OD matrices so that they match the total traffic. That eliminates estimation bias caused by the data coming from taxis. Since the real OD matrix estimations are not known, the reliability of the developed methods cannot be estimated. It is however clear that using FCD for the examined purposes is very feasible. For further research, it is recommended that the real OD matrices should be found. Other recommendations are for instance to enlarge the study area and add information to the FCD regarding the taxi's occupancy.