[Back]

M. Tulic, D. Bauer, W. Scherrer:
"Link and Route Travel Time Prediction Including the Corresponding Reliability in an Urban Network Based on Taxi Floating Car Data";
Poster: Transportation Research Board, 93rd Annual Meeting, Washington, DC.; 2014-01-12 - 2014-01-16.

This paper deals with the prediction of link travel speeds, link- and route-travel-times in urban networks based on taxi floating car data. The expected link travel speeds are modeled as functions of deterministic regressors such as daily profiles as well as seasonal components. The model specification is mainly based on automatic modeling procedures using information criteria. The travel speeds are highly heteroskedastic and thus the variances of the travel speeds are modeled as a function of the number of measurements as well as the traffic state. Different VAR(1) models are proposed to take the spatial structure of the travel speeds into consideration and compared to each other as well as to the mean model. The proposed modeling methods are demonstrated using a testsite near the core city of Vienna equipped with a rich variety of urban traffic conditions. The main findings of our analysis are as follows. Taxi floating car measurements of local speeds are strongly heteroskedastic and this has to be taken into account for the estimation of models for the expected travel speeds. The modeling of the mean leaves no suggestion of remaining daily or weekly pattern while being superior to simple univariate AR(1) models. The variance model successfully captures the heteroskedasticity. The more complex models for link travel speeds including temporal and spatial correlation do not increase prediction accuracy consistently and hence display the fact that a sampling frequency of fifteen minutes for the floating car data in urban settings is too low to allow for temporal dependencies that could be exploited for prediction. In addition to predicted route travel times, a method for computation of route-travel-time uncertainty is introduced which shows variability over the day evidently for a highly frequented route.

link-travel-times, route-travel-times, urban network, floating car data, prediction