Predicting Travel Times in Dense and Highly Varying Road Traffic Networks using STARIMA Models.
Research output: Book/Report › Report
As road networks are becoming increasingly utilised, it is increasingly important to be able to accurately predict travel times for disseminating information among road users and to support traffic management decisions and planning. Over the past few years, a new way of making such predictions utilising the space-time autoregressive integrated moving average (STARIMA) model has been introduced. The results have been very promising so far with very good accuracy reported for prediction times of several tens of minutes. However, so far, the literature only concerns steady state freeflow or Manhattan grid based scenarios. In this paper, we generalise on the previous work and investigate how this approach performs in urban traffic scenarios. We investigate the models prediction accuracy both on a data set of measured travel times in the metropolitan Sydney region as well as a faithful representation of a large section of Sydney’s urban landscape. We analyse the performance of STARIMA under six level of service (LOS) in both settings and find that even though the model performs well in the steady state case, the basic approach is not suitable for modelling the urban traffic setting. We therefore propose to extend the STARIMA model with feedback control loops in order for the approach to be suitable for highly varying environments.
|Research areas and keywords||
Subject classification (UKÄ) – MANDATORY
|Publisher||University of Sydney|
|Number of pages||23|
|Publication status||Published - 2012|
|Name||School of Information Technologies Technical Reports|
No data available